Sociality and the life-mind continuity thesis - Dr. Tom Froese
Sociality and the life-mind continuity thesis - Dr. Tom Froese Sociality and the life-mind continuity thesis - Dr. Tom Froese
Sociality and the Life-Mind Continuity Thesis: A Study in Evolutionary Robotics Tom Froese Submitted for the degree of D.Phil. University of Sussex June 2009
- Page 2 and 3: Declaration I hereby declare that t
- Page 4 and 5: Table of contents 1 Introduction ..
- Page 6 and 7: 13.2 Primatology ..................
- Page 8 and 9: 1 Introduction Out of all the tradi
- Page 10 and 11: approach can systematically resolve
- Page 12 and 13: these agents with the role of inter
- Page 14 and 15: „top-down‟ perspective, namely
- Page 16 and 17: subjectivity at the heart of cognit
- Page 18 and 19: importance of AI for this ongoing s
- Page 20 and 21: hypotheses are viable. Empirical re
- Page 22 and 23: theories outside the field of AI, s
- Page 24 and 25: dimension of bodily self-regulation
- Page 26 and 27: egulation, or leads to the constitu
- Page 28 and 29: However, while Wheeler‟s appeal t
- Page 30 and 31: provides a tentative answer to thes
- Page 32 and 33: personal experience. They point out
- Page 34 and 35: interaction (Chapters 7 to 10). The
- Page 36 and 37: The most comprehensive framework ba
- Page 38 and 39: To be sure, a related response has
- Page 40 and 41: Varela 2002; Di Paolo 2003; 2005; F
- Page 42 and 43: emains controversial (cf. Thompson
- Page 44 and 45: under precarious conditions (cf. Di
- Page 46 and 47: In order for these considerations t
- Page 48 and 49: compensation. In the context of sen
- Page 50 and 51: terminology allows us to propose a
<strong>Sociality</strong> <strong>and</strong> <strong>the</strong> Life-Mind Continuity Thesis:<br />
A Study in Evolutionary Robotics<br />
<strong>Tom</strong> <strong>Froese</strong><br />
Submitted for <strong>the</strong> degree of D.Phil.<br />
University of Sussex<br />
June 2009
Declaration<br />
I hereby declare that this <strong>the</strong>sis has not been submitted, ei<strong>the</strong>r in <strong>the</strong> same or different<br />
form, to this or any o<strong>the</strong>r University for a degree.<br />
Signature:<br />
2 | P a g e
<strong>Sociality</strong> <strong>and</strong> <strong>the</strong> Life-Mind Continuity Thesis:<br />
A Study in Evolutionary Robotics<br />
<strong>Tom</strong> <strong>Froese</strong><br />
Summary<br />
The <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis holds that <strong>mind</strong> is prefigured in <strong>life</strong> <strong>and</strong> that <strong>mind</strong><br />
belongs to <strong>life</strong>. Its biggest challenge is <strong>the</strong> problem of scalability: how can <strong>the</strong> same<br />
explanatory framework that accounts for basic phenomena of <strong>life</strong> <strong>and</strong> <strong>mind</strong> be extended<br />
to incorporate <strong>the</strong> highest reaches of human cognition? So far <strong>the</strong>re has been little<br />
systematic response to this „cognitive gap‟. The main argument of this <strong>the</strong>sis is that <strong>the</strong><br />
problem appears insurmountable because of <strong>the</strong> prevalent focus on <strong>the</strong> individual agent<br />
alone, <strong>and</strong> that it can start to be addressed by an appreciation of <strong>the</strong> constitutive role of<br />
sociality for <strong>mind</strong> <strong>and</strong> behavior. This argument is developed in a <strong>the</strong>oretical,<br />
experimental, <strong>and</strong> phenomenological manner. In terms of <strong>the</strong>ory, <strong>the</strong> enactive paradigm<br />
of cognitive science is developed in a novel direction by highlighting <strong>the</strong> specific<br />
manner in which <strong>the</strong> dynamics of <strong>the</strong> interaction process opens up new behavioral<br />
domains. This provides <strong>the</strong> motivation for using an evolutionary robotics methodology<br />
to syn<strong>the</strong>size a set of minimalist simulation models that are based on experiments in<br />
social psychology. A detailed dynamical analysis of <strong>the</strong>se models supports <strong>the</strong> enactive<br />
approach; <strong>the</strong> behavior of <strong>the</strong> agents is not an individual achievement alone but ra<strong>the</strong>r<br />
co-determined by <strong>the</strong>ir mutual interaction <strong>and</strong> organized effectively by this multi-agent<br />
interaction process. Some phenomenological observations complement <strong>the</strong>se results by<br />
indicating that <strong>the</strong> detached perceptual attitude that is characteristic of adult human<br />
perception is essentially an intersubjective <strong>and</strong> socially mediated ability. Finally, <strong>the</strong><br />
systemic <strong>and</strong> phenomenological insights are combined to provide <strong>the</strong> beginnings of a<br />
novel perspective on <strong>the</strong> origins of cumulative cultural development that gives fur<strong>the</strong>r<br />
support to <strong>the</strong> main argument of this <strong>the</strong>sis. It is concluded that <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong><br />
<strong>the</strong>sis is a viable working hypo<strong>the</strong>sis even when accounting for specifically human<br />
abilities, <strong>and</strong> that an appreciation of <strong>the</strong> constitutive role of sociality for <strong>life</strong> <strong>and</strong> <strong>mind</strong><br />
confirms it to be a serious contender for a unified <strong>the</strong>ory of cognitive science.<br />
Submitted for <strong>the</strong> degree of D.Phil.<br />
University of Sussex<br />
June 2009<br />
3 | P a g e
Table of contents<br />
1 Introduction ............................................................................................................... 8<br />
2 A brief history of cognitive science ........................................................................ 15<br />
2.1 Toward embodied-embedded cognitive science .............................................. 16<br />
2.2 Fur<strong>the</strong>r: Toward enactive cognitive science ..................................................... 22<br />
2.3 An empirical stalemate ..................................................................................... 27<br />
2.4 A phenomenological resolution ........................................................................ 30<br />
2.5 Summary .......................................................................................................... 32<br />
3 Enactive cognitive science ...................................................................................... 35<br />
3.1 The <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis ......................................................................... 35<br />
3.2 Constitutive autonomy is necessary for intrinsic teleology .............................. 39<br />
3.3 Adaptivity is necessary for sense-making ........................................................ 44<br />
3.4 Constitutive autonomy is necessary for sense-making ..................................... 48<br />
3.5 Summary .......................................................................................................... 53<br />
4 The enactive approach to social cognition .............................................................. 58<br />
4.1 The autonomy of <strong>the</strong> interaction process ......................................................... 58<br />
4.2 Social interaction .............................................................................................. 64<br />
4.3 Cultural interaction ........................................................................................... 71<br />
4.4 Summary .......................................................................................................... 75<br />
5 Beyond methodological individualism ................................................................... 77<br />
6 Studies in social psychology: A critical analysis .................................................... 80<br />
6.1 Body image <strong>and</strong> body schema .......................................................................... 81<br />
6.2 Case studies in social psychology .................................................................... 83<br />
6.2.1 Non-pathological face-to-face interaction ................................................. 83<br />
6.2.2 Facial imitation by human neonates .......................................................... 84<br />
6.2.3 Gesturing by a deafferented subject (I)..................................................... 87<br />
6.2.4 Perceptual crossing in a virtual space ....................................................... 90<br />
6.2.5 Deafferented subject under a blind ........................................................... 94<br />
6.2.6 Gesturing by a deafferented subject (II).................................................... 96<br />
6.2.7 Bodily coordination in a virtual space (I) ................................................. 99<br />
6.2.8 Bodily coordination in a virtual space (II) .............................................. 102<br />
6.3 An integrative motor <strong>the</strong>ory ........................................................................... 104<br />
4 | P a g e
6.4 Summary ........................................................................................................ 111<br />
7 Toward <strong>the</strong> syn<strong>the</strong>sis of minimally social behavior .............................................. 114<br />
7.1 Evolutionary robotics ..................................................................................... 114<br />
7.2 An integrative methodology ........................................................................... 120<br />
7.3 Implementation details ................................................................................... 124<br />
8 Investigating sensitivity to social contingency ..................................................... 128<br />
8.1 Methods .......................................................................................................... 130<br />
8.2 Results ............................................................................................................ 132<br />
8.3 Behavioral analysis ......................................................................................... 134<br />
8.4 Dynamical analysis ......................................................................................... 138<br />
8.5 Summary ........................................................................................................ 140<br />
9 Investigating <strong>the</strong> interaction process ..................................................................... 143<br />
9.1 Methods .......................................................................................................... 144<br />
9.2 Experiments .................................................................................................... 148<br />
9.2.1 Experimental setup 1: Original setup ...................................................... 148<br />
9.2.2 Experimental setup 2: Switched receptor fields ...................................... 157<br />
9.2.3 Experimental setup 3: Conflicting behaviors .......................................... 160<br />
9.3 Dynamical analysis ......................................................................................... 166<br />
9.4 Discussion ...................................................................................................... 172<br />
9.5 Summary ........................................................................................................ 175<br />
10 Investigating social interaction ............................................................................. 177<br />
10.1 Experimental setup 4: Infinitely small objects ........................................... 178<br />
10.2 Experimental setup 5: Maximally distant shadows .................................... 182<br />
10.3 Experimental setup 6: Coordinated behavior.............................................. 184<br />
10.4 Summary ..................................................................................................... 190<br />
10.5 Discussion ................................................................................................... 192<br />
11 Beyond methodological physicalism .................................................................... 197<br />
12 Phenomenological considerations ......................................................................... 201<br />
12.1 The phenomenology of perception ............................................................. 201<br />
12.2 The phenomenology of intersubjectivity .................................................... 209<br />
12.3 A phenomenologically informed <strong>continuity</strong> <strong>the</strong>sis ..................................... 213<br />
13 Toward an enactive approach to culture ............................................................... 218<br />
13.1 The „ratchet effect‟...................................................................................... 219<br />
5 | P a g e
13.2 Primatology ................................................................................................. 221<br />
13.3 Developmental <strong>and</strong> social psychology ....................................................... 224<br />
13.4 Evolutionary anthropology ......................................................................... 227<br />
14 Conclusion ............................................................................................................ 232<br />
15 References ............................................................................................................. 234<br />
6 | P a g e
Acknowledgments<br />
I would like to give special thanks to my supervisor Ezequiel Di Paolo for being a much<br />
needed critical filter for <strong>the</strong> stream of ideas that were produced by my irresistible urge<br />
to adopt <strong>the</strong> working assumption that <strong>the</strong>re is never an end to relevant context. I am also<br />
appreciative of my D.Phil. research committee Inman Harvey <strong>and</strong> Anil Seth for keeping<br />
me focused <strong>and</strong> on target. Many thanks are owed to my academic colleagues, especially<br />
everyone from <strong>the</strong> CCNR <strong>and</strong> PAICS, as well as <strong>the</strong> participants of <strong>the</strong> Life <strong>and</strong> Mind<br />
seminars for <strong>the</strong>ir many helpful discussions.<br />
Of course, I would not be where I am now without my parents Rainer <strong>and</strong> Sabine, <strong>and</strong><br />
my sister Nele. I dedicate this <strong>the</strong>sis to <strong>the</strong>m. I am also extremely grateful to my friends<br />
<strong>and</strong> especially to my partner Iliana for <strong>the</strong> constant support, encouragement <strong>and</strong> friendly<br />
background noise that kept me sane in those moments of crisis. Without <strong>the</strong> invaluable<br />
help of this extended family <strong>the</strong> completion of <strong>the</strong>sis would have not been possible.<br />
Some of <strong>the</strong> chapters of this <strong>the</strong>sis have benefited from extensive comments made by<br />
anonymous reviewers as a result of <strong>the</strong>ir being published elsewhere. In particular,<br />
Chapter 2 is based on a paper that appeared as <strong>Froese</strong> (2007). Chapter 3 includes large<br />
parts from <strong>Froese</strong> <strong>and</strong> Ziemke (2009), as well as some text from <strong>Froese</strong> <strong>and</strong> Di Paolo<br />
(2009). Chapter 4 grew out of some of <strong>the</strong> ideas presented in De Jaegher <strong>and</strong> <strong>Froese</strong><br />
(2009). A shorter version of Chapter 8 has previously been published as <strong>Froese</strong> <strong>and</strong> Di<br />
Paolo (2008a). Chapters 9 <strong>and</strong> 10 are based on <strong>Froese</strong> <strong>and</strong> Di Paolo (in press-a) <strong>and</strong> (in<br />
press-b), respectively. The content of Chapter 12 is largely taken from <strong>Froese</strong> <strong>and</strong> Di<br />
Paolo (2009). I am indebted to all <strong>the</strong> reviewers for <strong>the</strong>ir criticisms <strong>and</strong> comments on<br />
how to improve <strong>the</strong> manuscripts. Recognition is also due to Shaun Gallagher <strong>and</strong> Mike<br />
Beaton for providing detailed comments on Chapter 6 <strong>and</strong> 13, respectively. I would also<br />
like to give many thanks to my examiners Phil Husb<strong>and</strong>s <strong>and</strong> Mike Wheeler for <strong>the</strong>ir<br />
constructive feedback. Finally, I am grateful to Eörs Szathmáry <strong>and</strong> <strong>the</strong> Collegium<br />
Budapest for hosting me during <strong>the</strong> summer of 2006, where some of <strong>the</strong> initial ideas for<br />
Chapter 4 took shape, as well as to <strong>Tom</strong> Ziemke <strong>and</strong> <strong>the</strong> University of Skövde for<br />
hosting me for a few months in 2007 <strong>and</strong> 2008 (with <strong>the</strong> generous financial assistance<br />
of <strong>the</strong> euCognition network), where substantial parts of Chapter 3 were written.<br />
7 | P a g e
1 Introduction<br />
Out of all <strong>the</strong> traditional difficulties faced by mainstream cognitive science, <strong>the</strong> <strong>mind</strong>body<br />
problem has all <strong>the</strong> makings of morphing into a paradigm buster. Even though<br />
consciousness has recently started to become a hot topic in science, it is still not clear<br />
what precisely <strong>the</strong> nature of <strong>the</strong> dilemma is, let alone what form a systematic response<br />
should take. The status of <strong>the</strong> problem, recently selected as one of <strong>the</strong> top outst<strong>and</strong>ing<br />
problems in a special issue of Science (cf. Miller 2005), indicates that <strong>the</strong>re is more at<br />
stake than revising our underst<strong>and</strong>ing of mentality: it can potentially challenge a<br />
particular way of doing science that dates back to <strong>the</strong> problem‟s Cartesian origins in <strong>the</strong><br />
scientific revolution of <strong>the</strong> 17 th century.<br />
While most cognitive scientists continue <strong>the</strong> attempt to somehow get a grip on <strong>the</strong> <strong>mind</strong>body<br />
problem within <strong>the</strong> conventional Cartesian framework, this <strong>the</strong>sis is part of a<br />
growing trend to change <strong>the</strong> fundamental terms of <strong>the</strong> debate. More specifically, it<br />
builds on what has become known as enactive cognitive science, an approach which<br />
replaces <strong>the</strong> computer metaphor of <strong>mind</strong> with a focus on <strong>life</strong> – a phenomenon that<br />
incorporates body <strong>and</strong> <strong>mind</strong> as two aspects of a unified whole. By placing <strong>the</strong><br />
phenomenon of <strong>life</strong> at <strong>the</strong> heart of its conceptual framework, <strong>the</strong> enactive paradigm has<br />
turned <strong>the</strong> intractable <strong>mind</strong>-body problem into a novel research program that is based on<br />
<strong>the</strong> principles of biological autonomy <strong>and</strong> phenomenological philosophy.<br />
However, this shift in terms of <strong>the</strong> debate from computer science to what might be<br />
called „bio-phenomenology‟ has also made <strong>the</strong> enactive approach vulnerable to <strong>the</strong><br />
criticism that its foundational principles, which are largely based on minimal forms of<br />
<strong>life</strong>, are irrelevant for <strong>the</strong> interests of cognitive science. In particular, proponents of <strong>the</strong><br />
Cartesian mainstream, who prefer to treat <strong>the</strong> human <strong>mind</strong> as a computer, have argued<br />
that such biological foundations are essentially incapable of accounting for „higher‟<br />
cognitive faculties. And, indeed, even though <strong>the</strong> explicit working hypo<strong>the</strong>sis of <strong>the</strong><br />
enactive approach is that <strong>the</strong>re actually is <strong>continuity</strong> between <strong>life</strong> <strong>and</strong> <strong>mind</strong>, so far it has<br />
been difficult – if not impossible – to conceive of a satisfactory way to bridge <strong>the</strong><br />
„cognitive gap‟ that lies, for example, between <strong>the</strong> capacity for adaptive behavior of a<br />
simple bacterium <strong>and</strong> <strong>the</strong> ability for abstract cognition of an adult human being.<br />
8 | P a g e
Accordingly, in response to this situation <strong>the</strong> main goal of this <strong>the</strong>sis is to argue for two<br />
complementary claims: (i) that <strong>the</strong> apparent inconceivability of a satisfactory version of<br />
<strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis largely results from <strong>the</strong> widely unquestioned assumption<br />
of methodological individualism in cognitive science (an assumption which treats all<br />
cognition as essentially an individual achievement), <strong>and</strong> (ii) that <strong>the</strong> enactive approach<br />
has <strong>the</strong> means to bridge this cognitive gap in a principled manner through a systematic<br />
consideration of <strong>the</strong> constitutive role of sociality for <strong>mind</strong> <strong>and</strong> behavior.<br />
In Chapter 2 <strong>the</strong> stage for this twofold argument is set by means of a brief history of<br />
recent cognitive science, framed in terms of issues related to philosophy of science. In<br />
particular, several possibilities for irresolvable stalemates between different paradigms<br />
are identified. Special emphasis is placed on <strong>the</strong> role of research in artificial intelligence<br />
(AI) <strong>and</strong> robotics in breaking a longst<strong>and</strong>ing philosophical stalemate, <strong>and</strong> supporting <strong>the</strong><br />
subsequent turn toward more embodied-embedded approaches. It is argued that progress<br />
in this experimental domain is never<strong>the</strong>less still threatened by an empirical stalemate,<br />
which is related to <strong>the</strong> necessity of an observer to adopt some interpretative perspective<br />
in order to make sense of <strong>the</strong> experimental data. There is thus a need for analyzing <strong>the</strong><br />
constitutive conditions of our scientific perspective, a task which motivates <strong>the</strong> role of<br />
phenomenology for <strong>the</strong> development of <strong>the</strong> enactive paradigm. This background chapter<br />
<strong>the</strong>refore acts as a first introduction to <strong>the</strong> three main approaches pursued in <strong>the</strong> <strong>the</strong>sis,<br />
namely <strong>the</strong>oretical argumentation, experimental investigation, <strong>and</strong> phenomenological<br />
observation, as well as to some of <strong>the</strong> basic ideas of <strong>the</strong> enactive paradigm.<br />
This general introduction is followed in Chapter 3 by a more detailed description of <strong>the</strong><br />
conceptual framework of enactive cognitive science. The <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis is<br />
presented as a strong working hypo<strong>the</strong>sis that has <strong>the</strong> potential to become a unified<br />
<strong>the</strong>ory of cognitive science. Some outst<strong>and</strong>ing problems with <strong>the</strong> <strong>continuity</strong> <strong>the</strong>sis are<br />
identified, especially what we call <strong>the</strong> „cognitive gap‟: <strong>the</strong> seemingly insurmountable<br />
distance between <strong>the</strong> basic phenomena of <strong>life</strong> <strong>and</strong> <strong>the</strong> higher cognitive functions of adult<br />
human beings. It is suggested that this perceived problem is largely due to <strong>the</strong><br />
methodological individualism that is present in most cognitive science, <strong>and</strong> that a<br />
consideration of <strong>the</strong> constitutive role of sociality from <strong>the</strong> perspective of <strong>the</strong> enactive<br />
9 | P a g e
approach can systematically resolve this issue. As a first step toward this goal, <strong>the</strong><br />
biological foundations of enactive cognitive science are introduced, in particular <strong>the</strong><br />
notions of autonomy <strong>and</strong> sense-making, which denote a system‟s capacity to generate<br />
its own identity under precarious conditions <strong>and</strong> <strong>the</strong> capacity to adaptively regulate its<br />
interactions in relation to those conditions, respectively. The chapter finishes by briefly<br />
indicating how <strong>the</strong> notions of autonomy <strong>and</strong> sense-making inform <strong>the</strong> current version of<br />
<strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis.<br />
The second step follows in Chapter 4 which provides a critical analysis of what has<br />
already been published about <strong>the</strong> enactive approach to social cognition. The starting<br />
point is an appraisal of <strong>the</strong> claim that <strong>the</strong> defining aspect of social interaction is its<br />
autonomy, i.e. that <strong>the</strong> interaction process between two or more interacting agents can<br />
itself take on an autonomous organization <strong>and</strong> <strong>the</strong>reby effectively organize <strong>the</strong> behavior<br />
of those interactors to exp<strong>and</strong> (or constrain) <strong>the</strong>ir individual domains of interaction. It is<br />
argued that <strong>the</strong> autonomy of <strong>the</strong> interaction process is a necessary but not sufficient<br />
condition for social interaction, <strong>and</strong> that this necessary condition is better captured by<br />
<strong>the</strong> notion of „multi-agent interaction‟. Accordingly, a revised definition of social<br />
interaction is offered: it is a type of multi-agent interaction whereby an agent‟s action<br />
necessarily requires an appropriate response by ano<strong>the</strong>r agent for its completion. This<br />
co-regulation of activity opens up specifically social ways of sense-making, i.e. forms<br />
of participatory sense-making, <strong>and</strong> <strong>the</strong>reby introduces a qualitative change to <strong>the</strong> agents‟<br />
cognitive domains. However, this type of social interaction is still not sufficient to<br />
account for <strong>the</strong> specificity of cultural forms of interaction, which depend on pre-existing<br />
practices, <strong>and</strong> a provisional account of cultural interaction is suggested that takes such<br />
heteronomy into consideration. Each of <strong>the</strong>se transitions in sociality entails constitutive<br />
changes to <strong>the</strong> structures of agency which originally give rise to <strong>the</strong>m, <strong>and</strong> <strong>the</strong>reby lead<br />
to increases in an individual‟s behavioral capacity. In this way <strong>the</strong> enactive approach to<br />
social cognition provides a <strong>the</strong>oretical opening for a research program that addresses <strong>the</strong><br />
cognitive gap of <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis by means of a systematic investigation<br />
of <strong>the</strong> constitutive role of sociality for <strong>mind</strong> <strong>and</strong> behavior.<br />
The development of novel definitions for multi-agent systems <strong>and</strong> social interaction<br />
completes <strong>the</strong> <strong>the</strong>oretical part of <strong>the</strong> <strong>the</strong>sis. Chapter 5 provides a brief recap of how <strong>the</strong><br />
10 | P a g e
enactive approach to social cognition has responded to methodological individualism,<br />
<strong>and</strong> it situates this achievement in a wider scientific <strong>and</strong> historical context. The next part<br />
of <strong>the</strong> <strong>the</strong>sis is concerned with experimental evidence.<br />
The aim of Chapter 6 is to show that <strong>the</strong> enactive approach to social cognition can be<br />
used to provide a fresh interpretation of some important experiments in developmental<br />
<strong>and</strong> social psychology. The results of <strong>the</strong>se experiments are critically analyzed in order<br />
to reveal <strong>the</strong> significant problems that are faced by <strong>the</strong> traditional explanations based on<br />
methodological individualism. These problems serve as a motivation to broaden <strong>the</strong><br />
acceptable range of explanations to include a consideration of <strong>the</strong> constitutive role of<br />
<strong>the</strong> interaction process. It is argued that o<strong>the</strong>rwise even <strong>the</strong> integrative explanations in<br />
embodied-embedded cognitive science can be forced to return to <strong>the</strong> traditional method<br />
of postulating hypo<strong>the</strong>tical neuro-physiological structures to explain <strong>the</strong> existence of<br />
social phenomena. This chapter also provides <strong>the</strong> empirical backdrop for some of <strong>the</strong><br />
modeling experiments presented in subsequent chapters.<br />
The methodology for <strong>the</strong>se simulation models, namely <strong>the</strong> development of a mutually<br />
informative relationship between <strong>the</strong> artificial <strong>and</strong> empirical sciences, is introduced in<br />
Chapter 7. In particular, <strong>the</strong> aim is to promote a dialogue between evolutionary robotics<br />
<strong>and</strong> social psychology. The models are mainly used as useful tools for thinking that can<br />
challenge established positions <strong>and</strong> explanations, serve as proof of concepts, <strong>and</strong> lead to<br />
<strong>the</strong> generation of novel predictions <strong>and</strong> hypo<strong>the</strong>ses. In all cases <strong>the</strong> intention is to<br />
capture relevant phenomena in <strong>the</strong> most minimalist manner possible such that <strong>the</strong>se<br />
insights do not get lost in unnecessary complexity. The next three chapters of <strong>the</strong> <strong>the</strong>sis<br />
present novel modeling experiments.<br />
To begin with, Chapter 8 presents a model of a famous psychological experiment on<br />
infants‟ sensitivity to social contingency. The results demonstrate that, contrary to<br />
traditional expectations, it is not necessary to postulate innate cognitive modules in<br />
order to explain this capacity, <strong>and</strong> that a consideration of <strong>the</strong> interaction process itself<br />
could provide a more parsimonious explanation of <strong>the</strong> empirical data. Due to <strong>the</strong><br />
minimalism of <strong>the</strong> model it is also possible to give a detailed dynamical explanation of<br />
<strong>the</strong> evolved behavior. It is shown that it is <strong>the</strong> interaction process itself which invests<br />
11 | P a g e
<strong>the</strong>se agents with <strong>the</strong> role of interactors, because <strong>the</strong> mutual interaction perturbs <strong>the</strong>ir<br />
structure such that <strong>the</strong> behavior required for <strong>the</strong> interaction becomes possible. Isolated<br />
agents have a more limited behavioral domain.<br />
The aim of <strong>the</strong> model presented in Chapter 9 is to fur<strong>the</strong>r investigate how <strong>the</strong> interaction<br />
process itself can organize <strong>the</strong> behavior of individuals. This is achieved by modeling a<br />
recent psychological experiment that was also specifically designed for this purpose. A<br />
number of modifications to <strong>the</strong> original experimental design demonstrate <strong>the</strong> robustness<br />
of <strong>the</strong> interaction process to organize behaviors even under impaired <strong>and</strong> unfavorable<br />
conditions. The results of <strong>the</strong>se modifications lead to <strong>the</strong> generation of novel hypo<strong>the</strong>ses<br />
that are open to verification by future psychological experiments. The simplicity of <strong>the</strong><br />
model also allows a detailed dynamical description of <strong>the</strong> individuals‟ behavior <strong>and</strong> how<br />
this behavior is constituted by <strong>the</strong> interaction, <strong>the</strong>reby generating skepticism about<br />
traditional ways of schematizing sub-personal processes. It is shown that even simple<br />
multi-agent interactions can exp<strong>and</strong> individual behavioral domains by a process of codetermination<br />
of agential structures.<br />
Chapter 10 fur<strong>the</strong>r explores some of <strong>the</strong> implications of this model by fine-tuning <strong>the</strong><br />
experimental design so as to fur<strong>the</strong>r reduce <strong>the</strong> potential for agents to rely on individualbased<br />
behavioral strategies. The results of <strong>the</strong>se modifications lead to novel predictions<br />
about what precisely are <strong>the</strong> essential elements of <strong>the</strong> experimental setup of <strong>the</strong> original<br />
psychological study. Moreover, a simple modification to <strong>the</strong> task requiring coordinated<br />
behavior results in a model of social interaction, as defined in Chapter 4. The results<br />
demonstrate that this particular type of interaction process can fur<strong>the</strong>r increase <strong>the</strong><br />
behavioral repertoire of <strong>the</strong> agents. The model leads to a novel hypo<strong>the</strong>sis about <strong>the</strong><br />
minimal conditions for human participants of <strong>the</strong> psychological study to experience <strong>the</strong><br />
experimental situation as qualitatively social.<br />
In all of <strong>the</strong>se modeling experiments <strong>the</strong> minimalist approach afforded by evolutionary<br />
robotics is demonstrated as an effective antidote against <strong>the</strong> widespread assumption of<br />
methodological individualism, especially because it is possible to resolve doubts in a<br />
non-mysterious manner by providing detailed dynamical accounts of <strong>the</strong> constitutive<br />
role of <strong>the</strong> interaction process. This completes <strong>the</strong> experimental contribution of this<br />
12 | P a g e
<strong>the</strong>sis. It has been argued that <strong>the</strong> enactive approach to social cognition can provide a<br />
<strong>the</strong>oretical framework to close <strong>the</strong> cognitive gap. These modeling experiments have<br />
demonstrated that this <strong>the</strong>ory can be put into scientific practice <strong>and</strong> that <strong>the</strong> results are<br />
amenable to analysis in dynamical terms.<br />
In Chapter 11 this improved dynamical underst<strong>and</strong>ing of <strong>the</strong> constitutive role of<br />
sociality is placed into a wider scientific <strong>and</strong> historical context. It is argued that a simple<br />
rejection of methodological individualism based on this kind of systems <strong>the</strong>ory alone is<br />
not sufficient to break out of <strong>the</strong> conventional framework entirely. For that to happen it<br />
is also necessary to complement this work with ano<strong>the</strong>r defining aspect of enactive<br />
cognitive science, namely experiential considerations. This phenomenological approach<br />
reveals ano<strong>the</strong>r widespread assumption that has limited mainstream approaches to social<br />
cognition: <strong>the</strong> idea that <strong>the</strong> primary function of perception is to process information<br />
about an independent world of abstract physical quantities. We refer to this assumption<br />
as „methodological physicalism‟. It has led much mainstream research in <strong>the</strong> field of<br />
social cognition to be concentrated on <strong>the</strong> „problem of o<strong>the</strong>r <strong>mind</strong>s‟, i.e. <strong>the</strong> question of<br />
how underst<strong>and</strong>ing of o<strong>the</strong>rs is possible on <strong>the</strong> basis of perceiving <strong>the</strong>ir abstract physical<br />
details alone. This misguided but deeply engrained focus has regrettably come at <strong>the</strong><br />
expense of a more phenomenologically plausible research program.<br />
Fortunately, <strong>the</strong> enactive paradigm has <strong>the</strong> capacity to provide an effective remedy to<br />
methodological physicalism (<strong>and</strong> those aspects of methodological individualism that are<br />
derived from it) by appealing to careful phenomenological analyses of our immediate<br />
experience. Thus, Chapter 12 introduces some central insights of <strong>the</strong> phenomenology of<br />
intersubjectivity. In particular, <strong>the</strong> consideration of intersubjectivity is motivated by a<br />
critical analysis of our perception of objects. It is argued that <strong>the</strong> experience of an object<br />
as independent of our current perspective of concern is constitutively dependent on what<br />
Husserl calls „open intersubjectivity‟, i.e. <strong>the</strong> potential presence of o<strong>the</strong>r perspectives in<br />
<strong>the</strong> world. This opens up <strong>the</strong> way for more detailed observations of how o<strong>the</strong>r subjects<br />
appear in our experience, <strong>and</strong> how <strong>the</strong>ir presence impacts on how we make sense of <strong>the</strong><br />
world. In particular, it is argued that <strong>the</strong> categories of objectivity <strong>and</strong> subjectivity are<br />
impossible to appreciate experientially without open intersubjectivity. Accordingly, <strong>the</strong><br />
phenomenological perspective enables us to refine <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis from a<br />
13 | P a g e
„top-down‟ perspective, namely by starting from <strong>the</strong> specificity of human (inter-)<br />
subjectivity. This perspective provides novel insights into <strong>the</strong> qualitative dimensions of<br />
<strong>the</strong> <strong>continuity</strong> <strong>the</strong>sis that complement <strong>the</strong> „bottom-up‟ approach of previous chapters.<br />
The phenomenological return to our immediate experience provides not only a vantage<br />
point from which to question <strong>the</strong> validity of methodological physicalism, but also makes<br />
it possible for us to take a fresh perspective on some controversial empirical data. This<br />
is <strong>the</strong> task of Chapter 13, which completes <strong>the</strong> investigation of sociality by focusing on<br />
some crucial aspects of cumulative cultural development. More specifically, it proposes<br />
to turn <strong>the</strong> traditional framework in primatology <strong>and</strong> infant studies on its head by means<br />
of a novel explanation of cumulative cultural development based on <strong>the</strong> systemic <strong>and</strong><br />
phenomenological accounts of sociality of <strong>the</strong> enactive paradigm. Interestingly, this<br />
perspective reveals a blind spot in <strong>the</strong> primary literature, which leaves unaccounted <strong>the</strong><br />
capacity of humans (<strong>and</strong> enculturated chimpanzees) to perceive o<strong>the</strong>rs in terms of <strong>the</strong>ir<br />
abstract physical properties, an ability that is necessary for imitative learning (a primary<br />
mechanism of cultural development). Some empirical evidence is presented which, in<br />
combination with <strong>the</strong> phenomenological insights developed in Chapter 12, points to a<br />
socially mediated origin of this perceptual capacity. Finally, once cumulative cultural<br />
development is underway, it appears that it takes on properties that can be captured by<br />
concepts akin to <strong>the</strong> basic organizational principles of <strong>life</strong>.<br />
On <strong>the</strong> basis of <strong>the</strong> <strong>the</strong>oretical, experimental, <strong>and</strong> phenomenological insights developed<br />
in this <strong>the</strong>sis it is concluded in Chapter 14 that <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis of <strong>the</strong><br />
enactive paradigm is indeed a viable working hypo<strong>the</strong>sis for cognitive science. It has<br />
been shown that <strong>the</strong> organizational principles which can be derived from minimal forms<br />
of <strong>life</strong>, complemented by phenomenological considerations, can help us to underst<strong>and</strong> in<br />
a unified manner <strong>the</strong> processes which connect individual agency <strong>and</strong> simple interaction<br />
processes to human agency <strong>and</strong> cultural cognition. At <strong>the</strong> heart of this underst<strong>and</strong>ing lie<br />
<strong>the</strong> complementary notions of biological autonomy <strong>and</strong> enacted meaning. More work<br />
surely needs to be done, but this <strong>the</strong>sis has contributed to <strong>the</strong> beginnings of a research<br />
program that has <strong>the</strong> potential to provide a unified <strong>the</strong>ory of cognitive science. In fact,<br />
we can expect that this approach will not only impact how we scientifically approach<br />
<strong>life</strong>, <strong>mind</strong> <strong>and</strong> sociality, but also how we perceive ourselves, o<strong>the</strong>rs <strong>and</strong> <strong>the</strong> world.<br />
14 | P a g e
2 A brief history of cognitive science<br />
Over <strong>the</strong> last two decades <strong>the</strong> field of artificial intelligence (AI) has undergone some<br />
significant developments (cf. Anderson 2003; <strong>Froese</strong> & Ziemke 2009). Good oldfashioned<br />
AI (GOFAI) has faced considerable problems whenever it attempts to extend<br />
its domain beyond simplified „toy worlds‟ in order to address context-sensitive realworld<br />
problems in a robust <strong>and</strong> flexible manner (<strong>Dr</strong>eyfus 1981; 1972). A few wellknown<br />
examples are <strong>the</strong> commonsense knowledge problem (<strong>Dr</strong>eyfus 1991, p. 119), <strong>the</strong><br />
frame problem (McCarthy & Hayes 1969), <strong>and</strong> <strong>the</strong> symbol grounding problem (Harnad<br />
1990). These difficulties motivated <strong>the</strong> Brooksian revolution toward an embodied <strong>and</strong><br />
situated robotics in <strong>the</strong> early 1990s (Brooks 1991a; 1991b). Since <strong>the</strong>n this approach has<br />
been fur<strong>the</strong>r developed (e.g. Pfeifer & Scheier 1999; Pfeifer 1996; Brooks 1997), <strong>and</strong><br />
has also significantly influenced <strong>the</strong> emergence of a variety of o<strong>the</strong>r successful<br />
methodologies, such as <strong>the</strong> dynamical approach (e.g. Beer 1995a; 2003), evolutionary<br />
robotics (e.g. Harvey et al. 2005; Nolfi & Floreano 2000; Cliff, et al. 1993), <strong>and</strong><br />
organismically-inspired robotics (e.g. Di Paolo 2003; Iizuka & Di Paolo 2007a; 2008;<br />
Wood & Di Paolo 2008). These approaches are united by <strong>the</strong> claim that cognition is best<br />
understood as embodied <strong>and</strong> embedded in <strong>the</strong> sense that it emerges out of <strong>the</strong> dynamics<br />
of an extended brain-body-world systemic whole.<br />
These developments make it evident that <strong>the</strong> traditional GOFAI mainstream, with its<br />
emphasis on perception as representation <strong>and</strong> cognition as computation, is being<br />
challenged by <strong>the</strong> establishment of an alternative paradigm in <strong>the</strong> form of embodiedembedded<br />
AI. How is this major shift in AI related to <strong>the</strong> ongoing paradigm shift within<br />
<strong>the</strong> cognitive sciences 1 ? Section 2.1 analyzes <strong>the</strong> role of AI in <strong>the</strong> emergence of what<br />
has been called „embodied-embedded‟ cognitive science (e.g. Clark 1997; Wheeler<br />
2005). Recently, <strong>the</strong>re has also been a noticeable shift in interest toward „enactive‟<br />
cognitive science (e.g. Thompson 2007; Di Paolo, et al., in press), a paradigm which<br />
radicalizes <strong>the</strong> embodied-embedded approach by placing autonomous agency <strong>and</strong> lived<br />
1 Whe<strong>the</strong>r any of <strong>the</strong> major changes in AI or cognitive science are in fact paradigm shifts in <strong>the</strong> strict<br />
sense introduced by Kuhn (1962) is an interesting open question but beyond <strong>the</strong> scope of this chapter.<br />
Here <strong>the</strong> notion is used in <strong>the</strong> more general sense of a major shift in experimental practice <strong>and</strong> focus.<br />
15 | P a g e
subjectivity at <strong>the</strong> heart of cognitive science. How <strong>the</strong> field of AI relates to this fur<strong>the</strong>r<br />
shift is still in need of clarification. The rest of this chapter provides some initial steps in<br />
this direction by providing a general introduction to <strong>the</strong> conceptual framework of <strong>the</strong><br />
enactive paradigm (Sections 2.2). Never<strong>the</strong>less, some methodological worries still<br />
remain (Section 2.3). The brief history of cognitive science concludes with some<br />
remarks about <strong>the</strong> need for a practice-oriented phenomenology, especially when trying<br />
to promote a more widespread acceptance of <strong>the</strong> enactive approach (Section 2.4).<br />
2.1 Toward embodied-embedded cognitive science<br />
Much of contemporary cognitive science owes its existence to <strong>the</strong> founding of <strong>the</strong> field<br />
of AI in <strong>the</strong> late 1950s by <strong>the</strong> likes of Herbert Simon, Marvin Minsky, Allen Newell,<br />
<strong>and</strong> John McCarthy 2 . These researchers, along with Noam Chomsky, put forth ideas that<br />
were to become <strong>the</strong> major guidelines for <strong>the</strong> computational approach which has<br />
dominated <strong>the</strong> cognitive sciences since its inception (cf. Boden 2006a). In order to<br />
determine <strong>the</strong> impact of AI on <strong>the</strong> ongoing shift from such orthodox computationalism<br />
toward embodied-embedded cognitive science, it is necessary to briefly consider some<br />
of <strong>the</strong> central claims associated with <strong>the</strong>se competing <strong>the</strong>oretical frameworks.<br />
The paradigm that came into existence with <strong>the</strong> birth of AI, <strong>and</strong> which was essentially<br />
identified with cognitive science itself for <strong>the</strong> ensuing three decades <strong>and</strong> which still<br />
represents <strong>the</strong> mainstream today, is known as cognitivism (e.g. Fodor 1975). The<br />
cognitivist claim, that cognition is a form of computation (i.e. information processing<br />
through <strong>the</strong> manipulation of symbolic representations), is famously articulated in <strong>the</strong><br />
„Physical-Symbol System Hypo<strong>the</strong>sis‟ which holds that such a system has <strong>the</strong> necessary<br />
<strong>and</strong> sufficient means for general intelligent action (Newell & Simon 1976). From <strong>the</strong><br />
cognitivist perspective cognition is essentially a centrally controlled, disembodied, <strong>and</strong><br />
2 The origins of this early symbolic AI, <strong>and</strong> <strong>the</strong> computationalist cognitive science that was to be founded<br />
on it, can be traced to <strong>the</strong> influential cybernetics tradition of <strong>the</strong> „40s <strong>and</strong> „50s, which is best known for<br />
<strong>the</strong> work by Wiener, von Neumann <strong>and</strong> o<strong>the</strong>r participants of <strong>the</strong> Macy conferences (cf. Dupuy 2009). The<br />
enactive paradigm has related roots, though it was influenced more by British cyberneticists such as Pask<br />
<strong>and</strong> Ashby (cf. Husb<strong>and</strong>s, et al. 2008), as well as by <strong>the</strong> „second-order cybernetics‟ of von Foerster <strong>and</strong> its<br />
fur<strong>the</strong>r development into Maturana <strong>and</strong> Varela‟s „biology of cognition‟ (cf. Varela 1996a).<br />
16 | P a g e
decontextualized reasoning <strong>and</strong> planning algorithm as epitomized by abstract problem<br />
solving. Accordingly, <strong>the</strong> <strong>mind</strong> is conceptualized as a digital computer <strong>and</strong> cognition is<br />
viewed as fundamentally distinct from <strong>the</strong> embodied action of an autonomous agent that<br />
is situated within <strong>the</strong> continuous dynamics of its environment.<br />
The cognitivist orthodoxy remained unchallenged until connectionism arose in <strong>the</strong> early<br />
1980s (e.g. McClell<strong>and</strong>, Rumelhart et al. 1986). The connectionist alternative views<br />
cognition as <strong>the</strong> emergence of global states in a network of simple components, <strong>and</strong><br />
promises to address two practical shortcomings of cognitivism, namely by (i) increasing<br />
efficiency through parallel processing, <strong>and</strong> (ii) achieving greater robustness through<br />
distributed operations. Moreover, because it makes use of artificial neural networks as a<br />
metaphor for <strong>the</strong> <strong>mind</strong>, its <strong>the</strong>ories of cognition are often more biologically plausible.<br />
Never<strong>the</strong>less, connectionism still retains many cognitivist commitments. In particular, it<br />
maintains <strong>the</strong> idea that cognition is essentially a form of information processing in <strong>the</strong><br />
head which converts a set of inputs into an appropriate set of outputs in order to solve a<br />
given problem. In o<strong>the</strong>r words, “connectionism‟s disagreement with cognitivism was<br />
over <strong>the</strong> nature of computation <strong>and</strong> representation (symbolic for cognitivists,<br />
subsymbolic for connectionsists)” (Thompson 2007, p. 10), ra<strong>the</strong>r than over <strong>the</strong> notion<br />
of computationalism as such (see also Wheeler 2005, p. 75). Accordingly, most of<br />
connectionism can be regarded as constituting a part of orthodox cognitive science.<br />
Since <strong>the</strong> early 1990s this computationalist orthodoxy has begun to be challenged by <strong>the</strong><br />
emergence of embodied-embedded cognitive science (cf. Clark 1997; Wheeler 2005), a<br />
paradigm which claims that an agent‟s embodiment is constitutive of its perceiving,<br />
knowing <strong>and</strong> doing (e.g. Gallagher 2005; Noë 2004; Varela, et al. 1991; Thompson &<br />
Varela 2001). Fur<strong>the</strong>rmore, <strong>the</strong> computational hypo<strong>the</strong>sis has been confronted by <strong>the</strong><br />
dynamical hypo<strong>the</strong>sis that cognitive agents are best understood as dynamical systems<br />
(van Gelder 1998; van Gelder & Port 1995). Thus, while <strong>the</strong> embodied-embedded<br />
paradigm has retained <strong>the</strong> connectionist focus on self-organizing dynamic systems, it<br />
fur<strong>the</strong>r holds that cognition is a situated activity which spans a systemic totality<br />
consisting of an agent‟s brain, body, <strong>and</strong> world (e.g. Beer 2000). In order to assess <strong>the</strong><br />
17 | P a g e
importance of AI for this ongoing shift toward embodied-embedded cognitive science, it<br />
is helpful to first consider <strong>the</strong> potential impact of <strong>the</strong>ory for this shift alone.<br />
The <strong>the</strong>oretical premises of orthodox <strong>and</strong> embodied-embedded cognitive science can<br />
generally be seen as Cartesian <strong>and</strong> Heideggerian in character, respectively (cf. Wheeler<br />
2005; <strong>Dr</strong>eyfus 2007; Anderson 2003). The traditional Cartesian philosophy accepts <strong>the</strong><br />
assumption that any kind of phenomena can be reduced to a combination of more basic<br />
atomic elements which are <strong>the</strong>mselves irreducible. On this view cognition is seen as a<br />
general-purpose reasoning process by which a relevant representation of <strong>the</strong> world is<br />
assembled through <strong>the</strong> appropriate manipulation <strong>and</strong> transformation of basic mental<br />
states. Orthodox cognitive science adopts a similar kind of reductionism in that it<br />
assumes that symbolic (or, in <strong>the</strong> case of connectionism, sub-symbolic) structures are<br />
<strong>the</strong> basic representational elements which ground all mental states 3 , <strong>and</strong> that cognition is<br />
essentially treated as <strong>the</strong> appropriate computation of such representations which pick<br />
out facts about <strong>the</strong> physical world. What are <strong>the</strong> arguments against such a position?<br />
The Heideggerian critique starts from <strong>the</strong> phenomenological claim that <strong>the</strong> world is first<br />
<strong>and</strong> foremost experienced as a significant whole <strong>and</strong> that cognition is grounded in <strong>the</strong><br />
skilful disposition to respond flexibly <strong>and</strong> appropriately as dem<strong>and</strong>ed by contextual<br />
circumstances. <strong>Dr</strong>eyfus (1991, p. 117) has argued that such a position questions <strong>the</strong><br />
validity of <strong>the</strong> Cartesian approach in two fundamental ways. First, <strong>the</strong> claim of holism<br />
entails that <strong>the</strong> isolation of a specific part or element of our experience as an atomic<br />
entity appears as secondary because it already presupposes a background of significance<br />
as <strong>the</strong> context from which to make <strong>the</strong> isolation. From this point of view a reductionist<br />
attempt at reconstructing a meaningful whole by combining isolated parts appears<br />
nonsensical since <strong>the</strong> required atomic elements were created by stripping away exactly<br />
that contextual significance in <strong>the</strong> first place:<br />
3 In contrast to <strong>the</strong> Cartesian claim that mental stuff is ontologically basic, orthodox cognitive science<br />
holds that <strong>the</strong>se constitutive elements are not basic in any metaphysical sense because <strong>the</strong>y are fur<strong>the</strong>r<br />
reducible to binary logic. And, even though it is only this domain which ultimately constitutes <strong>the</strong> mental,<br />
<strong>the</strong>re is no problem of it being realized in a physical system. Never<strong>the</strong>less, this change in position does<br />
not make any difference with regard to Heidegger‟s critique.<br />
18 | P a g e
Facts <strong>and</strong> rules are, by <strong>the</strong>mselves, meaningless. To capture what Heidegger calls<br />
significance or involvement, <strong>the</strong>y must be assigned relevance. But <strong>the</strong> predicates<br />
that must be added to define relevance are just more meaningless facts. (<strong>Dr</strong>eyfus<br />
1991, p. 118)<br />
From <strong>the</strong> Heideggerian perspective it <strong>the</strong>refore appears that <strong>the</strong> Cartesian position is<br />
faced with a problem of infinite regress. Second, if we accept <strong>the</strong> claim of skills, namely<br />
that cognition is essentially grounded in a kind of skilful know-how or context-sensitive<br />
coping, <strong>the</strong>n <strong>the</strong> orthodox aim of reducing such behaviour into a formal set of<br />
input/output mappings which specify <strong>the</strong> manipulation <strong>and</strong> transformation of basic<br />
mental states appears to be hopelessly misguided.<br />
Judging from <strong>the</strong>se philosophical considerations it seems that <strong>the</strong> Heideggerian critique<br />
of <strong>the</strong> Cartesian tradition could have a significant impact on <strong>the</strong> paradigm shift from<br />
orthodox toward embodied-embedded cognitive science. However, since <strong>the</strong> two<br />
approaches have distinct underlying constitutive assumptions (e.g. reductionism vs.<br />
holism), <strong>the</strong>re exists no a priori <strong>the</strong>oretical argument which would force someone<br />
holding a Cartesian position to accept <strong>the</strong> Heideggerian critique from holism <strong>and</strong> skills.<br />
Similarly, it is not possible for <strong>the</strong> Cartesian <strong>the</strong>orist to prove that worldly significance<br />
can indeed be created through <strong>the</strong> appropriate manipulation <strong>and</strong> transformation of<br />
abstract <strong>and</strong> de-contextualized representational elements. The problem is that, like all<br />
rational arguments, both accounts of cognition are founded on a particular set of<br />
premises which one is at liberty to accept or reject. Thus, even if <strong>the</strong> development of a<br />
strong philosophical position is most likely a necessary factor in <strong>the</strong> success of <strong>the</strong><br />
embodied-embedded paradigm, it is by itself not sufficient. In o<strong>the</strong>r words, <strong>the</strong>re is a<br />
fundamental stalemate in <strong>the</strong> purely philosophical domain; a shift in constitutive<br />
assumptions cannot be engendered by argumentation alone.<br />
It has often been proposed that this <strong>the</strong>oretical stalemate has to be resolved in <strong>the</strong><br />
empirical domain of <strong>the</strong> cognitive sciences (e.g. <strong>Dr</strong>eyfus & <strong>Dr</strong>eyfus 1988; Clark 1997,<br />
p. 169; Wheeler 2005, p. 187). The authors of <strong>the</strong> Physical-Symbol System Hypo<strong>the</strong>sis<br />
(Newell & Simon 1976) <strong>and</strong> <strong>the</strong> Dynamical Hypo<strong>the</strong>sis (van Gelder 1998) are also in<br />
agreement that only sustained empirical research can determine whe<strong>the</strong>r <strong>the</strong>ir respective<br />
19 | P a g e
hypo<strong>the</strong>ses are viable. Empirical research in AI is <strong>the</strong>reby awarded <strong>the</strong> ra<strong>the</strong>r privileged<br />
position of being able to help resolve <strong>the</strong>oretical disputes which have plagued <strong>the</strong><br />
Western philosophical tradition for decades if not centuries 4 . This reciprocal<br />
relationship between AI <strong>and</strong> <strong>the</strong>ory has been captured with <strong>the</strong> slogan „underst<strong>and</strong>ing by<br />
building‟ (e.g. Pfeifer 1996; Pfeifer & Scheier 1999, p. 299).<br />
In what way has AI research managed to fulfill this role? It can do so negatively, such as<br />
when insurmountable problems appear in practice. <strong>Dr</strong>eyfus (1991, p. 119), for example,<br />
has argued that <strong>the</strong> Heideggerian philosophy of cognition has been vindicated because<br />
GOFAI faces significant difficulties whenever it attempts to apply its Cartesian<br />
principles to real-world situations which require robust, flexible, <strong>and</strong> context-sensitive<br />
behavior. In addition, he demonstrates that <strong>the</strong> Heideggerian arguments from holism<br />
<strong>and</strong> skills can provide powerful explanations of why this kind of AI has to wrestle with<br />
<strong>the</strong> frame <strong>and</strong> commonsense knowledge problems.<br />
But AI can also fulfill this role positively, as when philosophical assumptions lead to <strong>the</strong><br />
successful design <strong>and</strong> implementation of actual systems. Wheeler (2005, p. 188), for<br />
instance, argues compellingly that <strong>the</strong> growing success of embodied-embedded AI<br />
provides important experimental support for <strong>the</strong> shift toward a Heideggerian position in<br />
cognitive science. He suggests that Heidegger‟s claim that a cognitive agent is best<br />
understood from <strong>the</strong> perspective of „being-in-<strong>the</strong>-world‟ is put to <strong>the</strong> test by embodiedembedded<br />
AI experiments which investigate cognition as a dynamical process which<br />
emerges out of a brain-body-world systemic whole.<br />
In light of <strong>the</strong>se developments it seems fair to say that AI can have a significant impact<br />
on <strong>the</strong> ongoing shift from orthodox toward embodied-embedded cognitive science.<br />
However, while embodied-embedded AI has managed to overcome some of <strong>the</strong><br />
significant challenges faced by traditional GOFAI, it has also started to encounter some<br />
4 It is worth noting that <strong>the</strong>re are compelling arguments for claiming that <strong>the</strong> results generated by AI<br />
research are not „empirical‟ in <strong>the</strong> same way as those of <strong>the</strong> natural sciences, <strong>and</strong> that this is likely to<br />
weaken <strong>the</strong>ir impact outside <strong>the</strong> field. Never<strong>the</strong>less, it is still <strong>the</strong> case that AI, just like a good empirical<br />
experiment, can provide valuable tools for re-organizing <strong>and</strong> probing <strong>the</strong> internal consistency of a<br />
<strong>the</strong>oretical position (cf. Di Paolo, et al. 2000). We will return to this issue in Chapter 7.<br />
20 | P a g e
of its own limitations. Considering <strong>the</strong> seemingly insurmountable challenge to make <strong>the</strong><br />
artificial agents of current embodied-embedded AI behave in a more robust, flexible,<br />
<strong>and</strong> generally more <strong>life</strong>-like manner, particularly in <strong>the</strong> way that more complex living<br />
organisms do, <strong>the</strong> embodied robotics pioneer Brooks was led to entertain <strong>the</strong> following<br />
skeptical reflections on <strong>the</strong> topic:<br />
Perhaps we have all missed some organizing principle of biological systems, or<br />
some general truth about <strong>the</strong>m. Perhaps <strong>the</strong>re is a way of looking at biological<br />
systems which will illuminate an inherent necessity in some aspect of <strong>the</strong><br />
interactions of <strong>the</strong>ir parts that is completely missing from our artificial systems.<br />
[…] I am suggesting that perhaps at this point we simply do not get it, <strong>and</strong> that<br />
<strong>the</strong>re is some fundamental change necessary in our thinking in order that we<br />
might build artificial systems that have <strong>the</strong> levels of intelligence, emotional<br />
interactions, long term stability <strong>and</strong> autonomy, <strong>and</strong> general robustness that we<br />
might expect of biological systems. (Brooks 1997, p. 301)<br />
Has <strong>the</strong> field of AI managed to find this missing „organizing principle of biological<br />
systems‟ during <strong>the</strong> decade of research since Brooks‟ pronouncement? Unfortunately,<br />
we do not need to look far to find reasons for continued skepticism.<br />
The existential philosopher <strong>Dr</strong>eyfus, while mostly known in <strong>the</strong> field of AI for his<br />
scathing criticisms of GOFAI (e.g. <strong>Dr</strong>eyfus 1972), has recently referred to <strong>the</strong> current<br />
work in embodied-embedded AI as a „failure‟. He points to <strong>the</strong> lack of “a model of our<br />
particular way of being embedded <strong>and</strong> embodied such that what we experience is<br />
significant for us in <strong>the</strong> particular way that it is. That is, we would have to include in our<br />
program a model of a body very much like ours” (<strong>Dr</strong>eyfus 2007, p. 265). Similarly, Di<br />
Paolo (2003) has argued that embodied-embedded robots, while in many respects an<br />
improvement over traditional GOFAI, can never be truly autonomous. Moreover, <strong>the</strong><br />
mere presence of a physical body <strong>and</strong> a closed sensorimotor loop in such robots does<br />
not fully solve <strong>the</strong> problem of grounding meaning (cf. Ziemke 1999; 2001). These<br />
problems are even fur<strong>the</strong>r amplified because, while embodied-embedded AI has focused<br />
on establishing itself as a viable alternative to <strong>the</strong> traditional computational paradigm,<br />
relatively little effort has been made to connect its practical <strong>and</strong> experimental work with<br />
21 | P a g e
<strong>the</strong>ories outside <strong>the</strong> field of AI, such as with <strong>the</strong>oretical biology, in order to address<br />
issues of autonomy <strong>and</strong> embodiment (Ziemke 2007).<br />
It appears that <strong>the</strong>re is a growing awareness in <strong>the</strong> field of embodied-embedded AI that<br />
something crucial is still missing in <strong>the</strong> current implementations of cognitive systems,<br />
<strong>and</strong> that this shortcoming is likely related to <strong>the</strong>ir particular manner of embodiment (cf.<br />
Ziemke 2003). But what could this elusive factor be? What is so special about <strong>the</strong> body<br />
of living systems? In order to answer <strong>the</strong>se questions we need to shift our focus back to<br />
recent developments in <strong>the</strong> cognitive sciences.<br />
2.2 Fur<strong>the</strong>r: Toward enactive cognitive science<br />
The enactive paradigm originally emerged as a part of embodied-embedded cognitive<br />
science in <strong>the</strong> early 1990s with <strong>the</strong> publication of <strong>the</strong> influential book The Embodied<br />
Mind (Varela, et al. 1991). It has recently distinguished itself by more explicitly placing<br />
<strong>the</strong> phenomenon of <strong>life</strong> at <strong>the</strong> heart of cognitive science (e.g. Thompson 2007). In order<br />
to determine what is missing in current embodied-embedded AI, we will <strong>the</strong>refore<br />
consider how such work could contribute to <strong>the</strong> enactive account. In particular, we are<br />
interested in how it could inform <strong>the</strong>ories of how bodily activity relates to <strong>the</strong> <strong>mind</strong> at<br />
three interrelated „dimensions of embodiment‟: (i) bodily self-regulation, (ii) sensorymotor<br />
coupling, <strong>and</strong> (iii) intersubjective interaction (cf. Thompson & Varela 2001).<br />
While <strong>the</strong> development of such fully „enactive‟ AI is a significant challenge to existing<br />
AI methodologies, it has <strong>the</strong> potential of providing a fresh perspective on some of <strong>the</strong><br />
issues currently faced by <strong>the</strong> embodied-embedded approach.<br />
(i) Bodily self-regulation. This dimension of embodiment is central to <strong>the</strong> enactive<br />
paradigm in cognitive science, because its <strong>the</strong>oretical framework builds on <strong>the</strong> notion of<br />
biological autonomy (Di Paolo, et al., in press). Since embodied-embedded AI has<br />
always been involved in extensive studies of „autonomous systems‟ (e.g. Pfeifer &<br />
Scheier 1999), it might seem that such AI research is particularly destined to relate to<br />
<strong>the</strong> enactive paradigm in a mutually informative manner. Unfortunately, things are not<br />
as straightforward; <strong>the</strong> enactive account of biological autonomy has a very different<br />
view of what constitutes autonomy when compared to most embodied-embedded AI,<br />
22 | P a g e
which is why it is sometimes referred to more specifically as constitutive autonomy (cf.<br />
<strong>Froese</strong>, et al. 2007). Its distinctive approach can be traced to <strong>the</strong> notion of autopoiesis, a<br />
systems concept which originated in <strong>the</strong> <strong>the</strong>oretical biology of <strong>the</strong> 1970s (e.g. Maturana<br />
& Varela 1980). We will return to this concept in Chapter 3.<br />
In brief, we can say that <strong>the</strong> enactive paradigm broadly defines an autonomous agent as<br />
a self-producing network of processes which constitutes its own identity; <strong>the</strong><br />
paradigmatic example being a minimal living organism (cf. Di Paolo 2009). The<br />
existence of this self-constituted system is necessarily precarious, because it continually<br />
needs to sustain its own identity against <strong>the</strong> equalizing forces of its environment.<br />
<strong>Dr</strong>awing from <strong>the</strong> bio-philosophy of Hans Jonas (1968), it is claimed that such an<br />
autonomous system, one whose being is its own doing, should be conceived of as an<br />
individual in its own right. Moreover, as a consequence this process of self-constitution<br />
brings forth, in <strong>the</strong> same stroke of identity generation, what is outside of this identity,<br />
namely its world (cf. Thompson 2007, p. 153). In o<strong>the</strong>r words, it is proposed that <strong>the</strong><br />
continuous process of self-construction, which constitutes <strong>the</strong> autonomous system as a<br />
precarious individual, also furnishes it with a meaningful perspective on its physical<br />
environment. In sum, biological autonomy lies at <strong>the</strong> basis of sense-making (Weber &<br />
Varela 2002).<br />
It follows from <strong>the</strong>se considerations that today‟s robotic AI systems are not autonomous<br />
in <strong>the</strong> enactive sense. They do not constitute <strong>the</strong>ir own identity, <strong>and</strong> <strong>the</strong> only „identity‟<br />
which <strong>the</strong>y can be said to possess is projected onto <strong>the</strong>m by <strong>the</strong> observing researcher (cf.<br />
Bar<strong>and</strong>iaran, et al. 2009; <strong>Froese</strong>, et al. 2007). The popular methodology of evolutionary<br />
robotics, for example, presupposes that an „individual‟ is already defined by <strong>the</strong><br />
experimenter as <strong>the</strong> basis for selection by <strong>the</strong> evolutionary algorithm. And in <strong>the</strong><br />
dynamical approach to AI it is up to <strong>the</strong> investigator to distinguish which subpart of <strong>the</strong><br />
systemic whole actually constitutes <strong>the</strong> „agent‟ (Beer 1995a). The enactive notion of<br />
autonomous agency <strong>the</strong>refore poses a significant difficulty even for current embodiedembedded<br />
AI methodologies (<strong>Froese</strong> & Ziemke 2009).<br />
Never<strong>the</strong>less, it is worth noting that AI researchers do not have to syn<strong>the</strong>size actual<br />
living beings in order for <strong>the</strong>ir work to provide some relevant insights into <strong>the</strong><br />
23 | P a g e
dimension of bodily self-regulation. This misunderst<strong>and</strong>s <strong>the</strong> purpose of a good model<br />
(cf. Chapter 7, p. 114). Following <strong>the</strong> organismic approach first proposed by Di Paolo<br />
(2003; Di Paolo & Iizuka 2008), an initial step would be to investigate artificial systems<br />
with some kind of self-sustaining dynamic structures. In this manner embodiedembedded<br />
AI can move beyond its current focus on closed sensory-motor feedback<br />
loops by implementing systems which have a reciprocal link between internal<br />
organization <strong>and</strong> external behavior (cf. Iizuka & Di Paolo 2008). Indeed, <strong>the</strong>re are signs<br />
that a shift toward more concern with bodily self-regulation is starting to develop. This<br />
is demonstrated by an increasing interest in homeostasis as a regulatory mechanism for<br />
investigating, for example, sensory inversion (e.g. Di Paolo 2003), <strong>the</strong> emergence of<br />
sensory-motor coupling <strong>and</strong> development (e.g. Ikegami & Suzuki 2008; Wood & Di<br />
Paolo 2007), mechanisms of behavioral preference (e.g. Iizuka & Di Paolo 2007a), <strong>and</strong><br />
active perception (e.g. Harvey 2004). Of course, looking at <strong>the</strong> emergence of behavior<br />
from <strong>the</strong> perspective of modeling chemical self-assembly should be considered as well<br />
(e.g. Egbert & Di Paolo 2009), especially since <strong>the</strong> notion of autonomy is currently best<br />
understood in <strong>the</strong> chemical domain (<strong>Froese</strong>, et al. 2007).<br />
(ii) Sensory-motor coupling. Since sensory-motor embodiment or situatedness is <strong>the</strong><br />
research target of most current embodied-embedded AI, its results can have an impact<br />
on <strong>the</strong> sensory-motor <strong>the</strong>ories of <strong>the</strong> enactive paradigm. However, since <strong>the</strong> vast<br />
majority of such work is not concerned with how <strong>the</strong> constraints of constitutive<br />
autonomy are related to <strong>the</strong> emergence of sensory-motor behavior, it is not contributing<br />
to <strong>the</strong> enactive account of how an autonomous agent is able to bring forth its own<br />
relational domain (<strong>Froese</strong> & Ziemke 2009). To become more relevant in this respect,<br />
<strong>the</strong> field of system modeling needs to adapt its methodologies so as to deal with <strong>the</strong><br />
enactive proposal that an agent‟s sense-making is grounded in <strong>the</strong> active regulation of<br />
ongoing sensory-motor coupling in relation to <strong>the</strong> viability of a precarious, dynamically<br />
self-sustaining identity. So far this is an area which has been practically unexplored,<br />
although some promising work has begun from <strong>the</strong> perspective of evolutionary robotics<br />
(e.g. Di Paolo 2003; Iizuka & Di Paolo 2008). Ano<strong>the</strong>r route that shows potential,<br />
though radically different from <strong>the</strong> usual evolutionary robotics methodology, is to<br />
follow an incremental approach in simplified artificial chemistries, which has already<br />
24 | P a g e
een used to model <strong>the</strong> emergence of autonomous systems that move <strong>and</strong> can follow<br />
gradients (e.g. Ikegami & Suzuki 2008; Egbert & Di Paolo 2009).<br />
(iii) Intersubjective interaction. The considerations regarding sensory-motor<br />
embodiment can be extended to <strong>the</strong> domain of intersubjective interaction, since this<br />
dimension of embodiment also involves distinctive forms of sensory-motor coupling<br />
(Thompson & Varela 2001). An enactive account of social underst<strong>and</strong>ing based on this<br />
<strong>continuity</strong>, fur<strong>the</strong>r discussed in Chapter 3, has recently been outlined by Di Paolo,<br />
Rohde <strong>and</strong> De Jaegher (in press). They make <strong>the</strong> important suggestion that <strong>the</strong><br />
traditional focus on <strong>the</strong> embodiment of individual interactors needs to be complemented<br />
by an investigation of <strong>the</strong> interaction process that takes place between <strong>the</strong>m. This shift<br />
in focus enables <strong>the</strong>m to extend <strong>the</strong> enactive notion of sense-making into <strong>the</strong> realm of<br />
social cognition in <strong>the</strong> form of participatory sense-making (De Jaegher 2006), a shift we<br />
will specify in more detail in Chapter 4 <strong>and</strong> support in subsequent chapters.<br />
The development of such an account is important for embodied-embedded AI, because<br />
most of its current research remains limited to „lower-level‟ cognition. Exploring <strong>the</strong><br />
domain of social interaction might provide it with <strong>the</strong> necessary means to tackle <strong>the</strong><br />
problem of „scalability‟ (cf. Clark 1997, p. 101) by bridging <strong>the</strong> cognitive gap, in<br />
particular because such interaction can constitute new ways of sense-making that are not<br />
available to <strong>the</strong> individual alone (<strong>Froese</strong> & Di Paolo, in press; De Jaegher & <strong>Froese</strong><br />
2009). The challenge is to implement AI systems that constitute <strong>the</strong> social domain by<br />
means of an interaction process that is essentially embodied <strong>and</strong> situated, as opposed to<br />
<strong>the</strong> traditional means of formalized transmissions of abstract information over prespecified<br />
communication channels. Di Paolo, Rohde <strong>and</strong> De Jaegher review some initial<br />
work in this direction which demonstrates that <strong>the</strong>se models have <strong>the</strong> possibility to<br />
capture <strong>the</strong> rich dynamics of reciprocity that are left outside of traditional individualistic<br />
approaches. A more detailed review of this methodology is presented in Chapter 7.<br />
There is thus a possibility for modeling work to inform each of <strong>the</strong>se central dimensions<br />
of embodiment. However, it is debatable if AI research should be considered as enactive<br />
ra<strong>the</strong>r than merely embodied-embedded if it does not address some form of bodily self-<br />
25 | P a g e
egulation, or leads to <strong>the</strong> constitution of autonomy in novel domains of interaction 5 . In<br />
this sense <strong>the</strong> authors of The Embodied Mind perhaps got slightly carried away when<br />
<strong>the</strong>y referred to <strong>the</strong> emergence of Brooks‟s behaviour-based robotics as a “fully enactive<br />
approach to AI” (Varela, et al. 1991, p. 212). However, this is not to say that embodiedembedded<br />
AI does not have an impact on <strong>the</strong> shift toward <strong>the</strong> enactive framework, it<br />
certainly does, but only to <strong>the</strong> extent that <strong>the</strong>re is an overlap between <strong>the</strong> paradigms. Its<br />
current influence is <strong>the</strong>refore by no means as significant as it has been on <strong>the</strong> shift<br />
toward embodied-embedded cognitive science. For example, Thompson‟s recent book<br />
Mind in Life, which can be considered as a successor to The Embodied Mind, does not<br />
even include robotic AI as one of <strong>the</strong> cognitive science sub-disciplines from which it<br />
draws its insights (cf. Thompson 2007, p. 24). Of course, it goes without saying that all<br />
of <strong>the</strong>se dimensions of embodiment are open to fur<strong>the</strong>r refinement through artificial<br />
modeling, <strong>and</strong> that some initial work in this direction has already begun. Never<strong>the</strong>less,<br />
for AI to have a more significant impact on <strong>the</strong> ongoing shift toward enactive cognitive<br />
science, it must address some considerable methodological challenges (<strong>Froese</strong> &<br />
Ziemke 2009). The field needs to extend its current preoccupation with sensory-motor<br />
interaction in <strong>the</strong> behavioral domain to include a concern of <strong>the</strong> constitutive processes<br />
that give rise to that domain in living systems. Maybe Brooks (1997) was right when he<br />
suggested that in order for AI to be more <strong>life</strong>-like perhaps <strong>the</strong>re has to be some<br />
fundamental change in our thinking. Fortunately, such a change might be provided by<br />
<strong>the</strong> development of enactive AI (<strong>Froese</strong> 2007).<br />
Indeed, at <strong>the</strong> moment it seems more likely that <strong>the</strong> influence will run more strongly<br />
from enactive cognitive science to AI instead. Its account of autonomous agency, for<br />
example, has <strong>the</strong> potential to provide embodied-embedded AI with exactly <strong>the</strong> kind of<br />
bodily organizational principle that has been identified as missing by Brooks (2001). In<br />
addition, <strong>the</strong> enactive notion of sense-making, as a biologically grounded account of<br />
how a system must be embodied in order for its encounters to be experienced as<br />
significant, can be used as a response to <strong>Dr</strong>eyfus‟s vague requirement of a detailed<br />
5 In a similar manner it could be argued that since recent work in enactive perception (e.g. Noë 2004) is<br />
more concerned with sensory-motor contingencies than with autonomous agency or lived subjectivity,<br />
such work might be more usefully classified as part of embodied-embedded cognitive science. For a more<br />
in-depth discussion of this issue, cf. <strong>Froese</strong> <strong>and</strong> Ziemke (2009), Thompson (2005) <strong>and</strong> Torrance (2005).<br />
26 | P a g e
description of our body, which in terms of AI apparently has not even “a chance of<br />
being realized in <strong>the</strong> real world” (<strong>Dr</strong>eyfus 2007, p. 265). Fur<strong>the</strong>rmore, <strong>the</strong>re is a good<br />
possibility that <strong>the</strong> field‟s current restriction to „lower-level‟ cognition could be<br />
overcome in a principled manner by extending its existing research focus on sensorymotor<br />
embodiment to also include participatory sense-making. All of <strong>the</strong>se concepts<br />
will be introduced in more detail in Chapters 3 <strong>and</strong> 4, <strong>and</strong> some new AI models of social<br />
interaction will be presented in Chapters 7 to 10.<br />
Never<strong>the</strong>less, we can already now ask to what extent such modeling work can impact on<br />
<strong>the</strong> current developments in cognitive science? The following section argues that, while<br />
clearly an important aspect, results in AI are not sufficient to displace <strong>the</strong> orthodox<br />
mainstream on its own. More than just having to make Heideggerian AI more<br />
Heideggerian, as <strong>Dr</strong>eyfus (2007) proposes, Heideggerian cognitive science as a whole<br />
must become more Heideggerian by complementing its methodological focus on AI<br />
with considerations of phenomenology, a shift which coincides with a movement from<br />
embodied-embedded to enactive cognitive science.<br />
2.3 An empirical stalemate<br />
Over two decades ago <strong>Dr</strong>eyfus <strong>and</strong> <strong>Dr</strong>eyfus (1988) characterized GOFAI as a project in<br />
which <strong>the</strong> rationalist tradition had finally been put to an empirical test, <strong>and</strong> it had failed.<br />
Never<strong>the</strong>less, despite this supposed „failure‟ no alternative has yet succeeded in fully<br />
displacing <strong>the</strong> orthodox mainstream in AI or cognitive science. While it could be argued<br />
that more progress in embodied-embedded or enactive AI will eventually remedy this<br />
situation, a more serious problem becomes apparent when we consider why this<br />
perceived „failure‟ did not remove <strong>the</strong> orthodox framework from <strong>the</strong> mainstream. As<br />
Wheeler (2005, p. 185) points out, this did not happen for <strong>the</strong> simple reason that<br />
researchers are always at liberty to interpret practical problems as mere temporary<br />
difficulties which will eventually be eliminated through more scientific research <strong>and</strong><br />
additional technological development. Accordingly, Wheeler goes on to conclude that a<br />
resolution of <strong>the</strong> st<strong>and</strong>off must await fur<strong>the</strong>r empirical evidence.<br />
27 | P a g e
However, while Wheeler‟s appeal to more experimental data is applicable when <strong>the</strong>re is<br />
a need to resolve <strong>the</strong>oretical issues within a particular paradigm, it is not clear whe<strong>the</strong>r it<br />
is also valid when deciding between different paradigms: you always already have to<br />
choose (whe<strong>the</strong>r explicitly or not) one paradigm over <strong>the</strong> o<strong>the</strong>rs from which to interpret<br />
<strong>the</strong> data. Fur<strong>the</strong>rmore, <strong>the</strong> impact of this choice is significant:<br />
The conceptual framework that we bring to <strong>the</strong> study of cognition can have<br />
profound empirical consequences on <strong>the</strong> practice of cognitive science. It<br />
influences <strong>the</strong> phenomena we choose to study, <strong>the</strong> questions we ask about <strong>the</strong>se<br />
phenomena, <strong>the</strong> experiments we perform, <strong>and</strong> <strong>the</strong> ways in which we interpret<br />
<strong>the</strong> results of <strong>the</strong>se experiments. (Beer 2000, p. 91)<br />
Since data is only meaningful in a manner which crucially depends on <strong>the</strong> underlying<br />
premises of <strong>the</strong> investigator, <strong>the</strong> current empirical stalemate in AI appears to be partly<br />
due to a lack of empirical evidence, but also largely due to <strong>the</strong> fact that <strong>the</strong> impact of<br />
experimental results fundamentally depends on an interpretive aspect.<br />
Again, this is not to say that experimental evidence has no effect on moving forward a<br />
paradigm shift; of course, it is certainly helpful. Indeed, an important step will be to reinterpret<br />
<strong>the</strong> existing empirical evidence that has already been accumulated (a strategy<br />
we will pursue in Chapters 6 <strong>and</strong> 13) However, <strong>the</strong> point is simply that such evidence is<br />
a necessary but not sufficient condition for a successful paradigm shift. In o<strong>the</strong>r words,<br />
in order for experimental data to be turned into scientific knowledge it first has to be<br />
interpreted according to (often implicitly) chosen constitutive assumptions. Moreover,<br />
our premises even ground <strong>the</strong> manner in which we distinguish between noise <strong>and</strong> data 6 .<br />
It follows from this that <strong>the</strong> major cause of <strong>the</strong> st<strong>and</strong>off in <strong>the</strong> philosophical domain also<br />
plays a significant role in <strong>the</strong> current empirical stalemate: both domains of enquiry<br />
require an interpretative action on <strong>the</strong> part of <strong>the</strong> observer. And, more importantly,<br />
6 Consider, for example, <strong>the</strong> fact that <strong>the</strong> fossil record shows long periods of stasis interspersed with<br />
layers of rapid phyletic change. Someone who believes that evolution proceeds gradually will treat this<br />
fact as irrelevant noise (e.g. due to accidental differences in preservation), while someone who claims that<br />
it proceeds as punctuated equilibria will view it as supporting evidence (Eldredge & Gould 1972).<br />
28 | P a g e
while it is possible to influence this act of interpretation through research progress, its<br />
outcome cannot be fully determined by such external events since any kind of<br />
underst<strong>and</strong>ing always already presupposes interpretative activity. In addition, <strong>the</strong> impact<br />
of this potential influence is also limited because <strong>the</strong> significance of such advances<br />
might not become apparent if one does not already hold <strong>the</strong> kind of constitutive<br />
assumptions required to underst<strong>and</strong> <strong>the</strong>m appropriately.<br />
From <strong>the</strong> perspective of enactive cognitive science this constitutive role of interpretation<br />
for scientific activity is hardly surprising (Varela, et al. 1991, p. 10-12) 7 . In fact, at one<br />
point <strong>the</strong> enactive approach was actually called “<strong>the</strong> hermeneutic approach” (Thompson<br />
2007, p. 24), <strong>and</strong> it can even ground <strong>the</strong>se epistemological reflections in <strong>the</strong> biology of<br />
autonomy by claiming that a living system always constitutes its own perspective of<br />
value on <strong>the</strong> world (we will return to this idea in Chapter 3). Never<strong>the</strong>less, <strong>the</strong>se<br />
considerations give a ra<strong>the</strong>r bleak outlook for <strong>the</strong> possibility of actively generating a<br />
successful paradigm shift in <strong>the</strong> cognitive sciences. At this point it might seem relatively<br />
futile to worry about such abstract problems <strong>and</strong> better to just get on with <strong>the</strong> work.<br />
Considering <strong>the</strong> overall state of affairs this is in many respects a sensible <strong>and</strong> pragmatic<br />
course of action, <strong>and</strong> one that is evidently also pursued in this <strong>the</strong>sis. Never<strong>the</strong>less, in<br />
order to better set <strong>the</strong> stage for <strong>the</strong> final chapters in this <strong>the</strong>sis, especially for Chapters<br />
11 <strong>and</strong> 12 on phenomenology, it will be useful to paint <strong>the</strong> bigger picture at least in a<br />
broad outline. For it is still <strong>the</strong> case that we at least implicitly choose a paradigm for our<br />
research. However, if rational argument combined with empirical data is still not<br />
sufficient to establish this choice, <strong>the</strong>n what is it that determines which premises are<br />
assumed? And how can this elusive factor be influenced? The rest of this chapter<br />
7 The philosophy of science that is associated with <strong>the</strong> enactive paradigm is typically a combination of <strong>the</strong><br />
operational epistemology of Maturana (1988) <strong>and</strong> <strong>the</strong> phenomenological ontology of Heidegger (1927)<br />
<strong>and</strong> <strong>the</strong> later Husserl (1936). See, for example, <strong>the</strong> excellent paper by Bitbol (2002). This view of <strong>the</strong><br />
scientific method fits nicely with <strong>the</strong> content of <strong>the</strong> enactive approach, but whe<strong>the</strong>r it is <strong>the</strong> most<br />
compatible one is still open to debate. To be sure, it clearly differs from <strong>the</strong> view of science adopted by<br />
<strong>the</strong> sensory-motor „enactive‟ approaches that prefer to retain a realist stance (Pascal & O‟Regan 2008;<br />
Wheeler 2005). The difference, in essence, is that realism necessitates a role for mental representations.<br />
29 | P a g e
provides a tentative answer to <strong>the</strong>se questions by focusing on a crucial aspect of<br />
enactive cognitive science that has not been addressed so far.<br />
2.4 A phenomenological resolution<br />
The enactive account of autonomous agency as expressed in terms of systems biology is<br />
complemented by a concern with <strong>the</strong> first-person point of view, by which is meant <strong>the</strong><br />
subjectively lived experience associated with cognitive <strong>and</strong> mental events (cf. Varela &<br />
Shear 1999). This culmination of <strong>the</strong> recent developments in <strong>the</strong> cognitive sciences is<br />
illustrated in Figure 2-1.<br />
Enactive<br />
Embodied / Embedded<br />
Connectionist<br />
Cognitivist<br />
Today 1990s 1980s 1970s<br />
computational<br />
dynamic - emergent<br />
embodied - embedded<br />
living - lived<br />
Figure 2-1. This schematic summarizes <strong>the</strong> paradigm shift which is ongoing in cognitive science. There<br />
has been a systematic trend toward more inclusive frameworks which incorporate <strong>and</strong> ground <strong>the</strong><br />
previous insights in a more extended context. With enactive cognitive science we have finally returned to<br />
<strong>the</strong> point from which all of our investigations must necessarily originate in <strong>the</strong> first place, namely <strong>the</strong><br />
subjectivity of human existence: our lived experience as living beings.<br />
Since <strong>the</strong> enactive framework incorporates both biological agency (<strong>the</strong> living body) <strong>and</strong><br />
phenomenological subjectivity (<strong>the</strong> lived body), it has <strong>the</strong> capacity to recast <strong>the</strong><br />
traditional <strong>mind</strong>-body problem in terms of what has recently been called <strong>the</strong> „body-body<br />
30 | P a g e
problem‟ (Hanna & Thompson 2003). On this view <strong>the</strong> traditional „explanatory gap‟<br />
(Levine 1983) between our best explanation <strong>and</strong> „what it is like to be‟ that which is to<br />
be explained (Nagel 1974) is no longer absolute since <strong>the</strong> concepts of subjectively lived<br />
body <strong>and</strong> objective living body both require <strong>the</strong> notion of <strong>life</strong>. Though more work needs<br />
to be done to fully articulate <strong>the</strong> details, such as <strong>the</strong> development of an account that<br />
integrates <strong>the</strong>se dual aspects into a coherent conception of <strong>the</strong> embodied subject, this<br />
reformulation of <strong>the</strong> „hard problem‟ of consciousness (Chalmers 1996) can be seen as<br />
one of <strong>the</strong> major contributions of <strong>the</strong> enactive paradigm (cf. Torrance 2005).<br />
Never<strong>the</strong>less, it is not yet clear how a concern with subjective experience could provide<br />
us with a way to move beyond <strong>the</strong> stalemate that we have identified in <strong>the</strong> previous<br />
sections. Surely <strong>the</strong> enactive approach is just more philosophical <strong>the</strong>ory? However, to<br />
say this is to miss <strong>the</strong> point that it derives many of its crucial insights from a source that<br />
is quite distinct from st<strong>and</strong>ard <strong>the</strong>oretical or empirical enquiry, namely from careful<br />
phenomenological observations that have been gained through <strong>the</strong> principled<br />
investigation of <strong>the</strong> structure of our lived experience (see Ch. 2 in Thompson 2007 for<br />
an overview; for an introduction, cf. Gallagher & Zahavi 2008). But what about <strong>the</strong><br />
insights from which Heidegger originally deduced his claims? If his analysis of <strong>the</strong><br />
holistic structure of our Dasein or „being-in-<strong>the</strong>-world‟ (Heidegger 1927) is one of <strong>the</strong><br />
most influential accounts of <strong>the</strong> continental phenomenological tradition, <strong>the</strong>n why did it<br />
not succeed in convincing mainstream cognitive scientists? The regrettable answer is<br />
that while his claims have sometimes been probed in <strong>the</strong> philosophical or empirical<br />
domain, <strong>the</strong>re have not been many sustained <strong>and</strong> principled efforts in orthodox<br />
cognitive science to verify <strong>the</strong>ir validity in <strong>the</strong> phenomenological domain.<br />
If <strong>the</strong> enactive paradigm is to avoid a similar fate <strong>the</strong>n it needs to focus less on <strong>the</strong><br />
development of better, more enactive AI (an aim which will, to a large extent, already<br />
be pursued by embodied-embedded cognitive science), <strong>and</strong> more on <strong>the</strong> promotion of<br />
principled first-person phenomenological studies. Indeed, according to Di Paolo, Rohde<br />
<strong>and</strong> De Jaegher (in press) <strong>the</strong> central importance of experience is perhaps one of <strong>the</strong><br />
most revolutionary implications of <strong>the</strong> enactive approach, especially since a<br />
phenomenologically informed science goes beyond black marks on paper <strong>and</strong><br />
experimental procedures for measuring data, <strong>and</strong> dives straight into <strong>the</strong> realm of<br />
31 | P a g e
personal experience. They point out, for example, that no amount of rational argument<br />
will convince a reader of Jonas‟s claim that, as an embodied organism, he is concerned<br />
with his own existence if <strong>the</strong> reader cannot see this for himself. Thus, development of<br />
enactive cognitive science implicates an element of personal en-action.<br />
Accordingly, Varela <strong>and</strong> Shear (1999) outline <strong>the</strong> beginnings of a project where nei<strong>the</strong>r<br />
experience nor external mechanism have <strong>the</strong> final word, but ra<strong>the</strong>r st<strong>and</strong> to each o<strong>the</strong>r in<br />
a relationship of generative mutual constraints. They point out that <strong>the</strong> process of<br />
collecting phenomenological data requires disciplined training in <strong>the</strong> skilful exploration<br />
of one‟s lived experience. Such an endeavor to raise awareness might already be<br />
worthwhile in itself, but in <strong>the</strong> context of <strong>the</strong> stalemate in <strong>the</strong> cognitive sciences it<br />
comes with an added benefit. To be sure, it is still <strong>the</strong> case that phenomenological data<br />
first has to be interpreted from a particular point of view before it can be integrated into<br />
a conceptual framework. But, in a nutshell, generating such data also requires a change<br />
in our mode of experiencing. Moreover, this change in our experiential attitude is<br />
constituted by a change in our mode of being, <strong>and</strong> this in turn entails a change in our<br />
underst<strong>and</strong>ing (cf. Varela 1976). It is not primarily a matter of <strong>the</strong>oretical knowledge, or<br />
of deriving facts. Ra<strong>the</strong>r, it is this being, <strong>the</strong> structure of our everyday existence, which<br />
determines how we interpret our world. Of course, since we are autonomous agents this<br />
does not mean that actively practicing phenomenological inquiry necessarily commits<br />
us to an enactive approach. But perhaps by changing our awareness in this manner we<br />
will be able to underst<strong>and</strong> more fully <strong>the</strong> reasons, o<strong>the</strong>r than in terms of <strong>the</strong>ory <strong>and</strong><br />
empirical data, which are at <strong>the</strong> root of why we prefer one paradigm over ano<strong>the</strong>r.<br />
2.5 Summary<br />
The field of AI has had a significant impact on <strong>the</strong> ongoing shift from orthodox toward<br />
embodied-embedded cognitive science, especially because work in AI has made it<br />
possible for philosophical disputes to be addressed in an experimental manner.<br />
Conversely, enactive cognitive science can have a strong influence on AI because of its<br />
biologically <strong>and</strong> phenomenologically grounded account of autonomous agency, sensemaking,<br />
<strong>and</strong> social interaction. The development of such enactive AI, while challenging<br />
32 | P a g e
to current methodologies, has <strong>the</strong> potential to address some of <strong>the</strong> problems currently<br />
impeding significant progress in embodied-embedded AI.<br />
However, if this alternative paradigm is to be successful in actually displacing <strong>the</strong><br />
orthodox mainstream, <strong>the</strong>n it is likely that <strong>the</strong>oretical arguments <strong>and</strong> empirical evidence<br />
alone are necessary but not sufficient. For this shift to happen it will additionally be<br />
necessary that a phenomenological pragmatics is established as part of <strong>the</strong> accepted<br />
methodological toolbox of contemporary cognitive science (cf. Depraz, et al. 2003).<br />
This shift of focus from AI to phenomenology coincides with a shift from embodiedembedded<br />
to enactive cognitive science. Unfortunately, however, most of our current<br />
cognitive science institutions are not concerned with supporting first-person<br />
phenomenological inquiry in any principled manner. One promising opportunity for<br />
change is <strong>the</strong> increasing interest in sensory augmentation technology (cf. <strong>Froese</strong> &<br />
Spiers 2007). It will be one of <strong>the</strong> major challenges faced by those wanting to make <strong>the</strong><br />
enactive approach accepted as part of mainstream science to devise appropriate ways of<br />
overcoming this impasse. In this context, early day AI practitioner Terry Winograd‟s<br />
decision to turn toward teaching Heidegger in computer science courses at Stanford,<br />
after he became disillusioned with his pioneering work in symbolic language parsing<br />
(Winograd 1972), appears in a new light (<strong>Dr</strong>eyfus 1991, p. 119) 8 . In return, this shift in<br />
underst<strong>and</strong>ing also had a profound impact on his work in AI, which prefigured some of<br />
<strong>the</strong> concerns in embodied-embedded cognitive science (cf. Winograd & Flores 1986).<br />
The rest of this <strong>the</strong>sis will unfold according to <strong>the</strong> pattern established in this chapter. In<br />
<strong>the</strong> first part <strong>the</strong> <strong>the</strong>oretical framework of <strong>the</strong> enactive paradigm is presented in more<br />
detail (Chapter 3), with a special focus on its approach to sociality (Chapter 4). Some<br />
empirical evidence on social interaction will <strong>the</strong>n be re-interpreted from this <strong>the</strong>oretical<br />
perspective (Chapters 5 <strong>and</strong> 6). In <strong>the</strong> second part of <strong>the</strong> <strong>the</strong>sis this perspective is fur<strong>the</strong>r<br />
supported by means of an integrative evolutionary robotics methodology, which is used<br />
to syn<strong>the</strong>size a series of agent-based models that investigate <strong>the</strong> dynamics of social<br />
8 Note that it was Winograd‟s practical frustration with AI design that motivated his phenomenological<br />
<strong>and</strong> embodied-embedded turn. We will consider this kind of pedagogical value of engaging in AI-based<br />
research more fully in Chapter 7.<br />
33 | P a g e
interaction (Chapters 7 to 10). The results of <strong>the</strong>se models support <strong>the</strong> enactive approach<br />
to social cognition. Still, <strong>the</strong>y say almost nothing about what it is like for someone to be<br />
involved social situations. Accordingly, in <strong>the</strong> final part of <strong>the</strong> <strong>the</strong>sis this modeling<br />
approach is complemented by a phenomenological analysis of how our experience is<br />
modulated by <strong>the</strong> presence of o<strong>the</strong>rs (Chapters 11 <strong>and</strong> 12). The <strong>the</strong>sis finishes with a<br />
brief look at potential future work in related fields (Chapter 13), <strong>and</strong> a summary of what<br />
has been done (Chapter 14).<br />
34 | P a g e
3 Enactive cognitive science<br />
The aim of this chapter is to unpack <strong>the</strong> biological foundations of enactive cognitive<br />
science in more detail. First, <strong>the</strong> focus on basic biological principles is motivated by a<br />
closer consideration of <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis, which forms <strong>the</strong> <strong>the</strong>oretical<br />
backbone of enactive cognitive science. On this basis <strong>the</strong> fundamental notions of<br />
autopoiesis, organizational closure, <strong>and</strong> constitutive autonomy are introduced, followed<br />
by a consideration of <strong>the</strong> notion of sense-making <strong>and</strong> its necessary dependence on<br />
adaptivity <strong>and</strong> constitutive autonomy. Finally, all of <strong>the</strong>se notions are combined in order<br />
to indicate <strong>the</strong> broader framework of enactive cognitive science.<br />
3.1 The <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis<br />
A radical element of <strong>the</strong> recent embodied turn in cognitive science has become known<br />
as <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis (LMCT). The LMCT has been proposed in a wide<br />
variety of formulations (e.g. Di Paolo 2003; Godfrey-Smith 1996; Wheeler 1997;<br />
Stewart, in press; 1996; Maturana & Varela 1987). Most of <strong>the</strong>se essentially revolve<br />
around what has been called „strong‟ or „deep‟ <strong>continuity</strong>, i.e. that <strong>the</strong> phenomena of <strong>life</strong><br />
<strong>and</strong> <strong>mind</strong> have a common set of basic organizational properties:<br />
In more concrete terms, <strong>the</strong> <strong>the</strong>sis of strong <strong>continuity</strong> would be true if, for<br />
example, <strong>the</strong> basic concepts needed to underst<strong>and</strong> <strong>the</strong> organization of <strong>life</strong> turned<br />
out to be self-organization, collective dynamics, circular causal processes,<br />
autopoiesis, etc., <strong>and</strong> if those very same concepts <strong>and</strong> constructs turned out to be<br />
central to a proper scientific underst<strong>and</strong>ing of <strong>mind</strong>. (Clark 2001, p. 118)<br />
This version of <strong>the</strong> LMCT is especially attractive for embodied, dynamical approaches<br />
to cognitive science for obvious reasons: for if <strong>the</strong> <strong>the</strong>sis turns out to be correct, <strong>the</strong>n <strong>the</strong><br />
applicability of <strong>the</strong>se approaches is not only limited to mere low-level, „implementation‟<br />
details of adaptive behavior. Instead, <strong>the</strong>y would actually be providing <strong>the</strong> very<br />
foundations of a general <strong>the</strong>ory of <strong>mind</strong> <strong>and</strong> cognition, one that would also include <strong>the</strong><br />
highest reaches of human cognition (cf. Clark 2001, pp. 128-130).<br />
35 | P a g e
The most comprehensive framework based on <strong>the</strong> LMCT is currently being developed<br />
by enactive cognitive science (e.g. Di Paolo, et al., in press; Thompson 2007; 2004;<br />
Bar<strong>and</strong>iaran & Moreno 2006; 2008; <strong>Froese</strong> 2009). The enactive approach is just as<br />
interested in <strong>the</strong> single-cell organism, as <strong>the</strong> paradigmatic case of individual agency, as<br />
it is in human existence, as <strong>the</strong> paradigmatic case of enculturation. It is important to<br />
clarify from <strong>the</strong> start that this version of <strong>the</strong> LMCT does not involve a reductive form of<br />
<strong>continuity</strong>, whereby „higher-level‟ phenomena would be reduced to „lower-level‟ ones.<br />
The notion of autonomy, which is applicable to novel phenomena in each of <strong>the</strong> major<br />
transitions of <strong>life</strong>, guards against such trivialization. In o<strong>the</strong>r words, <strong>the</strong> enactive<br />
paradigm proposes a view of <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> where that <strong>continuity</strong> is more like an<br />
open-ended set of autonomous domains of dynamics that are partially decoupled <strong>and</strong><br />
constitutively interrelated by multiple interdependencies (cf. Di Paolo 2009). To use an<br />
example that will be discussed at length in this <strong>the</strong>sis, we can note that a process of<br />
social interaction is enabled <strong>and</strong> constrained by <strong>the</strong> behavior of autonomous individuals,<br />
but that <strong>the</strong> behavioral capacity of <strong>the</strong>se interacting individuals is simultaneously<br />
enabled <strong>and</strong> constrained by <strong>the</strong> dynamics of <strong>the</strong> autonomous interaction process (e.g.<br />
<strong>Froese</strong> & Di Paolo 2008). The characterization of this kind of interdependency as a<br />
form of <strong>continuity</strong> is justified by <strong>the</strong> fact that <strong>the</strong> same conceptual framework is applied<br />
at all levels, in this case both for <strong>the</strong> description of behavioral <strong>and</strong> social dynamics.<br />
Moreover, as Thompson (2007, p. 129) points out, <strong>the</strong> enactive approach goes fur<strong>the</strong>r<br />
than o<strong>the</strong>r <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>ories by following Hans Jonas‟ phenomenological<br />
claim that certain basic experiential categories that are needed to underst<strong>and</strong> human<br />
experience turn out to be applicable to <strong>life</strong> itself:<br />
The great contradictions which man discovers in himself – freedom <strong>and</strong><br />
necessity, autonomy <strong>and</strong> mortality – have <strong>the</strong>ir rudimentary traces in even <strong>the</strong><br />
most primitive forms of <strong>life</strong>, each precariously balanced between being <strong>and</strong> notbeing,<br />
<strong>and</strong> each already endowed with an internal horizon of „transcendence‟.<br />
(Jonas 1966, p. ix)<br />
In o<strong>the</strong>r words, <strong>the</strong> LMCT is not only based on an organizational (or behavioral)<br />
<strong>continuity</strong>, but also on a corresponding phenomenological <strong>continuity</strong>. In this manner our<br />
36 | P a g e
underst<strong>and</strong>ing of <strong>the</strong> phenomenon of <strong>life</strong> can be seen to comprise biology <strong>and</strong> cognitive<br />
science, as well as <strong>the</strong> philosophy of <strong>the</strong> organism <strong>and</strong> philosophy of <strong>mind</strong>. To be sure,<br />
<strong>the</strong> development of such a radical LMCT is not without its problems:<br />
The danger, of course, is that by stressing unity <strong>and</strong> similarity we may lose sight<br />
of what is special <strong>and</strong> distinctive. Mind may indeed participate in many of <strong>the</strong><br />
dynamic processes of <strong>life</strong>. But what about our old friends, <strong>the</strong> fundamentally<br />
reason-based transitions <strong>and</strong> <strong>the</strong> grasp of absent <strong>and</strong> <strong>the</strong> abstract characteristic of<br />
advanced cognition? (Clark 2001, pp. 118-119)<br />
We can unpack this concern into two related but distinct aspects, namely <strong>the</strong> problem of<br />
agency <strong>and</strong> scalability. Thus, on <strong>the</strong> one h<strong>and</strong>, <strong>the</strong>re is <strong>the</strong> morally <strong>and</strong> scientifically<br />
motivated worry that <strong>the</strong> LMCT “threatens to eliminate <strong>the</strong> idea of purposive agency<br />
unless it is combined with some recognition of <strong>the</strong> special way goals <strong>and</strong> knowledge<br />
figure in <strong>the</strong> origination of some of our bodily motions” (Clark 2001, p. 135). In<br />
response to this concern it is important to emphasize that <strong>the</strong> enactive approach is<br />
acutely aware of <strong>the</strong> problem of agency, <strong>and</strong> most of its efforts are directed toward<br />
gaining a better underst<strong>and</strong>ing of this phenomenon (cf. Di Paolo 2009; Moreno &<br />
Etxeberria 2005). Indeed, <strong>the</strong> very turn toward <strong>the</strong> LMCT is largely motivated by a<br />
perceived lack of any coherent notion of agency in current cognitive science, <strong>and</strong> <strong>the</strong><br />
possibility that a closer examination of biological autonomy can fill this gap.<br />
However, <strong>the</strong>re still remains ano<strong>the</strong>r problem that is closely associated with <strong>the</strong> LMCT:<br />
“What, in general, is <strong>the</strong> relation between <strong>the</strong> strategies used to solve basic problems of<br />
perception <strong>and</strong> action <strong>and</strong> those used to solve more abstract or higher level problems?”<br />
(Clark 2001, p. 135). Is it a question of mere complexity, of just having more of <strong>the</strong><br />
same kind of organizations <strong>and</strong> mechanisms? Then why is it seemingly impossible to<br />
properly address <strong>the</strong> hallmarks of human cognition with <strong>the</strong>se basic principles? In a<br />
recent paper, De Jaegher <strong>and</strong> <strong>Froese</strong> (2009) have referred to this missing link as <strong>the</strong><br />
„cognitive gap‟ of <strong>the</strong> LMCT. They propose that this gap is a symptom of <strong>the</strong> still<br />
prevalent methodological individualism of cognitive science (cf. Boden 2006b), i.e. an<br />
exclusive focus on individual agency, <strong>and</strong> that it can be addressed by taking <strong>the</strong> role of<br />
sociality into account.<br />
37 | P a g e
To be sure, a related response has been developed by „extended <strong>mind</strong>‟ <strong>the</strong>orists such as<br />
Clark, who proposes “to depict much of advanced cognition as rooted in <strong>the</strong> operation<br />
of <strong>the</strong> same basic kinds of capacity used for on-line, adaptive response, but tuned <strong>and</strong><br />
applied to <strong>the</strong> special domain of external <strong>and</strong>/or artificial cognitive aids” (2001, p. 141).<br />
However, <strong>the</strong>se efforts have largely focused on <strong>the</strong> role of language (e.g. Clark 2008,<br />
pp. 44-60) <strong>and</strong> technology (e.g. Clark 2003), <strong>the</strong>reby relating specifically human<br />
cognition with specifically human abilities <strong>and</strong> <strong>the</strong>ir cultural context. Thus, while this<br />
consideration of „cognitive technology‟ helps to spread <strong>the</strong> explanatory burden outside<br />
of <strong>the</strong> individual human agent, <strong>and</strong> <strong>the</strong>reby indeed makes basic embodied-embedded<br />
accounts more plausible, it still leaves <strong>the</strong> main cognitive gap of <strong>the</strong> LMCT largely<br />
unaddressed.<br />
It is certainly crucial to adopt an externalist view of cognition as a first step to make <strong>the</strong><br />
LMCT plausible. But this entails nothing more than a commitment to <strong>the</strong> hypo<strong>the</strong>sis of<br />
embodied-embedded cognitive science that cognition emerges out of <strong>the</strong> dynamics of a<br />
brain-body-world systemic whole (e.g. Beer 2000). What is additionally needed is a<br />
non-species-specific operational mechanism to account for <strong>the</strong> transformative potential<br />
of such cognitive extension. To be sure, <strong>the</strong> desirability of a more encompassing<br />
account is not denied by extended <strong>mind</strong> <strong>the</strong>orists. Clark (2005), for example, suggests<br />
<strong>the</strong> sound-amplifying burrow of <strong>the</strong> mole cricket as a loose analogy to <strong>the</strong> cognitiontransforming<br />
symbols found in human culture. But it is important to note that <strong>the</strong><br />
chirping cricket in its burrow is passively interacting with a static physical structure.<br />
Moreover, this example completely ignores <strong>the</strong> fact that human symbols only exist<br />
within a social context.<br />
Accordingly, De Jaegher <strong>and</strong> <strong>Froese</strong> (in press) would agree with Clark that “interactive<br />
complexity characterizes almost all forms of advanced human cognitive endeavor”<br />
(Clark 2001, p. 154), but <strong>the</strong>y argue that such interactive complexity is already<br />
prefigured in <strong>the</strong> interactive <strong>and</strong> social co-constitution of more basic cognitive domains.<br />
Even simple interactions between agents can give rise to an interaction process<br />
characterized by autonomous dynamics that self-sustain by modulating <strong>the</strong> behavior of<br />
<strong>the</strong> interactors. Here we have an example of cognitive extension that involves active<br />
38 | P a g e
coordination, dynamic emerging structures, <strong>and</strong> an interactive social context that<br />
removes <strong>the</strong> need for static physical structures. In <strong>the</strong> enactive approach to social<br />
cognition this transformative potential of <strong>the</strong> interaction process is <strong>the</strong> basis of what has<br />
been called „participatory sense-making‟ (De Jaegher 2006). In order to underst<strong>and</strong> this<br />
notion properly we will have to introduce <strong>the</strong> basic conceptual framework of enactive<br />
cognitive science in more detail, beginning with its origins in <strong>the</strong> autopoietic tradition.<br />
3.2 Constitutive autonomy is necessary for intrinsic teleology<br />
The notion of autopoiesis (from Greek: self-producing) as <strong>the</strong> minimal organization of<br />
<strong>the</strong> living first originated in <strong>the</strong> work of <strong>the</strong> Chilean biologists Maturana <strong>and</strong> Varela in<br />
<strong>the</strong> 1970s (e.g. Maturana & Varela 1980; for a more accessible introduction, cf.<br />
Maturana & Varela 1987). While <strong>the</strong> concept was developed in <strong>the</strong> context of<br />
<strong>the</strong>oretical biology, it was right from its inception also associated with computer<br />
simulations (Varela, et al. 1974) long before <strong>the</strong> term „artificial <strong>life</strong>‟ was first introduced<br />
in <strong>the</strong> late 1980s by Langton (1989). Nowadays <strong>the</strong> concept of autopoiesis continues to<br />
have a significant impact on <strong>the</strong> field of artificial <strong>life</strong> in both <strong>the</strong> computational <strong>and</strong><br />
chemical domain (see McMullin (2004) <strong>and</strong> Luisi (2003), respectively, for overviews of<br />
<strong>the</strong>se two kinds of approaches). Moreover, <strong>the</strong>re have been recent efforts of more tightly<br />
integrating <strong>the</strong> notion of autopoiesis into <strong>the</strong> overall framework of enactive cognitive<br />
science (e.g. Weber & Varela 2002; Thompson 2007; 2005; Di Paolo 2005; 2009;<br />
McGann 2007; Colombetti, in press). The reasons for this ongoing integration will be<br />
clarified in this chapter.<br />
What precisely is autopoiesis? During <strong>the</strong> time after <strong>the</strong> notion of autopoiesis was first<br />
coined in 1971 9 its exact definition has slowly evolved in <strong>the</strong> works of both Maturana<br />
<strong>and</strong> Varela (cf. Thompson 2007, pp. 99-101; Bourgine & Stewart 2004). For <strong>the</strong><br />
purposes of this article we will use a definition that has been used extensively by Varela<br />
in a series of publications throughout <strong>the</strong> 1990s (e.g. Varela 1991; 1992; 1997), but<br />
which has also been used as <strong>the</strong> definition of choice in more recent work (e.g. Weber &<br />
9 See Varela (1996a) <strong>and</strong> Maturana (2002) for more detailed accounts of <strong>the</strong> historical circumstances<br />
under which <strong>the</strong> notion of autopoiesis was first conceived <strong>and</strong> developed.<br />
39 | P a g e
Varela 2002; Di Paolo 2003; 2005; <strong>Froese</strong>, et al. 2007). This more-or-less st<strong>and</strong>ard<br />
definition states:<br />
An autopoietic system – <strong>the</strong> minimal living organization – is one that<br />
continuously produces <strong>the</strong> components that specify it, while at <strong>the</strong> same time<br />
realizing it (<strong>the</strong> system) as a concrete unity in space <strong>and</strong> time, which makes <strong>the</strong><br />
network of production of components possible. More precisely defined: an<br />
autopoietic system is organized (defined as a unity) as a network of processes of<br />
production (syn<strong>the</strong>sis <strong>and</strong> destruction) of components such that <strong>the</strong>se components:<br />
1. continuously regenerate <strong>and</strong> realize <strong>the</strong> network that produces <strong>the</strong>m, <strong>and</strong><br />
2. constitute <strong>the</strong> system as a distinguishable unity in <strong>the</strong> domain in which <strong>the</strong>y<br />
exist.<br />
(Varela 1997, p. 75)<br />
In addition to <strong>the</strong>se two explicit criteria for autopoiesis we can add ano<strong>the</strong>r important<br />
point, namely that <strong>the</strong> self-constitution of an identity entails <strong>the</strong> constitution of a<br />
relational domain between <strong>the</strong> system <strong>and</strong> its environment. The shape of this domain is<br />
not pre-given but ra<strong>the</strong>r co-determined by <strong>the</strong> organization of <strong>the</strong> system, as it is<br />
produced by that system, <strong>and</strong> its environment. Accordingly, any system which fulfils<br />
<strong>the</strong> criteria for autopoiesis also generates its own domain of possible interactions in <strong>the</strong><br />
same movement that gives rise to its emergent identity (Thompson 2007, p. 44).<br />
Considering that current embodied AI fails to fully capture what is needed for <strong>life</strong>-like,<br />
intentional agency (cf. Chapter 2), it is interesting to note that <strong>the</strong> autopoietic tradition<br />
has been explicitly referred to by Varela (1992) as a „biology of intentionality‟. In o<strong>the</strong>r<br />
words, for enactive cognitive science <strong>the</strong> phenomenon of autopoiesis not only captures<br />
<strong>the</strong> basic mode of identity of <strong>the</strong> living, but is moreover at <strong>the</strong> root at how living beings<br />
enact <strong>the</strong>ir world of significance. Thus, <strong>the</strong> notion of autopoiesis in many ways<br />
continues a particular philosophy of <strong>the</strong> organism, such as Kant‟s, von Uexküll‟s, <strong>and</strong><br />
Jonas‟ intuitions regarding <strong>the</strong> organization of <strong>the</strong> living (cf. <strong>Froese</strong> & Ziemke 2009, pp.<br />
476-479). However, as a more recent development, it has <strong>the</strong> added advantage that it<br />
formalizes <strong>the</strong>se intuitions in a systemic, operational manner.<br />
40 | P a g e
The term „operational‟ denotes that <strong>the</strong> autopoietic definition of <strong>life</strong> can be used to<br />
distinguish living from non-living entities on <strong>the</strong> basis of a concrete instance <strong>and</strong><br />
without recourse to wider contextual (e.g. functional, historical) considerations.<br />
Autopoiesis can be considered as a response to <strong>the</strong> question of how we can determine<br />
whe<strong>the</strong>r or not a system is a living being on <strong>the</strong> basis of what kind of system it is ra<strong>the</strong>r<br />
than on how it behaves or where it came from. As such it can be contrasted with<br />
functional (e.g. Nagel 1977) or historical (e.g. Millikan 1989) approaches to teleology.<br />
Already Kant (1790) speculated that since a living system is characterized by a form of selforganizing<br />
reciprocal causality, it follows that all relations of cause <strong>and</strong> effect in <strong>the</strong> system<br />
are also at <strong>the</strong> same time relations of means <strong>and</strong> purpose. More importantly, this reciprocal<br />
causality entails that such a natural purpose <strong>the</strong>n, as an interrelated totality of means <strong>and</strong><br />
goals, is strictly intrinsic to <strong>the</strong> organism (Weber & Varela 2002). Kant‟s philosophy thus<br />
provides <strong>the</strong> beginning of a <strong>the</strong>ory of <strong>the</strong> self-producing organization of <strong>life</strong>, which attempts<br />
to capture <strong>the</strong> observation that organisms generate <strong>the</strong>ir own goals. In o<strong>the</strong>r words, a living<br />
system, as an autopoietic system, is both cause <strong>and</strong> effect of itself, <strong>and</strong> <strong>the</strong>refore also of <strong>the</strong><br />
feedback systems underlying its goal-directed behavior. This intrinsic generation of goals is<br />
generally lacking in current AI systems (cf. Haselager 2005). Embodied-embedded AI made<br />
an advance when it included its systems within a sensory-motor loop (e.g. Cliff 1991), but<br />
<strong>the</strong>se systems are never<strong>the</strong>less lacking <strong>the</strong> kind of intrinsic teleology that is characteristic of<br />
biological systems.<br />
The paradigmatic instance of an autopoietic system is a minimal, living cell (Varela, et<br />
al. 1974), which is often cited as an illustration of <strong>the</strong> circularity that is inherent in<br />
metabolic self-production. In <strong>the</strong> case of <strong>the</strong> cell this circularity is expressed in <strong>the</strong> codependency<br />
between <strong>the</strong> (boundary) semi-permeable membrane <strong>and</strong> <strong>the</strong> (internal)<br />
metabolic network. The metabolic network constructs itself as well as <strong>the</strong> membrane,<br />
<strong>and</strong> <strong>the</strong>reby distinguishes itself as a unified system from <strong>the</strong> (external) environment. In<br />
turn, <strong>the</strong> membrane boundary makes <strong>the</strong> metabolism possible by preventing <strong>the</strong> network<br />
from fatally diffusing into <strong>the</strong> environment.<br />
While <strong>the</strong>re are cases in <strong>the</strong> literature where multi-cellular organisms are also classed as<br />
autopoietic systems in <strong>the</strong>ir own right, this is an issue that is far from trivial <strong>and</strong> still<br />
41 | P a g e
emains controversial (cf. Thompson 2007, pp. 105-107). For instance, would <strong>the</strong>re be a<br />
difference whe<strong>the</strong>r <strong>the</strong> multi-cellular organism is distinguished on <strong>the</strong> level of chemical<br />
processes, cells, or organs? Whatever <strong>the</strong> case, we intuitively want to say that such<br />
organisms meet <strong>the</strong> requirements for autonomy. A multi-cellular organism might be<br />
different from an autopoietic minimal entity in its mode of identity, but it is also<br />
essentially similar at an abstract level of organization: its activity demarcates it as an<br />
entity from its environment (Varela 1991).<br />
In <strong>the</strong> late 1970s Varela became dissatisfied with <strong>the</strong> way that <strong>the</strong> concept of autopoiesis<br />
was starting to be applied loosely to o<strong>the</strong>r systems, with its use even extended to nonmaterial<br />
systems such as social institutions. He complained that such characterizations<br />
“confuse autopoiesis with autonomy” (Varela 1979, p. 55). Never<strong>the</strong>less, <strong>the</strong>re was still<br />
a need to make <strong>the</strong> explanatory power offered by <strong>the</strong> systemic approach to autonomy<br />
available for use in o<strong>the</strong>r contexts than <strong>the</strong> molecular domain. Thus, while autopoiesis is<br />
a form of autonomy in <strong>the</strong> biochemical domain, “to qualify as autonomy, however, a<br />
system does not have to be autopoietic in <strong>the</strong> strict sense (a self-producing bounded<br />
molecular system)” (Thompson 2007, p. 44).<br />
Accordingly, Varela put forward <strong>the</strong> notion of organizational closure 10 by taking “<strong>the</strong><br />
lessons offered by <strong>the</strong> autonomy of living systems <strong>and</strong> convert <strong>the</strong>m into an operational<br />
characterization of autonomy in general, living or o<strong>the</strong>rwise” (Varela 1979, p. 55):<br />
We shall say that autonomous systems are organizationally closed. That is, <strong>the</strong>ir<br />
organization is characterized by processes such that<br />
1. <strong>the</strong> processes are related as a network, so that <strong>the</strong>y recursively depend on<br />
each o<strong>the</strong>r in <strong>the</strong> generation <strong>and</strong> realization of <strong>the</strong> processes <strong>the</strong>mselves, <strong>and</strong><br />
10 In recent literature <strong>the</strong> term organizational closure is often used more or less interchangeably with <strong>the</strong><br />
notion of operational closure. However, <strong>the</strong> latter seems better suited to describe any system which has<br />
been distinguished in a certain epistemological manner by an external observer, namely so as not to view<br />
<strong>the</strong> system under study as characterized by inputs/outputs, but ra<strong>the</strong>r as a self-contained system which is<br />
parametrically coupled to its environment. On this view, an organizationally closed system is a special<br />
kind of system, namely one which is characterized by some form of self-production or identity-generation<br />
when it is appropriately distinguished by an external observer in an operationally closed manner.<br />
42 | P a g e
2. <strong>the</strong>y constitute <strong>the</strong> system as a unity recognizable in <strong>the</strong> space (domain) in<br />
which <strong>the</strong> processes exist.<br />
(Varela 1979, p. 55)<br />
This definition of autonomy applies to multi-cellular organisms (Maturana & Varela<br />
1987, pp. 88-89), but moreover to a whole host of o<strong>the</strong>r systems such as <strong>the</strong> immune<br />
system, <strong>the</strong> nervous system, <strong>and</strong> even to social systems (Varela 1991). Maturana <strong>and</strong><br />
Varela (1987) introduced a couple of simple ideograms to denote systems which are<br />
characterized by organizational closure (Figure 3-1):<br />
Figure 3-1. Maturana <strong>and</strong> Varela‟s ideograms for autonomous systems, namely those systems which can<br />
be characterized by organizational closure. The ideogram on <strong>the</strong> left depicts a basic autonomous system:<br />
<strong>the</strong> closed arrow circle indicates <strong>the</strong> system with organizational closure, <strong>the</strong> rippled line its environment,<br />
<strong>and</strong> <strong>the</strong> bidirectional half-arrows <strong>the</strong> ongoing structural coupling between <strong>the</strong> two. The ideogram on <strong>the</strong><br />
right extends this basic picture by introducing ano<strong>the</strong>r organizational closure within <strong>the</strong> autonomous<br />
system, which could be <strong>the</strong> nervous system, for example.<br />
We will refer to <strong>the</strong> autonomy entailed by organizational closure as constitutive<br />
autonomy in order to demarcate it from <strong>the</strong> concept‟s more general usage (cf. <strong>Froese</strong>, et<br />
al. 2007). Since it does not specify <strong>the</strong> particular domain of <strong>the</strong> autonomous system, it is<br />
also to some extent more amenable to <strong>the</strong> sciences of <strong>the</strong> artificial, though some<br />
fundamental problems remain (cf. <strong>Froese</strong> & Di Paolo 2008b). For a more detailed<br />
description of how <strong>the</strong> notions of emergence through self-organization, constitutive<br />
autonomy, <strong>and</strong> autopoiesis relate to each o<strong>the</strong>r, see <strong>Froese</strong> <strong>and</strong> Ziemke (2009),<br />
especially Appendix C (p. 497).<br />
In summary, when we are referring to an autonomous system we denote a system<br />
composed of several processes that actively generate <strong>and</strong> sustain <strong>the</strong>ir systemic identity<br />
43 | P a g e
under precarious conditions (cf. Di Paolo & Iizuka 2008). The precariousness of <strong>the</strong><br />
identity is explicitly mentioned in order to emphasize that <strong>the</strong> system‟s identity is<br />
actively constituted by <strong>the</strong> system under conditions which tend toward its disintegration,<br />
<strong>and</strong> which is <strong>the</strong>refore constantly under threat of ceasing to exist. Accordingly, this<br />
working definition of constitutive autonomy captures <strong>the</strong> essential insights of both <strong>the</strong><br />
situation of <strong>the</strong> organism as described in <strong>the</strong> philosophical biology tradition, as well as<br />
<strong>the</strong> operational definitions provided by <strong>the</strong> autopoietic tradition. Both of <strong>the</strong>se traditions<br />
converge on <strong>the</strong> claim that it is this self-constitution of an identity, an identity that could<br />
at each moment become something different or disappear altoge<strong>the</strong>r, which grounds our<br />
underst<strong>and</strong>ing of intrinsic teleology (Weber & Varela 2002). Living systems are not just<br />
goal-directed because <strong>the</strong>y are feedback systems; <strong>the</strong>y are also <strong>the</strong> source of those goals<br />
because <strong>the</strong>y are autonomous systems. These considerations allow us to state <strong>the</strong> first<br />
core claim of <strong>the</strong> enactive paradigm as a systemic requirement (SR-1): autonomy is<br />
necessary <strong>and</strong> sufficient for intrinsic teleology 11 .<br />
3.3 Adaptivity is necessary for sense-making<br />
In contrast to <strong>the</strong> agents of embodied AI whose identity <strong>and</strong> domain of interactions are<br />
externally defined, constitutively autonomous systems (SR-1) bring forth <strong>the</strong>ir own<br />
identity <strong>and</strong> domain of interactions, <strong>and</strong> <strong>the</strong>reby constitute <strong>the</strong>ir own „problems to be<br />
solved‟ according to <strong>the</strong>ir particular affordances for action (SR-2). Such autonomous<br />
systems <strong>and</strong> <strong>the</strong>ir worlds st<strong>and</strong> in relation to each o<strong>the</strong>r through mutual specification or<br />
co-determination (Varela 1992). In o<strong>the</strong>r words, <strong>the</strong>re is a mutual dependence between<br />
<strong>the</strong> intentional agent (which must exist in some world) <strong>and</strong> its world (which can only be<br />
encountered by such an agent): in addition to self-production, <strong>the</strong>re is thus ano<strong>the</strong>r<br />
fundamental circularity at <strong>the</strong> core of intentionality (cf. McGann 2007).<br />
Fur<strong>the</strong>rmore, what an autonomous system does, due to its precarious mode of identity,<br />
is to treat <strong>the</strong> perturbations it encounters from a perspective of significance which is not<br />
11 <strong>Froese</strong> <strong>and</strong> Ziemke (2009) give a weaker version of SR-1, claiming that autonomy is merely necessary<br />
for intrinsic teleology. However, since any autonomous system always operates according to at least one<br />
internally defined goal, namely self-production, it is more accurate to claim sufficiency as well.<br />
44 | P a g e
intrinsic to <strong>the</strong> encounters <strong>the</strong>mselves. In o<strong>the</strong>r words, <strong>the</strong> meaning of an encounter is<br />
not determined by that encounter. Instead it is evaluated in relation to <strong>the</strong> ongoing<br />
maintenance of <strong>the</strong> self-constituted identity, <strong>and</strong> <strong>the</strong>reby acquires a meaning which is<br />
relative to <strong>the</strong> current situation of <strong>the</strong> agent <strong>and</strong> its needs. This process of meaning<br />
generation in relation to <strong>the</strong> perspective of <strong>the</strong> agent is what is meant by <strong>the</strong> notion of<br />
sense-making (Weber & Varela 2002). Translating this concept into von Uexküll‟s<br />
(1934) terms we could say that sense-making is <strong>the</strong> ongoing process of active<br />
constitution of an Umwelt for <strong>the</strong> organism.<br />
It is important to note that <strong>the</strong> significance which is continuously brought forth by <strong>the</strong><br />
endogenous activity of <strong>the</strong> autonomous agent is what makes <strong>the</strong> world, as it appears<br />
from <strong>the</strong> perspective of that agent, distinct from <strong>the</strong> physical environment of <strong>the</strong><br />
autonomous system, as it is distinguished by an external observer (Varela 1997). Sensemaking<br />
is <strong>the</strong> enaction of a meaningful world for <strong>the</strong> autonomous agent.<br />
Note that <strong>the</strong> enactive account of autonomy <strong>and</strong> sense-making entails that meaning is<br />
not to be found in <strong>the</strong> elements belonging to <strong>the</strong> environment or in <strong>the</strong> internal dynamics<br />
of <strong>the</strong> agent alone. Instead, meaning is an aspect of <strong>the</strong> relational domain established<br />
between <strong>the</strong> two (Di Paolo, et al. in press). It depends on <strong>the</strong> specific mode of codetermination<br />
that an autonomous system realizes with its specific environment, <strong>and</strong><br />
accordingly different modes of structural coupling will give rise to different meanings<br />
(Colombetti, in press). However, it is also important to note that <strong>the</strong> claim that meaning<br />
is grounded in such relations does not entail that meaning can be reduced to those<br />
relational phenomena. There is an asymmetry underlying <strong>the</strong> relational domain of an<br />
autonomous system since <strong>the</strong> very existence of that domain is continuously enacted by<br />
<strong>the</strong> endogenous activity of that system. In contrast to most embodied AI, where <strong>the</strong><br />
relational domain exists no matter what <strong>the</strong> system is or does, <strong>the</strong> relational domain of a<br />
living system is not pre-given but depends on precarious processes of self-production. It<br />
follows from this that any model that only captures <strong>the</strong> relational dynamics on <strong>the</strong>ir<br />
own, as is <strong>the</strong> case with most work on sensory-motor situatedness, will only be able to<br />
capture <strong>the</strong> functional aspects of <strong>the</strong> behavior. A functional model will not reproduce<br />
<strong>the</strong> intrinsic meaning such behavior would have for an autonomous system whose<br />
existence is constitutively linked with its relational domain.<br />
45 | P a g e
In order for <strong>the</strong>se considerations to be of more specific use for <strong>the</strong> development of better<br />
models of natural cognition, we need to unpack <strong>the</strong> notion of sense-making in more<br />
detail. Essentially, it requires that <strong>the</strong> perturbations which an autonomous agent<br />
encounters through its ongoing interactions must somehow acquire a valence that is<br />
related to <strong>the</strong> agent‟s viability. Varela (1992) has argued that <strong>the</strong> source of this worldmaking<br />
is always <strong>the</strong> breakdowns in autopoiesis. However, <strong>the</strong> concept of autopoiesis<br />
(or constitutive autonomy more generally) by itself allows no gradation – ei<strong>the</strong>r a<br />
system belongs to <strong>the</strong> class of such systems or it does not. The self-constitution of an<br />
identity can thus provide us only with <strong>the</strong> most basic kind of norm, namely that all<br />
events are good for that identity as long as <strong>the</strong>y do not destroy it (<strong>and</strong> <strong>the</strong> latter events<br />
do not carry any significance because <strong>the</strong>re will be no more identity to which <strong>the</strong>y could<br />
even be related). On this basis alone <strong>the</strong>re is no room for accounting for <strong>the</strong> different<br />
shades of meaning which are constitutive of an organism‟s Umwelt. Fur<strong>the</strong>rmore, <strong>the</strong><br />
operational definitions of autopoiesis <strong>and</strong> constitutive autonomy nei<strong>the</strong>r require that<br />
such a system can actively compensate for deleterious internal or external events, nor<br />
address <strong>the</strong> possibility that it can spontaneously improve its current situation. What is<br />
missing from <strong>the</strong>se definitions? How can we extend <strong>the</strong> meaningful perspective that is<br />
engendered by constitutive autonomy into a wider context of relevance?<br />
Di Paolo (2005) has recently proposed a resolution of this problem. He starts from <strong>the</strong><br />
observation that minimal autopoietic systems have a certain kind of tolerance or<br />
robustness: <strong>the</strong>y can sustain a certain range of perturbations as well as a certain range of<br />
internal structural changes before <strong>the</strong>y lose <strong>the</strong>ir autopoiesis, where <strong>the</strong>se ranges are<br />
defined by <strong>the</strong> organization <strong>and</strong> current state of <strong>the</strong> system. We can <strong>the</strong>n define <strong>the</strong>se<br />
ranges of non-fatal events as an autonomous system‟s viability set, which is “assumed to<br />
be of finite measure, bounded, <strong>and</strong> possibly time-varying” (Di Paolo 2005, p. 438).<br />
However, in order for an autopoietic system to actively improve its current situation, it<br />
must (i) be capable of determining how <strong>the</strong> ongoing structural changes are shaping its<br />
trajectory within its viability set, <strong>and</strong> (ii) have <strong>the</strong> capacity to regulate <strong>the</strong> conditions of<br />
this trajectory appropriately. These two criteria are provided by <strong>the</strong> property of<br />
adaptivity, for which Di Paolo (2005) provides <strong>the</strong> following definition:<br />
46 | P a g e
A system‟s capacity, in some circumstances, to regulate its states <strong>and</strong> its relation<br />
to <strong>the</strong> environment with <strong>the</strong> result that, if <strong>the</strong> states are sufficiently close to <strong>the</strong><br />
boundary of viability,<br />
1. Tendencies are distinguished <strong>and</strong> acted upon depending on whe<strong>the</strong>r <strong>the</strong><br />
states will approach or recede from <strong>the</strong> boundary <strong>and</strong>, as a consequence,<br />
2. Tendencies of <strong>the</strong> first kind are moved closer to or transformed into<br />
tendencies of <strong>the</strong> second <strong>and</strong> so future states are prevented from reaching <strong>the</strong><br />
boundary with an outward velocity.<br />
(Di Paolo 2005, p. 438)<br />
Similar to <strong>the</strong> case of robustness, <strong>the</strong> notion of adaptivity 12 implies tolerance of a range<br />
of internal <strong>and</strong> external perturbations. However, in this context it entails a special kind<br />
of context-sensitive tolerance which involves both actively monitoring perturbations<br />
<strong>and</strong> compensating for <strong>the</strong>ir tendencies, ra<strong>the</strong>r than mere homeostasis. In this context <strong>the</strong><br />
notion of active monitoring <strong>and</strong> compensating not only refers to <strong>the</strong> asymmetry of selfproduction,<br />
which entails that <strong>the</strong> system is <strong>the</strong> active source of activity, but also to an<br />
internal differentiation of operations that involves some partially decoupled, specialized<br />
adaptive mechanisms (cf. Bar<strong>and</strong>iaran & Moreno 2008). The explicit requirement of<br />
active monitoring is crucial for two reasons: (i) it allows <strong>the</strong> system to distinguish<br />
between positive <strong>and</strong> negative tendencies, <strong>and</strong> (ii) it ensures that <strong>the</strong> system can<br />
measure <strong>the</strong> type <strong>and</strong> severity of a tendency according to a change in <strong>the</strong> internal,<br />
regulative resources that are required for compensation of negative tendencies.<br />
It is important to note that <strong>the</strong> capacity for (i) does not contradict <strong>the</strong> organizational<br />
closure of <strong>the</strong> autonomous system because of (ii). In o<strong>the</strong>r words, <strong>the</strong> system does not<br />
have any special epistemic access to an independent (non-relational) environment, <strong>and</strong> it<br />
<strong>the</strong>refore does not violate <strong>the</strong> relational nature of constitutive autonomy, but this is not a<br />
problem since it only needs to monitor internal effort. Fur<strong>the</strong>rmore, it is worth<br />
emphasizing that <strong>the</strong> capacity for (ii) already implies <strong>the</strong> need for suitable<br />
12 Note that this form of adaptivity, as a special kind of self-regulatory mechanism, must be clearly<br />
distinguished from <strong>the</strong> more general notion of „adaptedness‟. This latter sense is usually used to indicate<br />
all viable behaviour that has evolutionary origins <strong>and</strong> contributes to reproductive success.<br />
47 | P a g e
compensation. In <strong>the</strong> context of sense-making we can <strong>the</strong>refore say that both elements,<br />
i.e. self-monitoring <strong>and</strong> appropriate regulation, are necessary to be able to speak of<br />
different kinds of meaning from <strong>the</strong> perspective of <strong>the</strong> organism. Thus, “if autopoiesis<br />
in <strong>the</strong> present analysis suffices for generating a natural purpose, adaptivity reflects <strong>the</strong><br />
organism‟s capability – necessary for sense-making – of evaluating <strong>the</strong> needs <strong>and</strong><br />
exp<strong>and</strong>ing <strong>the</strong> means towards that purpose” (Di Paolo 2005, p. 445).<br />
While it is likely that some form of adaptivity as it is defined here was assumed to be<br />
implicit in <strong>the</strong> definition of autopoiesis as constitutive of sense-making by Weber <strong>and</strong><br />
Varela. Never<strong>the</strong>less, it is useful to turn this implicit assumption into an explicit,<br />
operational specification. Di Paolo‟s work thus allows us to state <strong>the</strong> second core claim<br />
of <strong>the</strong> enactive paradigm in <strong>the</strong> form of ano<strong>the</strong>r systemic requirement (SR-2): adaptivity<br />
is necessary for sense-making. In <strong>the</strong> next section we will use a recent debate on <strong>the</strong><br />
relationship between autopoiesis <strong>and</strong> cognition to illustrate SR-1 <strong>and</strong> SR-2. It will be<br />
argued that autonomy <strong>and</strong> adaptivity are necessary <strong>and</strong> sufficient for sense-making.<br />
3.4 Constitutive autonomy is necessary for sense-making<br />
We have argued that <strong>the</strong> systemic requirements of autonomy <strong>and</strong> adaptivity are<br />
necessary for intrinsic teleology <strong>and</strong> sense-making, respectively. Toge<strong>the</strong>r <strong>the</strong>y are also<br />
necessary <strong>and</strong> sufficient for sense-making <strong>and</strong> adaptive agency, where <strong>the</strong> latter is<br />
defined as an agent capable of adaptive behavior in a purposeful <strong>and</strong> meaningful<br />
context. However, we are not making <strong>the</strong> stronger claim that autonomy <strong>and</strong> adaptivity<br />
are also sufficient conditions for cognitive agency. In fact, we expect that more systemic<br />
requirements will be added to this list as <strong>the</strong> enactive approach begins to address a<br />
wider range of phenomena. Some promising lines of research in this regard are <strong>the</strong><br />
development of an enactive approach to cognition (Bar<strong>and</strong>iaran & Moreno 2006), to<br />
emotion <strong>the</strong>ory (Colombetti, in press), to goals <strong>and</strong> goal-directedness (McGann 2007)<br />
<strong>and</strong> to social cognition (De Jaegher & Di Paolo 2007). All of <strong>the</strong>se developments are<br />
consistent <strong>and</strong> continuous with <strong>the</strong> fundamental notions of autonomy <strong>and</strong> sense-making<br />
as <strong>the</strong>y have been presented here, <strong>and</strong> we will return to some of <strong>the</strong>se additional<br />
requirements in Chapter 4. We can now summarize <strong>the</strong> insights of <strong>the</strong> previous sections<br />
as shown in Table 3-1.<br />
48 | P a g e
# Systemic requirement Entailment Normativity<br />
SR-1 autonomy intrinsic teleology uniform<br />
SR-2 adaptivity sense-making graded<br />
Table 3-1. Summary of <strong>the</strong> enactive approach to intentional agency, which includes at least two systemic<br />
requirements: (SR-1) autonomy is necessary <strong>and</strong> sufficient for intrinsic teleology, <strong>and</strong> (SR-2) adaptivity is<br />
necessary for sense-making. Since <strong>the</strong> viability constraints of adaptivity depend on <strong>the</strong> autonomous<br />
identity, we can say that both autonomy <strong>and</strong> adaptivity are necessary <strong>and</strong> sufficient for sense-making.<br />
When it comes to <strong>the</strong> practical challenge of how to go about realizing <strong>the</strong> two systemic<br />
requirements in <strong>the</strong> form of artificial systems it might be tempting to initially avoid<br />
tackling SR-1 <strong>and</strong> SR2 in combination. Would it not be better to implement <strong>the</strong>m as<br />
independent modules first <strong>and</strong> <strong>the</strong>n think about integrating <strong>the</strong>m later? Evolutionary<br />
roboticists in particular will want to avoid SR-1, which we will call <strong>the</strong> „hard problem‟<br />
of enactive AI (cf. <strong>Froese</strong> & Ziemke 2009), <strong>and</strong> first focus on <strong>the</strong> problem of SR-2<br />
alone. However, is it possible to design artificial systems with adaptivity as <strong>the</strong> basis for<br />
sense-making independently of constitutive autonomy? Conversely, those interested in<br />
modeling <strong>the</strong> chemical basis of autonomy might be inclined to approach things <strong>the</strong> o<strong>the</strong>r<br />
way around. Is autopoiesis perhaps not sufficient for sense-making after all?<br />
While an affirmative answer to <strong>the</strong>se questions might sound desirable for AI research,<br />
unfortunately things are not that simple. This is best illustrated by an analysis of <strong>the</strong><br />
relationship between autopoiesis <strong>and</strong> cognition as it has been presented by Bourgine <strong>and</strong><br />
Stewart (2004) in a paper which bases its insights on a ma<strong>the</strong>matical model of<br />
autopoiesis. Whereas traditionally it was held that „autopoiesis = <strong>life</strong> = cognition‟ (e.g.<br />
Maturana & Varela 1980; Stewart 1996; 1992), Bourgine <strong>and</strong> Stewart propose:<br />
Analytically, <strong>the</strong> interactions between a system <strong>and</strong> its environment can be<br />
subdivided into two sorts […]. Firstly, <strong>the</strong>re are those interactions that have<br />
consequences for <strong>the</strong> internal state of <strong>the</strong> organism: we may call <strong>the</strong>se type A<br />
interactions. Secondly, <strong>the</strong>re are those interactions that have consequences for<br />
<strong>the</strong> state of <strong>the</strong> (proximal) environment, or that modify <strong>the</strong> relation of <strong>the</strong><br />
system to its environment: we may call <strong>the</strong>se type B interactions. This<br />
49 | P a g e
terminology allows us to propose a definition of “cognition”: A system is<br />
cognitive if <strong>and</strong> only if type A interactions serve to trigger type B interactions<br />
in a specific way, so as to satisfy a viability constraint. (Bourgine & Stewart<br />
2004, p. 338)<br />
Of course, in most situations type A interactions can be termed „sensations‟, <strong>and</strong> type B<br />
interactions can be termed „actions‟ 13 . Bourgine <strong>and</strong> Stewart also note that <strong>the</strong>ir notion<br />
of „viability constraint‟ has been deliberately left vague so that <strong>the</strong>ir definition of<br />
cognition (similar to what we have been calling „adaptivity‟) can by „metaphorical<br />
extension‟ also be usefully applied to non-living systems (cf. Bourgine & Stewart 2004,<br />
pp. 338-339). This is supposed to make room for <strong>the</strong> „autonomous‟ robots designed in<br />
<strong>the</strong> field of artificial <strong>life</strong>.<br />
As a hypo<strong>the</strong>tical example <strong>the</strong>y describe a robot that navigates on <strong>the</strong> surface of a table<br />
by satisfying <strong>the</strong> constraints of nei<strong>the</strong>r remaining immobile nor falling off <strong>the</strong> edge.<br />
Since this robot is cognitive by definition when it satisfies <strong>the</strong> imposed viability<br />
constraint, but certainly not autopoietic, Bourgine <strong>and</strong> Stewart claim that autopoiesis is<br />
not a necessary condition for cognition (in contrast to what has been argued here in<br />
Sections 3.2 <strong>and</strong> 3.3). Fur<strong>the</strong>rmore, <strong>the</strong>y provide a ma<strong>the</strong>matical model of a simple<br />
chemical system, which <strong>the</strong>y maintain is autopoietic but for which it is never<strong>the</strong>less<br />
impossible to speak of „action‟ <strong>and</strong> „sensation‟ in any meaningful manner. Accordingly,<br />
<strong>the</strong>y also make <strong>the</strong> second claim that autopoiesis is not a sufficient condition for<br />
cognition, a claim which appears to be compatible with SR-2.<br />
However, while this last claim might sound at least vaguely analogous to Di Paolo‟s<br />
argument that minimal autopoiesis is insufficient to account for sense-making, <strong>the</strong>re are<br />
some important differences. It is worth noting that, as side effect of not fur<strong>the</strong>r<br />
restricting what is to count as a „viability constraint‟, Bourgine <strong>and</strong> Stewart‟s definition<br />
of cognition is different from Di Paolo‟s notion of sense-making for two important<br />
13 They fur<strong>the</strong>r clarify <strong>the</strong>ir position by stating: “It is only analytically that we can separate sensory inputs<br />
<strong>and</strong> actions; since <strong>the</strong> sensory inputs guide <strong>the</strong> actions, but <strong>the</strong> actions have consequences for subsequent<br />
sensory inputs, <strong>the</strong> two toge<strong>the</strong>r form a dynamic loop. Cognition, in <strong>the</strong> present perspective, amounts to<br />
<strong>the</strong> emergent characteristics of this dynamical system.” (Bourgine & Stewart 2004, p. 339)<br />
50 | P a g e
easons: (i) <strong>the</strong> viability constraint can be externally defined (as illustrated by <strong>the</strong><br />
example of <strong>the</strong> robot), <strong>and</strong> (ii) even if <strong>the</strong> viability constraint was intrinsic to <strong>the</strong><br />
cognitive system, <strong>the</strong>re is no explicit requirement for that system to actively measure<br />
<strong>and</strong> actively regulate its performance with regard to satisfying that constraint.<br />
To illustrate <strong>the</strong> consequences of (i) we can imagine defining an additional arbitrary<br />
constraint for <strong>the</strong> hypo<strong>the</strong>tical navigating robot, namely that it must also always stay on<br />
only one side of <strong>the</strong> table. Accordingly, we would have to treat it as cognitive as long as<br />
it happens to stay on that side, but as non-cognitive as soon as it moves to <strong>the</strong> o<strong>the</strong>r side<br />
of <strong>the</strong> table. Clearly, whe<strong>the</strong>r <strong>the</strong> robot stays on one side or <strong>the</strong> o<strong>the</strong>r does not make any<br />
difference to <strong>the</strong> system itself but only to <strong>the</strong> experimenter who is imposing <strong>the</strong> viability<br />
criteria (whe<strong>the</strong>r <strong>the</strong>se criteria are externalized into a component of <strong>the</strong> robot or not).<br />
Thus, <strong>the</strong> only overlap between Bourgine <strong>and</strong> Stewart‟s definition of cognition <strong>and</strong> Di<br />
Paolo‟s conception of sense-making is that both require <strong>the</strong> capacity for some form of<br />
sensory-motor interaction, a capacity which is not sufficient for grounding meaning by<br />
itself (<strong>Froese</strong> & Ziemke 2009).<br />
It will also be interesting from <strong>the</strong> perspective of AI to draw out <strong>the</strong> consequences of <strong>the</strong><br />
second difference (ii). Bourgine <strong>and</strong> Stewart‟s claim that a given interaction between a<br />
system <strong>and</strong> its environment “will not be cognitive unless <strong>the</strong> consequences for <strong>the</strong><br />
internal state of <strong>the</strong> system are employed to trigger specific actions that promote <strong>the</strong><br />
viability of <strong>the</strong> system” (2004, p. 338). How do we know what constitutes an action as<br />
opposed to mere physical change? They define actions as “those interactions that have<br />
consequences for <strong>the</strong> state of <strong>the</strong> (proximal) environment, or that modify <strong>the</strong> relation of<br />
<strong>the</strong> system to its environment” (2004, p. 338). However, this criterion is trivially met by<br />
all systems which are structurally coupled to <strong>the</strong>ir environment since any kind of<br />
interaction (whe<strong>the</strong>r originating from <strong>the</strong> system or <strong>the</strong> environment) changes <strong>the</strong><br />
relation of <strong>the</strong> system to its environment at some level of description. Thus, while <strong>the</strong>ir<br />
definition enables <strong>the</strong> movement of <strong>the</strong> hypo<strong>the</strong>tical navigating robot to be classed as an<br />
action, it also has <strong>the</strong> undesirable effect of making it impossible to distinguish whe<strong>the</strong>r<br />
it is <strong>the</strong> system or <strong>the</strong> environment that is <strong>the</strong> „agent‟ giving rise to this action. In order<br />
to remove this ambiguity we can adopt Di Paolo‟s view on adaptivity:<br />
51 | P a g e
[There is an] important distinction between structural coupling <strong>and</strong> <strong>the</strong><br />
regulation of structural coupling. The former is an ongoing happening, <strong>the</strong><br />
necessary outcome of non-lethal physical encounters between organism <strong>and</strong><br />
medium. Only <strong>the</strong> latter, <strong>the</strong> parametrical action that regulates coupling, fully<br />
deserves <strong>the</strong> name of behaviour because such regulation is done by <strong>the</strong><br />
organism […] as opposed to simply being undergone by it. Unregulated<br />
coupling is better described as suffering an exchange while behaviour is <strong>the</strong><br />
control <strong>and</strong> selection of what exchanges to suffer. (Di Paolo 2005, p. 442)<br />
As such, autonomy <strong>and</strong> <strong>the</strong> regulative capacity of adaptivity can account for <strong>the</strong> fact that<br />
“cognition requires a natural centre of activity on <strong>the</strong> world as well as a natural<br />
perspective on it” (Di Paolo 2005, p. 443). We have already seen that it is <strong>the</strong> principle<br />
of autonomy which introduces this required asymmetry: an autonomous system brings<br />
forth <strong>the</strong> relational domain that forms <strong>the</strong> basis for adaptive regulation by constituting<br />
its own identity, which is <strong>the</strong> reference point for its domain of possible interactions. It is<br />
essentially <strong>the</strong> lack of this relational asymmetry in Bourgine <strong>and</strong> Stewart‟s conception<br />
of cognition, which has made <strong>the</strong>ir proposal problematic.<br />
From this discussion we can conclude that autonomy <strong>and</strong> adaptivity are both necessary<br />
<strong>and</strong> sufficient for sense-making. Accordingly, while it might be desirable from a<br />
practical engineering <strong>and</strong> scientific approach to treat autonomy <strong>and</strong> adaptivity as<br />
separate requirements for sense-making, this abstraction might be <strong>the</strong> root problem for<br />
<strong>the</strong> continuing difficulties faced by embodied AI. Better models of natural cognition<br />
will require us to combine <strong>the</strong> two core systemic requirements of <strong>the</strong> enactive paradigm,<br />
SR-1 <strong>and</strong> SR-2, into one internally integrated system. Note, however, that <strong>the</strong> systemic<br />
requirements are not quite as constraining as <strong>the</strong>y might at first appear: an operational<br />
definition says nothing about how <strong>the</strong> required organization is structurally realized.<br />
The line of reasoning that we have developed in this section is fur<strong>the</strong>r supported by<br />
recent work on chemical autopoiesis by Bitbol <strong>and</strong> Luisi (2004). While <strong>the</strong>y broadly<br />
agree with Bourgine <strong>and</strong> Stewart‟s definition of cognition, provided that extended<br />
homeostasis is considered to be a special variety of sensory-motor behavior, <strong>the</strong>y<br />
never<strong>the</strong>less reject its proposed radical dissociation from autopoiesis. Thus, while<br />
52 | P a g e
Maturana <strong>and</strong> Varela‟s original provocative assertion was that autopoiesis is strictly<br />
equivalent to cognition, Bitbol <strong>and</strong> Luisi weaken this claim slightly by holding that<br />
minimal cognition requires both (i) <strong>the</strong> self-constitution of an identity (autonomy), <strong>and</strong><br />
(ii) dynamical interaction with <strong>the</strong> environment. Since <strong>the</strong>y maintain that minimal<br />
autopoiesis is sufficient for (i) but does not necessarily entail (ii), <strong>the</strong>ir position, like Di<br />
Paolo‟s, falls between <strong>the</strong> extreme positions of radical identity (e.g. Maturana & Varela<br />
1980; Stewart 1992; 1996) <strong>and</strong> radical dissociation (e.g. Bourgine & Stewart 2004).<br />
The upshot of <strong>the</strong>se last two sections is that <strong>the</strong> enactive paradigm might have <strong>the</strong><br />
conceptual tools to effectively diagnose <strong>the</strong> problems which have prevented embodied<br />
AI from designing more <strong>life</strong>-like artificial systems. In a nutshell, it turns out that<br />
sensory-motor interaction alone is not sufficient to ground intrinsic meaning or goal<br />
ownership, <strong>and</strong> it does not entail sense-making or cognition. The embodied AI approach<br />
has attempted to capture what biological agents do, e.g. <strong>the</strong>ir sensory-motor behavior,<br />
but while leaving out what <strong>the</strong>se agents are, e.g. autonomous <strong>and</strong> adaptive 14 . Ironically,<br />
<strong>the</strong> supposedly „Heideggerian AI‟ has ignored <strong>the</strong> question of being.<br />
3.5 Summary<br />
This has been a long chapter dealing mainly with issues that belong to <strong>the</strong>oretical<br />
biology. How do <strong>the</strong>se systemic foundations relate to <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong>? In order<br />
to answer this question we will briefly relate <strong>the</strong> main argument, i.e. that relational<br />
phenomena such as cognition, behavior, <strong>and</strong> sense-making cannot be decoupled from<br />
<strong>the</strong> operations of an autonomous <strong>and</strong> adaptive system without rendering <strong>the</strong>m<br />
intrinsically meaningless, to <strong>the</strong> rest of enactive cognitive science. Effectively, this<br />
summary will be a first pass through <strong>the</strong> enactive version of <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong>.<br />
As <strong>the</strong> starting point it is important to realize that an autonomous system, by generating<br />
its own identity in separation of what it is not, simultaneously generates <strong>the</strong> particular<br />
14 For a more extensive discussion of why natural agency <strong>and</strong> cognition entail autonomy <strong>and</strong> adaptivity in<br />
terms of more detailed biological considerations, see Bar<strong>and</strong>iaran <strong>and</strong> Moreno (2006; 2008) <strong>and</strong> Moreno<br />
<strong>and</strong> Etxeberria (2005). For <strong>the</strong> most up-to-date enactive account of agency that will likely be influential<br />
for future work, see Bar<strong>and</strong>iaran, Di Paolo <strong>and</strong> Rohde (2009).<br />
53 | P a g e
conditions by which it can relate to its environment. Indeed, it is this fundamental<br />
asymmetry of <strong>the</strong> organism-environment relationship which partly constitutes <strong>the</strong><br />
organism‟s perspective on that environment:<br />
Now, in this dialogic coupling between <strong>the</strong> living unity <strong>and</strong> <strong>the</strong> physicochemical<br />
environment, <strong>the</strong>re is a key difference on <strong>the</strong> side of <strong>the</strong> living since it has <strong>the</strong><br />
active role in this reciprocal coupling. In defining what it is as unity, in <strong>the</strong> very<br />
same movement it defines what remains exterior to it, that is to say, its<br />
surrounding environment. […] <strong>the</strong> autopoietic unity creates a perspective from<br />
which <strong>the</strong> exterior is one, which cannot be confused with <strong>the</strong> physical<br />
surroundings as <strong>the</strong>y appear to us as observers, <strong>the</strong> l<strong>and</strong> of physical <strong>and</strong> chemical<br />
laws simpliciter, devoid of such perspectivism. (Varela 1997, p. 78)<br />
Of course, this is not to say that we cannot provide a description of <strong>the</strong> organism <strong>and</strong> its<br />
environment in physicochemical terms 15 . The point is merely that this type of<br />
description does not exhaust <strong>the</strong> domain of phenomena with which biology should be<br />
concerned: “One could envisage <strong>the</strong> circularity metabolism-membrane entirely from <strong>the</strong><br />
outside (this is what most biochemists do). But this is not to deny that <strong>the</strong>re is, at <strong>the</strong><br />
same time, <strong>the</strong> instauration of a point of view provided by <strong>the</strong> self-construction” (Weber<br />
& Varela 2002, p. 116). More precisely, this generation of a point of view for <strong>the</strong><br />
organism consists in two essential aspects, namely <strong>the</strong> constitution of (i) an identity, <strong>and</strong><br />
(ii) a relationship to what is „o<strong>the</strong>r‟, whereby (i) acts as a reference point for (ii):<br />
In o<strong>the</strong>r words by putting at <strong>the</strong> center <strong>the</strong> autonomy of even <strong>the</strong> minimal<br />
cellular organism we inescapably find an intrinsic teleology in two<br />
complementary modes. First, a basic purpose in <strong>the</strong> maintenance of its own<br />
identity, an affirmation of <strong>life</strong>. Second, directly emerging from <strong>the</strong> aspect of<br />
concern to affirm <strong>life</strong>, a sense-creation purpose whence meaning comes to its<br />
15 Nor should Varela be misunderstood as implying that <strong>the</strong> physical surroundings that are described by<br />
scientists are devoid of perspectivism in that <strong>the</strong>y reflect an absolute reality (cf. Varela, et al. 1991, pp. 9-<br />
12). We can better underst<strong>and</strong> his point by realizing that <strong>the</strong> scientific description of <strong>the</strong> surroundings is<br />
generated precisely by stripping <strong>the</strong> world of its significance (cf. <strong>Dr</strong>eyfus 1991, pp. 112-121), a capacity<br />
that depends on a highly developed sensitivity to intersubjectivity (cf. Chapter 12).<br />
54 | P a g e
surrounding, introducing a difference between environment (<strong>the</strong> physical<br />
impacts it receives), <strong>and</strong> world (how that environment is evaluated from <strong>the</strong><br />
point of view established by maintaining an identity). (Weber & Varela 2002, p.<br />
117)<br />
Let us briefly consider <strong>the</strong>se two aspects in turn. First, <strong>the</strong>re is <strong>the</strong> notion of intrinsic<br />
teleology in terms of <strong>the</strong> organism‟s relation to its own identity, i.e. its capacity to<br />
constitute its own purposeful <strong>and</strong> goal-directed existence. Weber <strong>and</strong> Varela begin to<br />
derive this notion by combining Kant‟s conception of a natural purpose, namely <strong>the</strong><br />
idea that a self-organizing system that is both cause <strong>and</strong> effect of itself is also its own<br />
means <strong>and</strong> purpose (cf. Kant 1790, §64-65), with our modern underst<strong>and</strong>ing of<br />
autopoiesis. Then, by appealing to <strong>the</strong> philosophical biology of Jonas, <strong>the</strong>y move<br />
beyond Kant‟s conception of teleology as a useful regulative idea for <strong>the</strong> observer, <strong>and</strong><br />
posit teleology as intrinsic to <strong>the</strong> phenomenon of <strong>life</strong> itself. Indeed, Jonas argues that <strong>the</strong><br />
precarious situation of <strong>the</strong> living furnishes <strong>the</strong> organism with more than just an<br />
inherently purposeful existence: “The organism has to keep going, because to be going<br />
is its very existence – which is revocable – <strong>and</strong>, threatened with extinction, it is<br />
concerned in existing” (Jonas 1966, p. 126, emphasis added). He thus argues that it is<br />
<strong>the</strong> generation of a precarious identity through self-production that simultaneously<br />
enables <strong>the</strong> generation of existential values. Poetically expressed:<br />
The basic clue is that <strong>life</strong> says yes to itself. By clinging to itself it declares that it<br />
values itself. […] Are we <strong>the</strong>n, perhaps, allowed to say that mortality is <strong>the</strong><br />
narrow gate through which alone value – <strong>the</strong> addressee of a yes – could enter <strong>the</strong><br />
o<strong>the</strong>rwise indifferent universe? (Jonas 1992, p. 36)<br />
This brings us to <strong>the</strong> second aspect of <strong>the</strong> organism‟s perspective, namely its capacity<br />
for sense-creation or sense-making. This notion highlights that <strong>the</strong> generation of values<br />
always happens in <strong>the</strong> context of a particular organism-environment relationship. The<br />
internal <strong>and</strong> external encounters that perturb <strong>the</strong> process of identity generation take on a<br />
value in relation to this perturbation: “The perspective of a challenged <strong>and</strong> selfaffirming<br />
organism lays a new grid over <strong>the</strong> world: a ubiquitous scale of value. To have<br />
a world for an organism thus first <strong>and</strong> foremost means to have value which it brings<br />
55 | P a g e
forth by <strong>the</strong> very process of its identity” (Weber & Varela 2002, p. 118). In o<strong>the</strong>r words,<br />
an organism‟s world is first <strong>and</strong> foremost a meaningful context that is related to its<br />
particular manner of realizing its identity. To quote a famous example from Varela:<br />
“There is no food significance in sucrose except when a bacteria swims upgradient <strong>and</strong><br />
its metabolism uses <strong>the</strong> molecule in a way that allows its identity to continue” (1997, p.<br />
79). This nicely illustrates how meaningful behavior entails an autonomous identity.<br />
One of <strong>the</strong> most important consequences of <strong>the</strong> argument that we have developed in this<br />
chapter is that it strongly underlines <strong>the</strong> deep <strong>continuity</strong> between <strong>life</strong> <strong>and</strong> <strong>mind</strong>, <strong>and</strong> it is<br />
this <strong>continuity</strong> which forms <strong>the</strong> very core of <strong>the</strong> <strong>the</strong>oretical foundation of enactive<br />
cognitive science. In order to better illustrate this link between <strong>the</strong> systemic approach to<br />
biology, as it has been presented in this section of <strong>the</strong> paper, <strong>and</strong> <strong>the</strong> enactive approach<br />
as a cognitive science research program, we have adapted Thompson‟s (2004) five steps<br />
from <strong>life</strong> to <strong>mind</strong> for <strong>the</strong> present context. As a first rough pass we can say that:<br />
1. Life = constitutive autonomy + adaptivity 16<br />
2. Constitutive autonomy entails emergence of an identity<br />
3. Emergence of an adaptive identity entails emergence of a world<br />
4. Emergence of adaptive identity <strong>and</strong> world = sense-making<br />
5. Sense-making = cognition<br />
This chapter has provided only <strong>the</strong> basic <strong>the</strong>oretical elements for an underst<strong>and</strong>ing of<br />
<strong>the</strong>se five steps. We would only like to emphasize that, as Thompson points out, <strong>the</strong>se<br />
steps amount to an explicit hypo<strong>the</strong>sis about <strong>the</strong> natural roots of intentionality. In o<strong>the</strong>r<br />
words, <strong>the</strong>y form <strong>the</strong> basis of <strong>the</strong> claim that <strong>the</strong> „aboutness‟ of our cognition is not due<br />
to some presumed representational content that is matched to an independent external<br />
reality (by some designer or evolution), but is ra<strong>the</strong>r related to <strong>the</strong> significance that is<br />
continually enacted by <strong>the</strong> precarious activity of <strong>the</strong> organism during its ongoing<br />
encounters with <strong>the</strong> environment. Here we thus have <strong>the</strong> beginnings of how <strong>the</strong> enactive<br />
16 Step 1 is a more refined version of <strong>the</strong> traditional claim that „autopoiesis = <strong>life</strong>‟, though it is likely that<br />
<strong>the</strong> additional requirement of adaptivity was already implied by a looser conception of autopoiesis (cf.<br />
Section 3.3 of this chapter). Also, as will become clear in Section 4.2 of Chapter 4, it is more accurate to<br />
say that cognition entails sense-making, but that sense-making alone is not sufficient for cognition.<br />
56 | P a g e
approach might go about incorporating its two main foundations, namely<br />
phenomenological philosophy <strong>and</strong> systems biology, into one coherent <strong>the</strong>oretical<br />
framework (cf. Thompson 2007).<br />
Let us close this chapter by highlighting some issues for future research in this area. The<br />
concept of adaptivity has enabled us to become much clearer about what we mean by<br />
<strong>the</strong> notion of sense-making <strong>and</strong> its necessary conditions of realization. Our own<br />
perspective, <strong>and</strong> that of most o<strong>the</strong>r organisms, too, is evidently characterized by a whole<br />
range of different shades of meaning, <strong>and</strong> this phenomenal differentiation had to be<br />
matched in operational terms by a notion capable of giving rise to such graded<br />
distinctions. However, while this conceptual advance is an important accomplishment in<br />
itself, it is none<strong>the</strong>less just <strong>the</strong> beginning of <strong>the</strong> task of developing a more precise notion<br />
of sense-making. Indeed, <strong>the</strong> term‟s general applicability to all living beings, except<br />
perhaps for a few degenerate bacteria that have lost <strong>the</strong>ir adaptive mechanisms (cf.<br />
Bar<strong>and</strong>iaran & Moreno 2008, pp. 333-334), cries out for fur<strong>the</strong>r specification. It follows<br />
that more work needs to be done in order to account for different kinds of sense-making<br />
activities <strong>and</strong> <strong>the</strong>ir qualitative variations. How might we account for this variation?<br />
One possibility is to fur<strong>the</strong>r develop <strong>the</strong> emotive aspects of sense-making, for example<br />
in terms of a bodily cognitive-emotional form of underst<strong>and</strong>ing (e.g. Colombetti, in<br />
press). Such work is particularly important with respect to providing an enactive<br />
account of a special type of autonomous agent, namely animals, which are specifically<br />
characterized by motility, perception, <strong>and</strong> emotion (cf. Jonas 1966, pp. 99-107).<br />
Ano<strong>the</strong>r approach is to fur<strong>the</strong>r clarify <strong>the</strong> way in which sense-making is related to action<br />
(e.g. Thompson 2005). This can happen in terms of basic adaptivity (Di Paolo 2005), as<br />
well as for <strong>the</strong> kind of goals <strong>and</strong> goal-directedness that are specifically characteristic of<br />
human agency (e.g. McGann 2007), in particular with respect to <strong>the</strong> role of play (e.g. Di<br />
Paolo, et al., in press). All of <strong>the</strong>se are promising avenues of fur<strong>the</strong>r research. The<br />
approach pursued in <strong>the</strong> next chapter, however, is to clarify how sense-making is<br />
transformed by interaction in a social context, especially because this will enable us to<br />
develop a principled response to <strong>the</strong> problem of <strong>the</strong> LMCT‟s „cognitive gap‟.<br />
57 | P a g e
4 The enactive approach to social cognition<br />
In this chapter we refine <strong>the</strong> concepts introduced in <strong>the</strong> previous chapter in order to<br />
develop <strong>the</strong> enactive approach to social cognition from <strong>the</strong> most basic forms of interindividual<br />
interaction to cultural interaction. First, <strong>the</strong> notion of adaptive agency is<br />
introduced as <strong>the</strong> most basic form of agency that can become part of a multi-agent<br />
system, in which inter-individual interactions can <strong>the</strong>mselves take on an autonomous<br />
organization. This is followed by a consideration of <strong>the</strong> conditions for cognitive agency<br />
<strong>and</strong> properly social interaction. On this basis a possible definition of cultural interaction<br />
is suggested. Finally, <strong>the</strong> main arguments are summarized in relation to <strong>the</strong> <strong>life</strong>-<strong>mind</strong><br />
<strong>continuity</strong> <strong>the</strong>sis.<br />
4.1 The autonomy of <strong>the</strong> interaction process<br />
In Chapter 3 we defined an autonomous system to be a system composed of several<br />
processes that actively generate <strong>and</strong> sustain <strong>the</strong>ir systemic identity under precarious<br />
conditions. More precisely, an autonomous system is a network of processes in which<br />
each constituent process has as part of its enabling conditions one or more o<strong>the</strong>r<br />
processes in this network, <strong>and</strong> is itself also an enabling condition for one or more o<strong>the</strong>r<br />
processes. It is this organizational closure which underlies <strong>the</strong> self-constitution of an<br />
identity by that very system. The existence of this identity is qualified as being<br />
precarious in order to emphasize that <strong>the</strong> constituent processes would disintegrate in <strong>the</strong><br />
absence of this autonomous organization. This notion of autonomy is fundamental to <strong>the</strong><br />
enactive approach because it provides a way to naturalize <strong>the</strong> concept of an identity with<br />
intrinsic teleology. In o<strong>the</strong>r words, only to a system which is characterized by autonomy<br />
is it possible to attribute goal-states that belong to that system itself, ra<strong>the</strong>r than being<br />
imposed on <strong>the</strong> system from <strong>the</strong> outside by some external designer, structure or process.<br />
Intrinsic teleology should not be misunderstood in an anthropomorphic fashion; it is<br />
merely a way of specifying a certain quality of systemic behavior.<br />
58 | P a g e
Figure 4-1. The relationship between constitutive autonomy <strong>and</strong> adaptive agency: <strong>the</strong> autonomous system<br />
self-constitutes an identity which is conserved during structural coupling with its environment (black<br />
arrows); adaptive agency requires additional regulation by <strong>the</strong> system which is aimed at adjusting this<br />
coupling relationship appropriately (dotted arrows).<br />
It is important to emphasize again that <strong>the</strong> property of autonomy is a necessary but not a<br />
sufficient condition for adaptive agency (see Figure 4-1). Autonomy as such only<br />
ensures <strong>the</strong> passive conservation (homeostasis) of <strong>the</strong> self-constituted identity during<br />
structural coupling with <strong>the</strong> environment. In <strong>the</strong> previous chapter we argued that<br />
adaptivity, i.e. <strong>the</strong> capacity of an autonomous system to actively regulate its states in<br />
relation to self-constituted viability constraints, is also necessary for agency. We will<br />
now re-state this claim with more precision.<br />
As Bar<strong>and</strong>iaran <strong>and</strong> Moreno point out, adaptation can happen ei<strong>the</strong>r by means of <strong>the</strong><br />
internal reorganization of constructive processes, or by regulation of an extended<br />
interactive cycle; in both cases <strong>the</strong>re is some degree of decoupling from <strong>the</strong> basic<br />
constitutive processes: “we are now talking about two dynamic „levels‟ in <strong>the</strong> system:<br />
<strong>the</strong> constitutive level, which ensures ongoing self-construction, <strong>and</strong> <strong>the</strong> (now decoupled)<br />
interactive subsystem, which regulates boundary conditions of <strong>the</strong> former” (Bar<strong>and</strong>iaran<br />
& Moreno 2008, p. 332). It is only when <strong>the</strong> mechanisms of regulation operate by<br />
modulating structural coupling, such that adaptation is achieved through recursive<br />
interactions with <strong>the</strong> environment (interactive adaptivity), that we speak of adaptive<br />
agency. In contrast to internal compensation, this adaptive regulation of systemenvironment<br />
relations opens up a novel relational domain that can be traversed by<br />
means of behavior (i.e. regulated sensory-motor interactions).<br />
59 | P a g e
Of course, constitutive autonomy <strong>and</strong> interactive adaptivity are only <strong>the</strong> minimal<br />
conditions of agency (what we could call agency „mark I‟), exemplified, for example,<br />
by bacteria capable of performing chemotaxis. Many forms of <strong>life</strong> are likely to be more<br />
of a “veritable topology of processes of identity generation (intersecting, embedded,<br />
hiearchical, shared, etc.)” (Di Paolo 2009, p. 18). Still, <strong>the</strong> phenomenon of adaptive<br />
agency is sufficient to allow us to consider a simple extension to <strong>the</strong> basic scenario<br />
shown in Figure 4-1, namely by introducing two adaptive agents into a shared<br />
environment. This change results in <strong>the</strong> situation depicted in Figure 4-2.<br />
Figure 4-2. The relationship between two adaptive agents sharing <strong>the</strong> same environment: <strong>the</strong> manner in<br />
which one agent‟s movements affect <strong>the</strong> environment can result in changes to sensory stimulation for <strong>the</strong><br />
o<strong>the</strong>r agent, <strong>and</strong> vice versa, creating <strong>the</strong> basis for a multi-agent recursive interaction.<br />
The sensory stimulation of a solitary agent is largely determined by its own structure<br />
<strong>and</strong> movements, thus giving rise to a closed sensory-motor feedback loop. This closed<br />
loop makes it possible for <strong>the</strong> agent to engage in sensory-motor coordination so as to<br />
structure its own perceptual space (cf. Pfeifer & Scheier 1999, p. 377-434). However, in<br />
<strong>the</strong> case where two adaptive agents share a particular environment toge<strong>the</strong>r, one agent‟s<br />
movements can affect that environment in such a way that it results in changes of<br />
sensory stimulation for <strong>the</strong> o<strong>the</strong>r agent, <strong>and</strong> vice versa. Moreover, when <strong>the</strong>se changes<br />
in stimulation for one agent in turn lead to changes in its movement that change <strong>the</strong><br />
stimulation for <strong>the</strong> o<strong>the</strong>r agent, <strong>and</strong> so forth in a way that recursively sustains this<br />
mutual interaction, <strong>the</strong> result is a special kind of interaction process. This process can be<br />
characterized as an autonomous structure in <strong>the</strong> relational domain that is constituted by<br />
coordinated behaviors. Accordingly, we can simplify Figure 4-2 slightly by focusing on<br />
<strong>the</strong> autonomy of this interaction process, as shown in Figure 4-3.<br />
60 | P a g e
Figure 4-3. The autonomy of <strong>the</strong> interaction process: it is possible that when two adaptive agents share an<br />
environment <strong>and</strong> <strong>the</strong>y engage in sensory-motor interaction, that <strong>the</strong>ir activities become entwined in such a<br />
manner that <strong>the</strong>ir mutual interaction results in an autonomous interaction process.<br />
Famous examples of <strong>the</strong> emergence of autonomous structures from <strong>the</strong> interactions<br />
between adaptive agents are <strong>the</strong> slime molds (mycetozoans). Spores of Physarum begin<br />
<strong>life</strong> as unicellular amoebae, <strong>and</strong> multiply while feeding on bacteria. If <strong>the</strong>y encounter<br />
<strong>the</strong> correct mating type, <strong>the</strong>y can form zygotes which grow into large plasmodia, a<br />
unified <strong>life</strong> form containing many nuclei that are not separated by cell membranes. A<br />
plasmodium is even capable of migrating toward more favorable conditions by shifting<br />
concentrations of protoplasm. Note, however, that in this particular case <strong>the</strong> emergence<br />
of an autonomous structure in <strong>the</strong> relational domain actually coincides with <strong>the</strong><br />
dissolution of autonomy of <strong>the</strong> constitutive cells. It is <strong>the</strong>refore better described as a<br />
transformation between two types of adaptive agent, where <strong>the</strong> multi-agent interaction<br />
between amoebae is limited to a transitional phase. In <strong>the</strong> case of <strong>the</strong> slime mold<br />
Dycostelium, however, amoeboid individuals are capable of forming a fructiferous body<br />
without cellular fusion, <strong>and</strong> with a clear diversity of cellular types. To be sure, all multicellular<br />
organisms are clear examples that mutual interactions between adaptive agents<br />
(as defined above) can lead to <strong>the</strong> emergence of structures that are autonomous in <strong>the</strong>ir<br />
own right (cf. Maturana & Varela 1987, pp. 74-89).<br />
Note that <strong>the</strong> self-organized emergence of multi-cellularity highlights <strong>the</strong> importance of<br />
considering a developmental systems perspective (e.g. Oyama 2009) for overcoming <strong>the</strong><br />
„cognitive gap‟ of <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis. Thus, at first sight <strong>the</strong> task of<br />
establishing this <strong>continuity</strong> on <strong>the</strong> basis of insights gained from minimal, single-cell<br />
forms of <strong>life</strong> appears to equate <strong>the</strong> „gap‟ with <strong>the</strong> whole history of <strong>life</strong> on earth. Surely it<br />
would be better to start with something of medium complexity, as often practiced by<br />
61 | P a g e
embodied AI? But notice that whereas Brooks‟ insect-like robots still face an immense<br />
phylogenetic gap (<strong>and</strong> hence <strong>the</strong> provocative title of Kirsh„s (1991) paper „Today <strong>the</strong><br />
earwig tomorrow man?‟), <strong>the</strong> single-cell models often favored by <strong>the</strong> enactive paradigm<br />
can be viewed as confronting us with an ontogenetic gap instead. With this shift in<br />
perspective <strong>the</strong> cognitive gap has been narrowed from <strong>the</strong> whole extent of evolutionary<br />
history, to <strong>the</strong> developmental <strong>life</strong>span of a single human individual.<br />
While <strong>the</strong> integration of <strong>the</strong> developmental systems approach <strong>and</strong> <strong>the</strong> enactive paradigm<br />
is beyond <strong>the</strong> scope of this <strong>the</strong>sis, we will never<strong>the</strong>less focus our efforts on addressing<br />
<strong>the</strong> cognitive gap of <strong>the</strong> <strong>continuity</strong> <strong>the</strong>sis in terms of <strong>life</strong>time changes in behavior. More<br />
precisely, <strong>the</strong> task will be to investigate how „higher-level‟ autonomous identities can<br />
appear due to <strong>the</strong> coordinated interaction between two or more adaptive agents, <strong>and</strong> can<br />
be realized as more or less stable structures. In <strong>the</strong> most general terms we can define<br />
such a type of interaction as follows:<br />
Multi-agent interaction is <strong>the</strong> regulated coupling between at least two adaptive<br />
agents, where <strong>the</strong> regulation is aimed at aspects of <strong>the</strong> coupling itself so that it<br />
constitutes an emergent autonomous organization in <strong>the</strong> domain of relational<br />
dynamics, without destroying in <strong>the</strong> process <strong>the</strong> autonomy of <strong>the</strong> agents<br />
involved (though <strong>the</strong> latter‟s scope can be augmented or reduced).<br />
This definition is based on a related one proposed by De Jaegher <strong>and</strong> Di Paolo 17 , but it<br />
puts more specific requirements on <strong>the</strong> necessary form of agency (adaptive), <strong>and</strong> refers<br />
to this type of interaction as „multi-agent‟ ra<strong>the</strong>r than „social‟. The motivation for this<br />
distinction is that it gives us a more fine-grained conceptual h<strong>and</strong>le on <strong>the</strong> variety of<br />
phenomena that involve more than one agent, including a more specific definition of <strong>the</strong><br />
social which we will develop later in this chapter. Note that <strong>the</strong> definition of multi-agent<br />
17 De Jaegher <strong>and</strong> Di Paolo‟s definition reads: “Social interaction is <strong>the</strong> regulated coupling between at<br />
least two autonomous agents, where <strong>the</strong> regulation is aimed at aspects of <strong>the</strong> coupling itself so that it<br />
constitutes an emergent autonomous organization in <strong>the</strong> domain of relational dynamics, without<br />
destroying in <strong>the</strong> process <strong>the</strong> autonomy of <strong>the</strong> agents involved (though <strong>the</strong> latter‟s scope can be<br />
augmented or reduced)” (2007, p. 493). We will consider this definition more fully in Section 4.2.<br />
62 | P a g e
interaction is sufficiently abstract so as not to be limited to interactions between singlecell<br />
organisms. It is equally applicable to social interaction between humans as well.<br />
It is helpful to illustrate this idea briefly by means of a simple concrete case study which<br />
will be described in more detail later (cf. Chapter 6, p. 90). A recent psychological<br />
experiment by Auvray, Lenay, <strong>and</strong> Stewart (2009) has investigated <strong>the</strong> dynamics of<br />
human interaction under minimal conditions. Two participants were asked to locate<br />
each o<strong>the</strong>r in a simple 1-D virtual environment using only left-right movement <strong>and</strong> an<br />
all-or-nothing tactile feedback mechanism, which indicated whe<strong>the</strong>r <strong>the</strong>ir virtual<br />
„avatar‟ was overlapping any objects within <strong>the</strong> virtual space. They could encounter<br />
three types of objects: (i) a static object, (ii) <strong>the</strong> avatar of <strong>the</strong> o<strong>the</strong>r participant, <strong>and</strong> (iii) a<br />
„shadow‟ copy of <strong>the</strong> o<strong>the</strong>r participant‟s avatar that exactly mirrored <strong>the</strong> o<strong>the</strong>r‟s<br />
movement at a displaced location. Since all objects were of <strong>the</strong> same size <strong>and</strong> only<br />
generated an all-or-nothing tactile response, <strong>the</strong> only way to differentiate between <strong>the</strong>m<br />
was through <strong>the</strong> interaction dynamics that <strong>the</strong>y afforded. And, indeed, participants did<br />
manage to locate each o<strong>the</strong>r successfully because ongoing perceptual crossing afforded<br />
<strong>the</strong> most stable situation under <strong>the</strong>se circumstances. Thus, even though <strong>the</strong> participants<br />
„failed‟ to achieve <strong>the</strong> task individually, i.e. <strong>the</strong>re was no significant difference between<br />
<strong>the</strong>ir clicking response to <strong>the</strong> o<strong>the</strong>r‟s avatar <strong>and</strong> <strong>the</strong> o<strong>the</strong>r‟s shadow (cf. Auvray, et al.<br />
2009, p. 39), <strong>the</strong>y managed to solve <strong>the</strong> task because of <strong>the</strong> self-sustaining dynamics of<br />
<strong>the</strong> interaction process.<br />
Di Paolo <strong>and</strong> De Jaegher (2007; 2008) suggest ano<strong>the</strong>r paradigmatic example, namely<br />
any situation in which <strong>the</strong> individual interactors are attempting to stop interacting, but<br />
where <strong>the</strong> interaction process self-sustains even in spite of this intention. That can easily<br />
occur, for instance, when two people attempt to walk past each o<strong>the</strong>r in a corridor, but<br />
happen to move in mirroring directions at <strong>the</strong> same time. They <strong>the</strong>reby co-create a<br />
symmetrical coordinated relation, which is likely to result in <strong>the</strong>m moving in mirroring<br />
directions again, thus leading to fur<strong>the</strong>r interaction. In this case <strong>the</strong> individual intention<br />
of terminating <strong>the</strong> interaction process is actually prevented from being realized due to<br />
<strong>the</strong> emerging coordination patterns at <strong>the</strong> inter-individual level. In o<strong>the</strong>r words, in <strong>the</strong>se<br />
kinds of cases <strong>the</strong> overall organization of <strong>the</strong> interaction subsumes <strong>the</strong> individual actions<br />
of <strong>the</strong> interactors in such a way that <strong>the</strong> identity of <strong>the</strong> interactive situation is retained, at<br />
63 | P a g e
least temporarily, despite <strong>the</strong>ir efforts to <strong>the</strong> contrary. Accordingly, De Jaegher <strong>and</strong> Di<br />
Paolo suggest that <strong>the</strong> reciprocal relationship between <strong>the</strong> two autonomous domains,<br />
namely <strong>the</strong> individual <strong>and</strong> <strong>the</strong> interactional, may more easily be studied in situations<br />
where <strong>the</strong>y are in conflict.<br />
In sum, with this definition of multi-agent interaction we have taken a first step toward<br />
an enactive approach to social cognition. The co-constitution of an interactive cycle by<br />
two adaptive agents is a necessary (but not sufficient) condition. It is worth emphasizing<br />
that <strong>the</strong> insufficiency of multi-agent interactions with regard to sociality does not make<br />
it meaningless to investigate <strong>the</strong>m in <strong>the</strong>ir own right. On <strong>the</strong> contrary, it is clear that <strong>the</strong><br />
effects of such an interaction process are irreducible to individual capacities, <strong>and</strong> that<br />
<strong>the</strong>y can never<strong>the</strong>less significantly shape an individual‟s behavioral domain. Indeed, this<br />
intermediate level between <strong>the</strong> individual <strong>and</strong> <strong>the</strong> social <strong>the</strong>refore works in favor of <strong>the</strong><br />
<strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis because it shows <strong>the</strong> transformative potential of basic multiagent<br />
interactions even without <strong>the</strong> presence of sociality (i.e. without <strong>the</strong> need for <strong>the</strong><br />
presence of o<strong>the</strong>rs as such). A multi-agent system <strong>the</strong>refore provides <strong>the</strong> foundation for<br />
<strong>the</strong> emergence of more involved interactions.<br />
4.2 Social interaction<br />
The notion of „multi-agent interaction‟ has provided us with a general way of<br />
characterizing interactions between adaptive agents that result in autonomous structures,<br />
<strong>and</strong> which can radically alter <strong>the</strong> behavioral domains of <strong>the</strong> interacting individuals.<br />
However, as it st<strong>and</strong>s <strong>the</strong> notion is too broad to capture what is specific about social<br />
interactions. As a first step, we can note that <strong>the</strong>re is a mismatch of values: failure to<br />
regulate a social interaction does not necessarily imply a direct failure of material selfmaintenance.<br />
However, for an adaptive agent this independence of social purpose is<br />
impossible because its capacity for regulating interactions is, while partially decoupled<br />
from constructive processes, still too closely tied to its metabolic existence. The norms<br />
that are constitutive of its regulatory activity, while being potentially constrained by <strong>the</strong><br />
dynamics of multi-agent interaction, cannot be specifically social norms because <strong>the</strong>ir<br />
success is largely determined by basic energetic <strong>and</strong> material needs. What is needed for<br />
64 | P a g e
sociality is <strong>the</strong> creation of a new domain that can have its own internal coherency. The<br />
foundation for this social domain is provided by <strong>the</strong> cognitive domain.<br />
What is cognition? This question could be <strong>the</strong> topic of ano<strong>the</strong>r whole <strong>the</strong>sis, so we will<br />
restrict ourselves here to mentioning some essential aspects. In order to define what is<br />
special about <strong>the</strong> cognitive, we can fortunately draw on <strong>the</strong> work of Bar<strong>and</strong>iaran <strong>and</strong><br />
Moreno (2006; 2008) who have recently argued that cognition is given by <strong>the</strong> adaptive<br />
preservation of a dynamical network of autonomous sensory-motor structures sustained<br />
by continuous interactions with <strong>the</strong> environment <strong>and</strong> <strong>the</strong> body:<br />
The hierarchical decoupling achieved through <strong>the</strong> electrochemical functioning of<br />
neural interactions <strong>and</strong> <strong>the</strong>ir capacity to establish a highly connected <strong>and</strong> nonlinear<br />
network of interactions provides a dynamic domain with open-ended<br />
potentialities, not limited by <strong>the</strong> possibility of interference with basic metabolic<br />
processes (unlike diffusion processes in unicellular systems <strong>and</strong> plants). It is<br />
precisely <strong>the</strong> open-ended capacity of this high-dimensional domain that opens<br />
<strong>the</strong> door to spatial <strong>and</strong> temporal self-organization in neural dynamics <strong>and</strong><br />
generates an extremely rich dynamic domain mediating <strong>the</strong> interactive cycle,<br />
overcoming some limitations of previous sensorimotor control systems.<br />
(Bar<strong>and</strong>iaran & Moreno 2008, p. 338)<br />
A paradigmatic example of such structures are habits, which encompass partial aspects<br />
of <strong>the</strong> nervous system, physiological <strong>and</strong> structural systems of <strong>the</strong> body <strong>and</strong> patterns of<br />
behavior <strong>and</strong> processes in <strong>the</strong> environment (Di Paolo 2003). Due to <strong>the</strong> partial<br />
hierarchical decoupling of <strong>the</strong> electro-chemical activity of <strong>the</strong> nervous system from<br />
metabolic-constructive processes, <strong>the</strong> normative regulation of sensory-motor interaction<br />
is underdetermined by basic material <strong>and</strong> energetic needs. This is because <strong>the</strong> stability<br />
of a cognitive structure largely depends on <strong>the</strong> activity of <strong>the</strong> nervous system as well as<br />
<strong>the</strong> way <strong>the</strong> structure is coupled to sensory-motor correlations. In sum, following on<br />
from Bar<strong>and</strong>iaran <strong>and</strong> Moreno, we can define cognitive interaction as follows:<br />
Cognitive interaction is <strong>the</strong> regulated coupling between a dynamic agent <strong>and</strong> its<br />
environment, where <strong>the</strong> regulation is aimed at aspects of <strong>the</strong> coupling itself so<br />
65 | P a g e
that it constitutes an emergent autonomous organization in <strong>the</strong> domain of internal<br />
<strong>and</strong> relational dynamics, without destroying in <strong>the</strong> process <strong>the</strong> autonomy of <strong>the</strong><br />
agent (though <strong>the</strong> latter‟s scope can be augmented or reduced).<br />
It is important to emphasize once more that <strong>the</strong> minimal form of agency required for<br />
cognitive interaction (what we call „dynamic‟ or „cognitive‟ agency) is more complex<br />
than that provided by adaptive agency, though it is difficult to capture this difference in<br />
operational terms. Effectively, <strong>the</strong>re is a need for a hierarchically decoupled domain of<br />
dynamics that can generate its own, non-metabolic goals (e.g. determined by neurodynamic<br />
forms of autonomy), <strong>and</strong> be able to regulate its own activity <strong>and</strong> <strong>the</strong><br />
organism‟s sensory-motor behavior accordingly. Partly this is already a possibility for<br />
adaptive agents, since <strong>the</strong> mechanisms of adaptive regulation are partially decoupled<br />
from <strong>the</strong> metabolic-constructive processes. But <strong>the</strong> behavior of <strong>the</strong>se agents is limited<br />
because <strong>the</strong> regulatory goals are largely determined by metabolic needs, ra<strong>the</strong>r than by<br />
<strong>the</strong> activity that is generated via sensory-motor interaction <strong>and</strong> within <strong>the</strong> adaptive<br />
mechanism itself. Cognition, as an open-ended domain of behavior, only becomes<br />
possible when <strong>the</strong> adaptive mechanism is partially decoupled from <strong>the</strong> rest of <strong>the</strong> body<br />
in such a way that it is possible for autonomous structures to arise via recurrent<br />
dynamics (cf. Bar<strong>and</strong>iaran & Moreno 2006, p. 180). This form of dynamic agency<br />
(what we might call agency „mark II‟), is typically based on <strong>the</strong> nervous system.<br />
Once dynamic agency is in place it is possible that <strong>the</strong> continuation of certain patterns<br />
of sensory-motor interaction become goals in <strong>the</strong>mselves, for example due to <strong>the</strong><br />
autonomous dynamic structures which <strong>the</strong>y induce in neural activity. Moreover, <strong>the</strong>se<br />
patterns can involve coordination with ano<strong>the</strong>r agent in multi-agent system. Thus, only<br />
an agent capable of cognitive interaction can help to give rise to a social domain that is<br />
defined by its own specific normativity 18 . But is cognitive interaction in a multi-agent<br />
18 Might this be <strong>the</strong> beginning of a radicalization of <strong>the</strong> „social brain hypo<strong>the</strong>sis‟ (cf. Dunbar 1998)? The<br />
question is why a form of <strong>life</strong> should evolve that is controlled by a system whose operations are largely<br />
decoupled from its essential metabolic (self-relative) values, especially since this immediacy increases <strong>the</strong><br />
precariousness of <strong>the</strong> organism. But perhaps this is <strong>the</strong> price to pay for being able to regulate behavior in<br />
relation to social (o<strong>the</strong>r-relative) values? Is sociality related to <strong>the</strong> origin of <strong>the</strong> nervous system as such?<br />
66 | P a g e
system necessary <strong>and</strong> sufficient for that interaction to be called social? What is <strong>the</strong><br />
precise role of <strong>the</strong> o<strong>the</strong>r agent? De Jaegher <strong>and</strong> Di Paolo rightly insist that:<br />
if <strong>the</strong> autonomy of one of <strong>the</strong> interactors were destroyed, <strong>the</strong> process would<br />
reduce to <strong>the</strong> cognitive engagement of <strong>the</strong> remaining agent with his non-social<br />
world. The „o<strong>the</strong>r‟ would simply become a tool, an object, or a problem for his<br />
individual cognition (such a situation would epitomise what we have diagnosed<br />
traditional perspectives on social cognition as suffering from: namely, <strong>the</strong> lack<br />
of a properly social level). (De Jaegher & Di Paolo 2007, p. 492)<br />
It is certainly <strong>the</strong> case that <strong>the</strong> o<strong>the</strong>r agent must remain autonomous for an interaction to<br />
be characterized as social. The question that remains, however, is whe<strong>the</strong>r a cognitive<br />
interaction between two or more dynamic agents in a multi-agent system is also a<br />
sufficient criterion. Can such a cognitive inter-agent interaction capture what is specific<br />
about sociality in <strong>the</strong> sense presumably intended by De Jaegher <strong>and</strong> Di Paolo? What is<br />
needed is a notion of <strong>the</strong> social that not only excludes interactions that destroy <strong>the</strong><br />
autonomy of <strong>the</strong> o<strong>the</strong>r, but also exclude those situations in which <strong>the</strong> o<strong>the</strong>r is simply<br />
encountered as a mere tool, object or problem to be solved by an individual‟s cognitive<br />
ability (if <strong>the</strong> o<strong>the</strong>r appears as something to be encountered at all).<br />
Unfortunately, <strong>the</strong> notion of an interaction in a multi-agent system of dynamic agents is<br />
not specific enough. There are situations in which dynamic agents can interact (such<br />
that all of De Jaegher <strong>and</strong> Di Paolo‟s requirements are fulfilled), but in which <strong>the</strong> o<strong>the</strong>r<br />
agent is simply treated as part of <strong>the</strong> non-social environment. A famous example is <strong>the</strong><br />
cognitive domain of an autistic person who is embedded within <strong>the</strong> social world of<br />
o<strong>the</strong>rs, but who does not perceive this sociality as such.<br />
An illustration of this possibility is provided by <strong>the</strong> psychological experiment by<br />
Auvray <strong>and</strong> colleagues (briefly described above, also cf. Chapter 6, p. 90), whereby <strong>the</strong><br />
participants constitute an autonomous interaction process, but without actually being<br />
able to meaningfully differentiate between <strong>the</strong> socially contingent <strong>and</strong> non-contingent<br />
situations. What this example demonstrates is that it is not sufficient for two cognitive<br />
agents to give rise to an autonomous interaction process if <strong>the</strong>y are to break out of <strong>the</strong>ir<br />
67 | P a g e
individual cognitive domains. While <strong>the</strong> behavior of <strong>the</strong> participants is, unbeknownst to<br />
<strong>the</strong>m, guided by <strong>the</strong> dynamics of <strong>the</strong> interaction process to an appropriate solution to <strong>the</strong><br />
given task, <strong>the</strong>ir sense-making ability remains qualitatively unaffected with respect to its<br />
solitary point of reference. It is impossible for individuals to distinguish between <strong>the</strong><br />
movements of <strong>the</strong> o<strong>the</strong>r participant <strong>and</strong> its copy, even though <strong>the</strong>y are „collectively‟<br />
solving <strong>the</strong> task due to <strong>the</strong> dynamics of <strong>the</strong> multi-agent system.<br />
In sum, multi-agent interaction between dynamic agents is a necessary but not sufficient<br />
condition for <strong>the</strong> constitution of social significance. Since we have argued that it is<br />
regulation of structural coupling which is constitutive of <strong>the</strong> qualitative aspect of sensemaking<br />
activity (cf. Chapter 3, p. 35), we need to take a closer look at this regulative<br />
aspect. But what kind of regulation is characteristic of a social interaction such that it<br />
attains meaning as a social event? What is needed is way of defining <strong>the</strong> operational<br />
basis of participatory sense-making:<br />
If regulation of social coupling takes place through coordination of movements,<br />
<strong>and</strong> if movements – including utterances – are <strong>the</strong> tools of sense-making, <strong>the</strong>n<br />
our proposal is: social agents can coordinate <strong>the</strong>ir sense-making in social<br />
encounters. […] This is what we call participatory sense-making: <strong>the</strong><br />
coordination of intentional activity in interaction, whereby individual sensemaking<br />
processes are affected <strong>and</strong> new domains of social sense-making can be<br />
generated that were not available to each individual on her own. (De Jaegher &<br />
Di Paolo 2007, p. 497)<br />
This „regulation of social coupling‟ is precisely what has to be made explicit in De<br />
Jaegher <strong>and</strong> Di Paolo‟s definition of „social interaction‟ if it is to do all <strong>the</strong> intended<br />
work. For if we cannot find a qualitative difference in terms of <strong>the</strong> regulation of<br />
coupling <strong>the</strong>n <strong>the</strong> constitution of a novel social domain of sense-making will remain<br />
mysterious. What specific regulatory process is involved in <strong>the</strong> coordination of<br />
intentional activity during social interaction?<br />
Let us proceed by means of a concrete example. De Jaegher <strong>and</strong> Di Paolo (2008),<br />
drawing on Fogel (1993), provide an insightful description of a paradigmatic social act:<br />
68 | P a g e
<strong>the</strong> act of giving. Fogel describes a filmed session between a 1-year-old baby <strong>and</strong> his<br />
mo<strong>the</strong>r, in which <strong>the</strong> infant extends his arms with an object, <strong>and</strong> keeps <strong>the</strong>m relatively<br />
stationary only to gently release <strong>the</strong> object as <strong>the</strong> mo<strong>the</strong>r‟s h<strong>and</strong> takes hold of it. From<br />
this description it is already evident that giving has an essentially different structure of<br />
behavior that distinguishes it from merely individual cognitive engagements. In essence,<br />
in order for <strong>the</strong> act to be completed successfully, it requires acceptance from <strong>the</strong> o<strong>the</strong>r<br />
agent. In a more recent paper Di Paolo comments:<br />
Assuming for a moment that <strong>the</strong> infant is <strong>the</strong> initiator of <strong>the</strong> act, we realise that<br />
he must create an opening by his action that may only be completed by <strong>the</strong><br />
action of <strong>the</strong> mo<strong>the</strong>r. The giving involves more than orientation of <strong>the</strong> mo<strong>the</strong>r‟s<br />
sense-making; it involves a request for her not only to orient towards <strong>the</strong> new<br />
situation, but also to create an activity that will bring <strong>the</strong> act to completion. In<br />
o<strong>the</strong>r words: to take up <strong>the</strong> invitation for an intention to be shared. […] an<br />
invitation to participate is experienced as a request to create an appropriate<br />
closure of a sense-making activity that was not originally hers. To accept this<br />
request is to produce <strong>the</strong> „o<strong>the</strong>r half of <strong>the</strong> act‟ bringing it to a successful<br />
completion. (Di Paolo, in press; emphasis added)<br />
On <strong>the</strong> basis of <strong>the</strong> act of giving we can now make explicit what was already implicit in<br />
<strong>the</strong> enactive approach to social interaction that was first proposed by De Jaegher <strong>and</strong> Di<br />
Paolo (2007). The regulation involved in social interaction is indeed of a special kind:<br />
one cognitive agent‟s regulation of interaction creates an opening for an act that can<br />
only be realized through <strong>the</strong> complementary regulation of interaction by ano<strong>the</strong>r. In<br />
o<strong>the</strong>r words, social interaction is a manifestation of co-regulation. More precisely, we<br />
can provide <strong>the</strong> following definition:<br />
Social interaction is <strong>the</strong> co-regulated coupling between at least two dynamic<br />
agents, where <strong>the</strong> regulation is aimed at aspects of <strong>the</strong> coupling itself so that:<br />
1. It constitutes an emergent autonomous organization in <strong>the</strong> domain of internal<br />
<strong>and</strong> relational dynamics, without destroying in <strong>the</strong> process <strong>the</strong> autonomy of<br />
<strong>the</strong> agents involved (though <strong>the</strong> latter‟s scope can be augmented or reduced),<br />
<strong>and</strong><br />
69 | P a g e
2. An agent‟s regulation of coupling can only be completed by <strong>the</strong> coordinated<br />
regulation of at least one o<strong>the</strong>r agent.<br />
With respect to <strong>the</strong> goal of fur<strong>the</strong>r developing <strong>the</strong> notion of sense-making, two aspects<br />
of this definition are particularly noteworthy: (i) since <strong>the</strong> interacting agents are<br />
autonomous systems <strong>and</strong> <strong>the</strong>y adaptively regulate this interaction, it follows that <strong>the</strong>y<br />
engage with each o<strong>the</strong>r in terms of sense-making, <strong>and</strong> (ii) since <strong>the</strong> regulation of <strong>the</strong><br />
interaction by one agent changes not only its own coupling but also that of <strong>the</strong> o<strong>the</strong>r<br />
agent, it follows that <strong>the</strong> agents can enable <strong>and</strong> constrain each o<strong>the</strong>r‟s sense-making. But<br />
even though sense-making can be modulated by <strong>the</strong> interaction process in this manner,<br />
it essentially remains an individual affair if all we are dealing with is a multi-agent<br />
system. It only takes on a social significance when it is <strong>the</strong> result of co-regulation in <strong>the</strong><br />
strict sense, such that it could not be achieved by individual regulation alone. It is <strong>the</strong><br />
addition of <strong>the</strong> second requirement, i.e. of necessary co-regulation, that gives meaning<br />
to <strong>the</strong> notion of participatory sense-making as such.<br />
It is worth emphasizing <strong>the</strong> basic idea of this proposal again: if agents mutually enable<br />
<strong>and</strong> constrain <strong>the</strong>ir sense-making activities in a multi-agent system, <strong>the</strong>y can certainly<br />
open up behavioral domains that would have o<strong>the</strong>rwise remained inaccessible to <strong>the</strong><br />
individual agents. This is nicely illustrated by <strong>the</strong> psychological experiment conducted<br />
by Auvray <strong>and</strong> colleagues (2009), where <strong>the</strong> relative stability <strong>and</strong> instability of <strong>the</strong><br />
interaction process causes <strong>the</strong> participants to succeed at a task that <strong>the</strong>y are individually<br />
incapable of solving. But <strong>the</strong> fact that participants are equally likely to „locate‟ <strong>the</strong> o<strong>the</strong>r<br />
during mutual interaction as when interacting with <strong>the</strong> irresponsive mobile lure also<br />
shows that, while <strong>the</strong> interaction process has organized <strong>the</strong>ir behavior appropriately, it<br />
has not affected <strong>the</strong>ir sense-making activity. To <strong>the</strong> participants <strong>the</strong>re is no meaningful<br />
difference between <strong>the</strong> two situations. For that to happen <strong>the</strong> task must be changed such<br />
that an intended activity of one participant can only become realized by <strong>the</strong> coordinated<br />
activity of <strong>the</strong> o<strong>the</strong>r. Only <strong>the</strong>n can we properly speak of participatory sense-making <strong>and</strong><br />
expect a qualitative difference in experience.<br />
Of course, <strong>the</strong>se two scenarios are not mutually exclusive. Participatory situations<br />
necessarily emerge out of <strong>the</strong> interactions of a multi-agent system, <strong>and</strong> may in turn<br />
70 | P a g e
influence <strong>the</strong> structure of that system so as to lead to fur<strong>the</strong>r openings for co-regulated<br />
participation. Di Paolo, for example, suggests that when we remove <strong>the</strong> assumption that<br />
<strong>the</strong> infant intentionally originated <strong>the</strong> act of giving we open up <strong>the</strong> possibility even<br />
richer degrees of interaction: “A certain movement extending <strong>the</strong> object in <strong>the</strong> direction<br />
of <strong>the</strong> mo<strong>the</strong>r, without yet intending to give it, may now be opportunistically invested<br />
with a novel meaning through joint sense-making. Latent intentions become crystallised<br />
through <strong>the</strong> joint activity so that not only <strong>the</strong> completion of <strong>the</strong> act is achieved toge<strong>the</strong>r,<br />
but also its initiation” (Di Paolo, in press). In o<strong>the</strong>r words, interaction in a multi-agent<br />
system can not only extend <strong>the</strong> relational domains of <strong>the</strong> individual agents (i.e. <strong>the</strong>ir<br />
cognitive <strong>and</strong> behavioral capacities), but also lead to a novel way of participatory sensemaking<br />
(via co-regulation of activities) that inaugurates a social domain specific to <strong>the</strong>ir<br />
history of interactions.<br />
4.3 Cultural interaction<br />
The act of giving, as a paradigmatic social act, is widespread throughout <strong>the</strong> animal<br />
kingdom, most often in <strong>the</strong> context of parenting (e.g. giving food) or courtship (e.g.<br />
making more or less arbitrary offerings). As such, it is one of <strong>the</strong> most fundamental<br />
social acts on <strong>the</strong> basis of which o<strong>the</strong>r forms of sociality can develop. The act itself does<br />
not presuppose much <strong>and</strong>, following De Jaegher <strong>and</strong> Di Paolo‟s interpretation of <strong>the</strong><br />
infant giving an object to its mo<strong>the</strong>r, it is possible that none of <strong>the</strong> interactors<br />
intentionally originated <strong>the</strong> act. An arbitrary exchange can be subsequently invested<br />
with social significance when its joint completion changes <strong>the</strong> very meaning of <strong>the</strong><br />
relationship to that of „giver‟ <strong>and</strong> „receiver‟.<br />
However, do <strong>the</strong> abstract categories of „giver‟ <strong>and</strong> „receiver‟ actually have any meaning<br />
in <strong>the</strong> animal kingdom apart from <strong>the</strong>ir use by human beings? Typically, we would<br />
expect that <strong>the</strong> roles are much more concretely situated in non-human cases of social<br />
interaction, e.g. as „feeder‟ <strong>and</strong> „fed‟ or „courter‟ <strong>and</strong> „courted‟. The example of <strong>the</strong><br />
object exchange between <strong>the</strong> infant <strong>and</strong> its mo<strong>the</strong>r thus points to <strong>the</strong> need for some<br />
additional clarification. Where do <strong>the</strong> norms which guide <strong>the</strong> mo<strong>the</strong>r‟s response to <strong>the</strong><br />
infant‟s behavior come from? And how do <strong>the</strong>y provide a measure for <strong>the</strong> successful<br />
completion of <strong>the</strong> act as a whole?<br />
71 | P a g e
It is here that <strong>the</strong> socio-cultural background, in which <strong>the</strong> interactors <strong>and</strong> <strong>the</strong> unfolding<br />
interaction process are embedded, comes into play (cf. Steiner & Stewart 2009). Indeed,<br />
<strong>the</strong> mo<strong>the</strong>r might be moved to accept <strong>the</strong> held object because that is „what one does‟<br />
when offered something by ano<strong>the</strong>r. From her perspective, treating <strong>the</strong> gesture as <strong>the</strong><br />
infant‟s attempt to „give‟ <strong>the</strong> object is a natural way of making sense of <strong>the</strong> situation,<br />
<strong>and</strong> this sense-making is implicitly achieved in terms of a pre-established social<br />
practice. Moreover, this meaning, once it has been actualized in <strong>the</strong> situation, is not lost<br />
on <strong>the</strong> infant, ei<strong>the</strong>r, who has now discovered a novel way of interacting with his<br />
mo<strong>the</strong>r. In o<strong>the</strong>r words, to characterize this example as a social interaction alone misses<br />
<strong>the</strong> fact that we are dealing with a process of enculturation.<br />
The appeal to a pre-existing order of shared practices indicates that an approach to<br />
social cognition which only focuses on <strong>the</strong> momentary constitution of norms in social<br />
interaction is not sufficient to capture <strong>the</strong> whole of sociality. In particular, it is missing<br />
what is specific about those social interactions that unfold within a cultural context. As<br />
Steiner <strong>and</strong> Stewart emphasize, <strong>the</strong> latter kind of social interactions also necessarily<br />
involve a form of „heteronomy‟, i.e. <strong>the</strong> abiding by a heritage of pre-established social<br />
structures. Indeed, <strong>the</strong> claim that <strong>the</strong>re are heteronomous cultural values that guide our<br />
behavior <strong>and</strong> underst<strong>and</strong>ing points to a more general phenomenon, since enculturation<br />
has similarly profound effects on our solitary behavior. A castaway like Robinson<br />
Crusoe does not immediately cease to behave like an Englishman when he finds himself<br />
socially isolated on a tropical isl<strong>and</strong>. Enculturation thus involves at least some form of<br />
internalization of heteronomy (cf. Vygotsky 1978).<br />
In terms of <strong>the</strong> social, Steiner <strong>and</strong> Stewart argue that only enculturated forms of<br />
interaction deserve to be called social interactions, in order to distance <strong>the</strong>m from <strong>the</strong><br />
kind of multi-agent interactions that are paradigmatic of De Jaegher <strong>and</strong> Di Paolo‟s<br />
approach. However, while we agree that <strong>the</strong> latter approach is too inclusive, which is<br />
why we have re-conceptualized it as merely a necessary condition for social interaction<br />
(i.e. in terms of multi-agent interaction), Steiner <strong>and</strong> Stewart‟s approach is also overly<br />
exclusive. They make sociality a specifically human phenomenon, <strong>the</strong>reby excluding<br />
everything from <strong>the</strong> so-called social insects to our closest primate relatives.<br />
72 | P a g e
In contrast to both of <strong>the</strong>se approaches, <strong>the</strong> definition of social interaction that we have<br />
provided in <strong>the</strong> previous section takes up a middle ground. On <strong>the</strong> one h<strong>and</strong>, it excludes<br />
cognitive interactions that merely contingently happen to involve ano<strong>the</strong>r agent, but on<br />
<strong>the</strong> o<strong>the</strong>r h<strong>and</strong> it includes co-regulated interactions that are not already guided by preestablished<br />
cultural norms. Of course, this is not to deny that Steiner <strong>and</strong> Stewart are<br />
correct in insisting that <strong>the</strong>re is something special about many human forms of sociality,<br />
including <strong>the</strong>ir heteronomous character, but this specificity is perhaps better captured by<br />
<strong>the</strong> notion of culture ra<strong>the</strong>r than by sociality as such.<br />
While <strong>the</strong> enactive paradigm has acknowledged <strong>the</strong> constitutive role of <strong>the</strong> cultural<br />
context for <strong>life</strong> <strong>and</strong> <strong>mind</strong> (cf. Thompson 2007; Steiner & Stewart 2009; Di Paolo 2008;<br />
in press), so far <strong>the</strong>re has been no attempt to provide an operational definition of those<br />
social interactions whose unfolding is partially determined by a pre-existing sociocultural<br />
background. Never<strong>the</strong>less, <strong>the</strong> target phenomenon is starting to be clarified <strong>and</strong><br />
<strong>the</strong> bottom-up approach of <strong>the</strong> enactive paradigm is systematically developing an<br />
explanation in a step-by-step manner. In this chapter we have taken <strong>the</strong> important step<br />
of clarifying social interactions as being a special kind of multi-agent interaction where<br />
meaning is co-created by <strong>the</strong> joint action of <strong>the</strong> interactors.<br />
An important problem that remains is to explain how such social interaction is shaped<br />
by „external‟ cultural values. In response we can note that one way to begin to<br />
underst<strong>and</strong> <strong>the</strong> heteronomy of cultural structures, <strong>and</strong> <strong>the</strong> one pursued in this <strong>the</strong>sis, is to<br />
first consider <strong>the</strong> autonomy of a social interaction process in more detail. After all, this<br />
autonomy is, when viewed from <strong>the</strong> perspective of <strong>the</strong> interacting agents, also a form of<br />
heteronomy that has its own intrinsic teleology, <strong>and</strong> which can enable <strong>and</strong> constrain <strong>the</strong><br />
behavior of <strong>the</strong> individual agents. Of course, future work will need to determine more<br />
precisely what is special about <strong>the</strong> heteronomy of culture. In particular, how is it<br />
possible that behavior implicitly adheres to cultural norms even when o<strong>the</strong>rs are not<br />
immediately present? But even here we should be able to approach this problem from<br />
<strong>the</strong> perspective of social interaction, especially social learning. If we want to know how<br />
culture can shape our behavior even outside of an immediate social context, <strong>the</strong>n we<br />
first need to better underst<strong>and</strong> how an agent involved in a social interaction, faced with<br />
73 | P a g e
<strong>the</strong> heteronomy of ano<strong>the</strong>r agent <strong>and</strong> <strong>the</strong> heteronomy of <strong>the</strong> interaction process itself,<br />
can undergo a change in behavior that we would call learning.<br />
A final question to consider is whe<strong>the</strong>r <strong>the</strong> constitutive impact of cultural values is not a<br />
problem for <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis. Do we not have to provide a biological<br />
foundation for <strong>the</strong>se values? Yes <strong>and</strong> no. Yes, in <strong>the</strong> sense that <strong>the</strong>se values can only<br />
exist for certain kinds of sense-making agents, <strong>and</strong> <strong>the</strong>se agents are biological in that<br />
<strong>the</strong>y are alive (autonomous <strong>and</strong> adaptive). No, in <strong>the</strong> sense that this is not a reduction of<br />
cultural values to <strong>the</strong>ir biological conditions of possibility; <strong>the</strong> socio-cultural domain<br />
retains its own autonomy. As such, <strong>the</strong> emergence of <strong>the</strong> heteronomy of culture is <strong>the</strong><br />
appearance of ano<strong>the</strong>r dis<strong>continuity</strong> in <strong>the</strong> system of discontinuities which constitutes<br />
<strong>life</strong>, <strong>mind</strong>, <strong>and</strong> sociality. More specifically, <strong>the</strong> <strong>continuity</strong> <strong>the</strong>sis is preserved because<br />
<strong>the</strong> heteronomy of culture turns out to be mutually interdependent with <strong>the</strong> heteronomy<br />
of sociality, <strong>and</strong> <strong>the</strong> same conceptual framework of autonomy, which forms <strong>the</strong> very<br />
foundation of <strong>the</strong> enactive paradigm, is applicable to both.<br />
It is already clear that, like <strong>the</strong> previous transitions along <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong>, a<br />
cognitive agent‟s entrance into a cultural domain is both enabling <strong>and</strong> constraining. It is<br />
constraining because taking part in shared practices requires <strong>the</strong> alignment of an<br />
individual‟s autonomy with a pre-established. But despite this constraining, or ra<strong>the</strong>r –<br />
because of it, <strong>the</strong>re is also an expansion of possibilities. A good example of this is play,<br />
<strong>the</strong> freedom of which lies in a players‟ capability to create new meaningful constraints<br />
by which it can steer its sense-making activity <strong>and</strong> set new laws for itself <strong>and</strong> o<strong>the</strong>rs to<br />
follow (Di Paolo, et al. in press). Moreover, by inaugurating a historical trace of shared<br />
individual <strong>and</strong> social practices that can go beyond an individual‟s <strong>life</strong>time, cultural<br />
interaction provides <strong>the</strong> foundation for cumulatively building on previous more or less<br />
viable ways of living.<br />
The subsequent chapters of this <strong>the</strong>sis will investigate <strong>the</strong> dynamics of social interaction<br />
more generally, but we will return to some speculations about <strong>the</strong> possible mechanisms<br />
of cumulative cultural development in Chapter 13.<br />
74 | P a g e
4.4 Summary<br />
In this chapter we have traced <strong>the</strong> conceptual framework of <strong>the</strong> enactive paradigm from<br />
autonomy to culture, as summarized in Figure 4-4.<br />
Autonomy<br />
Adaptivity<br />
Agency I<br />
Agency II<br />
Operational specificity<br />
<strong>Sociality</strong><br />
Culture<br />
Qualitative change<br />
Social cognition<br />
Cognition<br />
Behavior<br />
Sense-making<br />
Intrinsic teleology<br />
Figure 4-4. This schematic summarizes <strong>the</strong> relationships between core concepts of <strong>the</strong> enactive paradigm<br />
as we have developed <strong>the</strong>m in this chapter. Any inner layer necessarily depends on all of <strong>the</strong> outer layers.<br />
Thus, for each phenomenon specified at <strong>the</strong> bottom of a layer (e.g. „sense-making‟), <strong>the</strong> operational<br />
requirements specified at <strong>the</strong> top of that layer, including those of all previous outer layers, toge<strong>the</strong>r form<br />
its necessary <strong>and</strong> sufficient conditions (e.g. „autonomy‟ <strong>and</strong> „adaptivity‟). „Agency I‟ refers to an<br />
autonomous system that achieves adaptation not only through internal re-organization (i.e. adaptivity,<br />
more generally), but also by regulation of its structural coupling (adaptive agency). „Agency II‟ denotes a<br />
form of adaptive agency, whereby <strong>the</strong> norms of <strong>the</strong> regulation of structural coupling are underdetermined<br />
by metabolic criteria alone (dynamic agency). As <strong>the</strong> operational specificity increases with each inner<br />
layer, we can attribute an expansion of qualitative existence to <strong>the</strong> system. The central layer, culture, is<br />
still in need of fur<strong>the</strong>r clarification in both operational <strong>and</strong> phenomenological terms.<br />
This figure shows that <strong>the</strong> enactive paradigm indeed promotes a form of <strong>life</strong>-<strong>mind</strong><br />
<strong>continuity</strong> that is not exhausted by unification into one conceptual framework. In o<strong>the</strong>r<br />
words, all of <strong>the</strong> latter, more specialized phenomena (inner layers) depend necessarily<br />
75 | P a g e
(<strong>and</strong> not just historically, i.e. evolutionarily <strong>and</strong> developmentally) on <strong>the</strong> existence of all<br />
of <strong>the</strong> former, more inclusive phenomena (outer layers). Note, however, that even<br />
though every new domain emerges on <strong>the</strong> basis of activity in <strong>the</strong> preceding domains, it<br />
cannot be reduced to that enabling activity. This operational asymmetry between<br />
successive domains is what <strong>the</strong> recurring concepts of partial decoupling, emergence,<br />
<strong>and</strong> autonomy provide. It is also what guarantees that we are actually dealing with a<br />
non-reductive <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong>, ra<strong>the</strong>r than a progression of heuristics that could be<br />
collapsed into a purely metabolic level on <strong>the</strong> basis of a more advanced science.<br />
Of course, we should not misunderst<strong>and</strong> this operational asymmetry as prescribing a<br />
one-sided interaction between <strong>the</strong> different phenomenal domains. On <strong>the</strong> contrary, once<br />
<strong>the</strong> different domains of activity have been established for an agent, <strong>the</strong>ir relationship is<br />
not one of hierarchical dependence, but ra<strong>the</strong>r of multiple interdependence. For any<br />
agent it is possible (<strong>and</strong> likely) that its activities in <strong>the</strong> different domains all mutually<br />
constrain <strong>and</strong> enable each o<strong>the</strong>r in various non-trivial ways. Thus, even cultural norms<br />
can be re-inscribed back into <strong>the</strong> normativity operative on <strong>the</strong> metabolic level (Di Paolo,<br />
in press). For example, I might take up drinking due to <strong>the</strong> kind of socio-cultural<br />
environment of which I am a part, but that appropriated activity might itself become<br />
sustained as a self-constituting habit, <strong>and</strong> that habit can even begin to re-organize my<br />
metabolism in such a way that it reinforces <strong>the</strong> frequency of my drinking behavior,<br />
which <strong>the</strong>reby starts to shape <strong>the</strong> socio-cultural environment faced by o<strong>the</strong>rs around me,<br />
perhaps making <strong>the</strong>m more inclined to take up drinking as well. Accordingly, we can<br />
identify multiple, interdependent, mutually enabling <strong>and</strong> constraining autonomous<br />
systems within <strong>and</strong> across different phenomenal domains. In general, working out how<br />
<strong>the</strong>se multiple interdependencies precisely operate, <strong>and</strong> how <strong>the</strong>y combine to bring forth<br />
coherent forms of agency is one of <strong>the</strong> most important research problems for enactive<br />
cognitive science. In particular, it remains to be explained how it is possible for us to<br />
reflectively live out a unified existence, though it is likely that this has to do with our<br />
interactions in a linguistic domain (Di Paolo 2009, p 19) <strong>and</strong> processes of self-o<strong>the</strong>r codetermination<br />
more generally (Thompson 2001).<br />
76 | P a g e
5 Beyond methodological individualism<br />
A fundamental assumption of mainstream cognitive science is that <strong>the</strong> individual agent<br />
(whatever that may mean in <strong>the</strong> absence of a mainstream definition of agency) is <strong>the</strong><br />
correct unit of analysis for underst<strong>and</strong>ing <strong>life</strong>, <strong>mind</strong>, cognition, <strong>and</strong> behavior, as well as<br />
all social phenomena. This approach has been termed “methodological individualism”<br />
(cf. Boden 2006b), after <strong>the</strong> doctrine in social science that was introduced by Max<br />
Weber in <strong>the</strong> beginning of <strong>the</strong> 20 th century, <strong>and</strong> continued by Friedrich von Hayek <strong>and</strong><br />
Karl Popper among o<strong>the</strong>rs. The history of this doctrine in social science is complex; it<br />
became embroiled in highly politicized debates, largely because it was often invoked as<br />
a way of discrediting historical materialism (Heath 2009). In scientific terms <strong>the</strong> central<br />
claim of methodological individualism is that social phenomena must be explained by<br />
showing how <strong>the</strong>y result from individual actions, which in turn must be explained<br />
through reference to <strong>the</strong> intentional states that motivate <strong>the</strong> individual actors.<br />
More recently, <strong>the</strong> validity of this widespread assumption in cognitive science is starting<br />
to be questioned on <strong>the</strong> basis of research from a variety of its sub-disciplines. The idea<br />
that cognition is at least partly constituted by social interactions <strong>and</strong> cultural context has<br />
received support from cognitive anthropology (Hutchins 1995), developmental <strong>and</strong><br />
social psychology (cf. <strong>Tom</strong>asello 1999; Lindblom & Ziemke 2003), social studies<br />
(Pentl<strong>and</strong> 2007), primatology (Savage-Rumbaugh, et al. 2005; <strong>Tom</strong>asello 2000),<br />
interaction studies (Auvray, et al. 2009; Di Paolo, et al. 2008), as well as philosophical<br />
<strong>and</strong> phenomenological considerations (e.g. Zahavi 2001). In terms of underst<strong>and</strong>ing<br />
social cognition this shift is expressed by positing embodied interaction as <strong>the</strong> primary<br />
mechanism of our underst<strong>and</strong>ing of o<strong>the</strong>r <strong>mind</strong>s, ra<strong>the</strong>r than <strong>the</strong>oretical inference or<br />
empathic simulation (cf. Gallagher 2001).<br />
One formidable challenge that needs to be addressed by <strong>the</strong> critics of methodological<br />
individualism is to build a framework that enables <strong>the</strong>m to specify precisely <strong>and</strong> in a<br />
non-mysterious manner how it is possible for social phenomena to play a constitutive<br />
role in <strong>the</strong> unfolding of individual behavior (De Jaegher 2009). In Chapters 2 to 4 we<br />
have provided <strong>the</strong> foundation for this task by introducing <strong>and</strong> developing <strong>the</strong> conceptual<br />
framework of <strong>the</strong> enactive approach to social cognition. Interestingly, <strong>the</strong> historical<br />
77 | P a g e
oots of this approach were never that far away from methodological individualism (Di<br />
Paolo 2008). The influential traditions of radical constructivism (e.g. von Glasersfeld<br />
1984) <strong>and</strong> especially second-order cybernetics (e.g. von Foerster 1973) were certainly<br />
concerned with <strong>the</strong> existence of o<strong>the</strong>rs, but largely as a one-sided response to <strong>the</strong> specter<br />
of solipsism that was haunting <strong>the</strong>ir subject-centered worldview. Maturana <strong>and</strong> Varela‟s<br />
(1980) biology of cognition continued <strong>the</strong> work of <strong>the</strong>se traditions (cf. Varela 1996a),<br />
but offered an essential improvement. They insisted on a clear logical accounting in<br />
biology that separated constitutive (individual) <strong>and</strong> relational (interactive) phenomena<br />
into two non-intersecting domains, whereby <strong>the</strong> relational phenomena may constrain <strong>the</strong><br />
constitutive processes <strong>and</strong> vice versa. This opened <strong>the</strong> door to a fuller appreciation of<br />
<strong>the</strong> role of interaction in a social, linguistic, <strong>and</strong> cultural context for <strong>the</strong> development of<br />
higher cognitive functions (e.g. Maturana, et al. 1995), even leading to <strong>the</strong> radical<br />
conclusion that “we are constituted in language in a continuous becoming that we bring<br />
forth with o<strong>the</strong>rs” (Maturana & Varela 1987, pp. 234-235). Never<strong>the</strong>less, this strong<br />
idea of self-o<strong>the</strong>r co-determination has remained marginalized in <strong>the</strong> biology of<br />
cognition, in particular because its doctrine of non-intersecting domains simply leaves it<br />
unexplained precisely how relational phenomena constrain processes in <strong>the</strong> constitutive<br />
domain. It thus remains unclear how <strong>the</strong> social is afforded any proper constitutive role.<br />
A detailed comparative analysis of how <strong>the</strong> enactive approach relates to <strong>the</strong>se traditions<br />
is desirable, but unfortunately beyond <strong>the</strong> scope of this <strong>the</strong>sis. In its early formulations it<br />
was certainly still afflicted by a lingering methodological individualism, though in<br />
recent work this has become less of a problem 19 . As should have become clear from <strong>the</strong><br />
preceding chapters, one crucial difference is that <strong>the</strong> notion of identity conservation has<br />
been replaced by a focus on precariousness <strong>and</strong> normativity. The profound implication<br />
of this simple change in focus is that a richer grasp of <strong>the</strong> interrelationship between <strong>the</strong><br />
constitutive <strong>and</strong> relational domains is now conceivable, to <strong>the</strong> extent that it becomes<br />
possible to think about how socio-cultural values can transform even basic metabolic<br />
processes (Di Paolo, in press). Enactive cognitive science has thus traded <strong>the</strong> framework<br />
19 The related tradition of <strong>the</strong> enactive or, more precisely, sensory-motor approach to perception (e.g. Noë<br />
2004; O‟Regan & Noë 2001), which took inspiration from <strong>the</strong> early work by Varela <strong>and</strong> colleagues, has<br />
remained committed to methodological individualism. We will return to this point in Chapter 12.<br />
78 | P a g e
of methodological individualism for an approach that is more akin to <strong>the</strong> dialectical <strong>and</strong><br />
historical materialism of Vygotsky‟s psychology, as first proposed by Marx <strong>and</strong> Engels,<br />
though stripped of its metaphysical pretense to universality <strong>and</strong> instead embedded in<br />
systems thinking <strong>and</strong> a closed-loop epistemology.<br />
However, it is one thing to say that social interaction plays a constitutive role, <strong>and</strong><br />
ano<strong>the</strong>r to say exactly how it plays that role. The focus of Chapters 6 to 10 is <strong>the</strong>refore<br />
to demonstrate more concretely <strong>the</strong> constitutive interplay between <strong>the</strong> individual <strong>and</strong><br />
interactional levels. First, <strong>the</strong> problems faced by <strong>the</strong> doctrine of methodological<br />
individualism in accounting for a range of psychological phenomena are highlighted,<br />
<strong>and</strong> a possible role of <strong>the</strong> interaction process is indicated (Chapter 6). In order to get a<br />
better underst<strong>and</strong>ing of how it is possible for <strong>the</strong> dynamics of <strong>the</strong> interaction process to<br />
be constitutive of an individual‟s behavior we introduce evolutionary robotics as a<br />
methodology to generate complete models of minimal complexity (Chapter 7). This is<br />
followed by a range of novel modeling experiments which show concretely <strong>and</strong> in<br />
ma<strong>the</strong>matically analyzable terms how individual <strong>and</strong> interaction levels are interrelated<br />
in multi-agent systems (Chapters 8 to 10).<br />
79 | P a g e
6 Studies in social psychology: A critical analysis<br />
The aim of this chapter is to show that <strong>the</strong> enactive approach to social cognition can be<br />
used to provide a fresh perspective on some important experiments in developmental<br />
<strong>and</strong> social psychology. Most contemporary studies of social cognition attempt to explain<br />
<strong>the</strong> widespread phenomenon of bodily coordination, for example facial imitation or<br />
gestural exchange, in terms of a combination of three factors pertaining to <strong>the</strong> individual<br />
interlocutors. These factors consist of two kinds of self-perception <strong>and</strong> one kind of<br />
o<strong>the</strong>r-perception: (i) visual self-perception, (ii) proprioceptive self-perception, <strong>and</strong> (iii)<br />
visual o<strong>the</strong>r-perception. As we will see, most traditional explanations have no problems<br />
accounting for bodily coordination as long as <strong>the</strong>y can appeal to ei<strong>the</strong>r or both kinds of<br />
self-perception <strong>and</strong> some form of o<strong>the</strong>r-perception. If this is not possible, for instance in<br />
some pathological cases, additional ad hoc neural systems are typically postulated to fill<br />
<strong>the</strong> explanatory gap. From <strong>the</strong> perspective of <strong>the</strong> enactive approach to social cognition,<br />
it appears that what is missing from <strong>the</strong>se traditional explanations in psychology is an<br />
appreciation of <strong>the</strong> role of <strong>the</strong> interaction process in organizing individual behavior.<br />
In order to motivate a consideration of <strong>the</strong> interaction process for explaining empirical<br />
data in developmental <strong>and</strong> social psychology we will proceed as follows. First, some of<br />
<strong>the</strong> main concepts used by traditional approaches, i.e. body image, body schema <strong>and</strong><br />
proprioception, need to be clarified. This is followed by a discussion of several case<br />
studies which map out empirically <strong>the</strong> various possibilities of <strong>the</strong> conceptual space<br />
afforded by <strong>the</strong> combinations of <strong>the</strong>se three concepts. On this basis an attempt is made<br />
to adopt a more progressive approach to social cognition in recent cognitive science,<br />
namely <strong>the</strong> integrative <strong>the</strong>ory of gesture, to make sense of this data. Never<strong>the</strong>less, some<br />
potential problems are identified. Finally, it is suggested that <strong>the</strong> enactive approach to<br />
social cognition, with its focus on <strong>the</strong> constitutive role of <strong>the</strong> interaction process, is a<br />
promising c<strong>and</strong>idate to resolve <strong>the</strong>se difficulties.<br />
Note that it is beyond <strong>the</strong> scope of this chapter to develop a full response to each of <strong>the</strong><br />
chosen case studies. However, <strong>the</strong> modeling experiments presented in <strong>the</strong> next chapters<br />
indicate one potential methodology of how to start going about this in a more systematic<br />
80 | P a g e
manner. They begin to develop <strong>the</strong> conceptual language of dynamics that would be<br />
needed for a more thorough review of <strong>the</strong>se psychological studies.<br />
6.1 Body image <strong>and</strong> body schema<br />
We will follow Gallagher <strong>and</strong> Cole (1995) in making a conceptual distinction between<br />
two aspects of embodiment. On <strong>the</strong> one h<strong>and</strong> <strong>the</strong>re is <strong>the</strong> body image (BI), which<br />
consists of a system of perceptions, attitudes, <strong>and</strong> beliefs pertaining to one‟s own body.<br />
On <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>, <strong>the</strong>re is <strong>the</strong> body schema (BS), namely a system of sensory-motor<br />
capacities that functions without conscious awareness or <strong>the</strong> necessity of perceptual<br />
monitoring. In o<strong>the</strong>r words, “<strong>the</strong> difference between body image <strong>and</strong> body schema is<br />
like <strong>the</strong> difference between a perception (or conscious monitoring) of movement <strong>and</strong> <strong>the</strong><br />
actual accomplishment of movement, respectively” (Gallagher 2005, p. 24). There is<br />
empirical evidence that this conceptual distinction does indeed pick out two different<br />
aspects of our embodiment, <strong>and</strong> some of this evidence will be presented as part of <strong>the</strong><br />
case studies.<br />
Ano<strong>the</strong>r important concept that we need to define more clearly is that of proprioception,<br />
which is often used to refer to somatic information about joint position, limb extension,<br />
as well as bodily position <strong>and</strong> body posture more generally. This notion of somatic<br />
information can be meant in <strong>the</strong> form of a pre-reflective pragmatic awareness which,<br />
while not taking <strong>the</strong> body as an object, still contributes a certain spatial structure to <strong>the</strong><br />
perceptual body image. But <strong>the</strong> notion can also be used to refer to a non-conscious<br />
process, whereby physiological stimuli activating peripheral proprioceptors, which in<br />
turn are registered at certain strategic sites in <strong>the</strong> brain, operate as part of <strong>the</strong> system that<br />
constitutes <strong>the</strong> body schema. Following Gallagher (2005, p. 46) we will refer to <strong>the</strong>se<br />
two different ways of conceptualizing proprioception in terms of proprioceptive<br />
awareness (PA) <strong>and</strong> proprioceptive information (PI), respectively.<br />
The distinction between body image <strong>and</strong> body schema is related to <strong>the</strong> important<br />
phenomenological distinction between <strong>the</strong> body-as-object <strong>and</strong> <strong>the</strong> body-as-subject (cf.<br />
Legr<strong>and</strong> 2006). In <strong>the</strong> former case our embodiment is reflectively experienced by means<br />
of sensory perception, <strong>and</strong> in <strong>the</strong> latter case it is experientially transparent as a means of<br />
81 | P a g e
eing-in-<strong>the</strong>-world. Note, however, that in contrast to <strong>the</strong> notion of body schema, which<br />
refers to an integrated system of physiological capacities <strong>and</strong> proprioceptive information<br />
that remains outside our awareness, <strong>the</strong> body-as-subject is a phenomenological notion.<br />
As such it refers to a structure of our lived experience, for example our pre-reflective<br />
proprioceptive awareness.<br />
With <strong>the</strong>se distinctions in place we can now return to <strong>the</strong> three factors traditionally used<br />
to explain <strong>the</strong> existence of bodily coordination. First, we have visual self-perception,<br />
which is an essential part of <strong>the</strong> body image we have of ourselves (Self-BI). And <strong>the</strong>n<br />
<strong>the</strong>re is proprioceptive self-perception, which is usually taken as an essential part of <strong>the</strong><br />
sub-personal mechanisms that are outside of our awareness (PI). And finally <strong>the</strong>re is<br />
visual o<strong>the</strong>r-perception, which forms an essential part of <strong>the</strong> body image that we have of<br />
<strong>the</strong> o<strong>the</strong>r interlocutor (O<strong>the</strong>r-BI). On <strong>the</strong> basis of <strong>the</strong>se three factors we can now<br />
perform a simple meta-analysis of <strong>the</strong> literature by focusing on case studies that map out<br />
<strong>the</strong> space of possibilities, as summarized in Table 6-1.<br />
Case: Self-BI: O<strong>the</strong>r-BI: PI: Example of bodily coordination:<br />
1 √ √ √ Non-pathological face-to-face interaction<br />
2 √ √ Facial imitation by human neonates<br />
3 √ √ Gesturing by a deafferented subject (I)<br />
4 √ √ Perceptual crossing in a virtual space<br />
5 √ Deafferented subject under a blind<br />
6 √ Gesturing by a deafferented subject (II)<br />
7 √ Body imitation in a virtual space (I)<br />
8 Body imitation in a virtual space (II)<br />
Table 6-1. A list of case studies (1-8) that spans all possible combinations of <strong>the</strong> three factors<br />
traditionally used to explain bodily coordination: a self-directed body image (Self-BI), an o<strong>the</strong>r-directed<br />
body image (O<strong>the</strong>r-BI), <strong>and</strong> proprioceptive information (PI). Each case study provides a brief description<br />
of an instance of bodily coordination <strong>and</strong> <strong>the</strong> particular factors which would be available in terms of <strong>the</strong><br />
traditional explanatory framework. Note that <strong>the</strong>re are some instances of coordination which appear to fall<br />
outside <strong>the</strong> scope of this traditional framework. Never<strong>the</strong>less, all of <strong>the</strong>se cases allow <strong>the</strong> possibility of<br />
responsive interaction between <strong>the</strong> participants, <strong>and</strong> thus retain a role for <strong>the</strong> interaction process.<br />
82 | P a g e
6.2 Case studies in social psychology<br />
We will now exp<strong>and</strong> on <strong>the</strong> information provided in Table 6-1 by providing a more<br />
detailed description of <strong>the</strong> example of bodily coordination for each case, as well as <strong>the</strong><br />
kind of explanations that have been offered to account for <strong>the</strong>m. The focus on bodily<br />
coordination is justified because it is one of <strong>the</strong> most basic forms of inter-individual<br />
interaction. In all of <strong>the</strong> cases <strong>the</strong> aim is to use <strong>the</strong> empirical evidence to develop an<br />
underst<strong>and</strong>ing of <strong>the</strong> necessary <strong>and</strong> sufficient conditions for <strong>the</strong> establishment of bodily<br />
coordination between cognitive agents. What will emerge out of this analysis is a more<br />
precise grasp of <strong>the</strong> explanatory gaps in <strong>the</strong> traditional framework of social psychology,<br />
<strong>and</strong> a sense of <strong>the</strong> kind of explanations that can potentially be offered by <strong>the</strong> enactive<br />
approach to social cognition.<br />
6.2.1 Non-pathological face-to-face interaction<br />
Let us begin this analysis of psychological studies of bodily coordination with a brief<br />
consideration of unencumbered, everyday social interaction between human beings. It is<br />
well known in social psychology that unconscious gestural imitation is prevalent during<br />
human interactions, <strong>and</strong> that <strong>the</strong> particular unfolding of this imitation even influences<br />
<strong>the</strong> meaning of <strong>the</strong> social encounter (e.g. LaFrance 1982). In this unconstrained case <strong>the</strong><br />
full explanatory framework of traditional social psychology is available to account for<br />
this phenomenon of bodily coordination, as represented schematically in Figure 6-1.<br />
This basic case covers a whole range of social phenomena, including what has been<br />
described as primary intersubjectivity (Trevar<strong>the</strong>n 1979), secondary intersubjectivity<br />
(Trevar<strong>the</strong>n & Hubley 1978), as well as joint attention <strong>and</strong> linguistic interaction<br />
(<strong>Tom</strong>asello 1988). Indeed, out of all <strong>the</strong> listed cases this is <strong>the</strong> one that allows one of <strong>the</strong><br />
richest forms of embodied interaction to emerge between <strong>the</strong> participants, i.e. <strong>the</strong> full<br />
range of human intersubjectivity. However, because explanations of <strong>the</strong>se forms of<br />
bodily coordination can appeal to any combination of <strong>the</strong> three factors, it is difficult to<br />
sort out which of <strong>the</strong>m are necessary <strong>and</strong>/or sufficient conditions.<br />
83 | P a g e
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Proprioception<br />
(PI)<br />
Motor Act<br />
Figure 6-1. A schematic of <strong>the</strong> three factors that are traditionally used to explain bodily coordination<br />
during social interaction: (i) <strong>the</strong> body image that we have of ourselves (Self-BI), which is largely formed<br />
through visual perception, (ii) <strong>the</strong> proprioceptive information (PI) of one‟s bodily posture <strong>and</strong> movement,<br />
<strong>and</strong> (iii) <strong>the</strong> visual perception of <strong>the</strong> o<strong>the</strong>r‟s body (O<strong>the</strong>r-BI).<br />
In order to better determine whe<strong>the</strong>r any combination of <strong>the</strong> three traditional factors<br />
depicted in Figure 6-1 is necessary <strong>and</strong>/or sufficient to explain <strong>the</strong> phenomenon of<br />
bodily coordination in general we can consider a series of more restrictive case studies<br />
which systematically eliminate <strong>the</strong>ir potential influence.<br />
6.2.2 Facial imitation by human neonates<br />
It is well known in developmental psychology that human infants are good at imitating a<br />
wide variety of gestures performed by adult experimenters (e.g. Meltzoff & Moore<br />
1977; 1989). Moreover, it has been shown that even newborn infants less than an hour<br />
old can successfully imitate facial gestures such as mouth-opening <strong>and</strong> tongue<br />
protrusion (Meltzoff & Moore 1983). In general, <strong>the</strong> range of imitative capacities<br />
exhibited by young infants is too extensive <strong>and</strong> specific to be explained in terms of predetermined<br />
innate sensory-motor structures (BS) alone. Indeed, <strong>the</strong> fact that infants<br />
retain this imitative capacity even when <strong>the</strong> experimenters introduce considerable delays<br />
between <strong>the</strong> presentation of a stimulus <strong>and</strong> <strong>the</strong> infant‟s ability to respond implies that<br />
early imitation is not entirely stimulus bound, directly triggered, or reflexive (cf.<br />
Meltzoff & Moore 1997, p. 182).<br />
84 | P a g e
In <strong>the</strong> case of adult imitation it is possible to provide an explanation in <strong>the</strong> traditional<br />
framework that is based on a comparison between self-BI <strong>and</strong> o<strong>the</strong>r-BI. But how can we<br />
explain <strong>the</strong> infants‟ ability to imitate facial gestures given that <strong>the</strong>y have not yet had <strong>the</strong><br />
chance to construct a body image based on visual perception of <strong>the</strong>ir face? For <strong>the</strong>orists<br />
who hold that perception is unorganized in early infancy even this very phenomenon<br />
itself appears to be impossible:<br />
Thus since <strong>the</strong> child cannot see his own face, <strong>the</strong>re will be no imitation of<br />
movements of <strong>the</strong> face at this stage. […] For imitation of such movements to be<br />
possible, <strong>the</strong>re must be co-ordination of visual schemas with tactilo-kines<strong>the</strong>tic<br />
schemas. (Piaget 1962, p. 45; quoted by Gallagher 2005, p. 68)<br />
However, that such imitation by even very young infants is indeed possible has now<br />
been conclusively demonstrated. How are we to explain this phenomenon? Meltzoff <strong>and</strong><br />
Moore propose that early facial imitation is based on „active intermodal mapping‟<br />
(AIM), whereby <strong>the</strong> target matching process is captured by a proprioceptive (PI)<br />
feedback loop. In essence, <strong>the</strong> experiments on neonate imitation demonstrate, contrary<br />
to <strong>the</strong> traditional position of <strong>the</strong>orists such as Piaget (1962), Merleau-Ponty (1945/1962)<br />
<strong>and</strong> o<strong>the</strong>rs, that a functioning body schema is present at least from birth onwards, if not<br />
even earlier. The AIM model posits that a PI feedback loop from <strong>the</strong> infant‟s motor acts<br />
enables it to perform <strong>the</strong> appropriate gesture even without visual awareness of its own<br />
face. The equivalence between <strong>the</strong> target o<strong>the</strong>r-BI <strong>and</strong> <strong>the</strong> infant‟s PI is <strong>the</strong>refore<br />
determined by means of a supra- or intermodal matching process (Meltzoff & Moore<br />
1997). A simple schematic of <strong>the</strong> AIM hypo<strong>the</strong>sis is shown in Figure 6-2.<br />
85 | P a g e
Visual Perception of Target<br />
Adult Facial Acts<br />
Supramodal<br />
of<br />
Representation<br />
Acts<br />
Equivalence<br />
Detector<br />
Proprioceptive<br />
Information<br />
Infant Motor Acts<br />
Figure 6-2. A conceptual schematic of <strong>the</strong> „active intermodal mapping‟ (AIM) hypo<strong>the</strong>sis adapted from<br />
Meltzoff <strong>and</strong> Moore (1997). They make use of a traditional form of explanation: <strong>the</strong> imitative ability of<br />
human neonates is accounted for by appealing to an intermodal equivalence detector between <strong>the</strong><br />
neonate‟s visual percept of <strong>the</strong> o<strong>the</strong>r (O<strong>the</strong>r-BI) <strong>and</strong> its proprioception (PI).<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Body schema (BS)<br />
Proprioception<br />
(PI)<br />
Motor act<br />
Figure 6-3. A revised version of <strong>the</strong> schematic in Figure 6-1. In <strong>the</strong> case of neonate imitation traditional<br />
explanations cannot appeal to <strong>the</strong> existence of a body image based on visual perception of <strong>the</strong> neonate‟s<br />
body (Self-BI). Accordingly, Meltzoff <strong>and</strong> Moore (1977) propose <strong>the</strong>ir „active intermodal mapping‟<br />
(AIM) hypo<strong>the</strong>sis which is illustrated in Figure 6-2.<br />
Apparently <strong>the</strong> existence of such a general matching process does not require extensive<br />
periods of learning appropriate intermodal correlations since neonatal imitation has been<br />
demonstrated with infants even less than a month old (Meltzoff & Borton 1979). Here<br />
86 | P a g e
we thus have an example of how bodily coordination can be explained in <strong>the</strong> traditional<br />
framework by appealing to <strong>the</strong> existence of an O<strong>the</strong>r-BI <strong>and</strong> PI alone without <strong>the</strong> need<br />
for a perceptually formed self-BI as well. This possibility is illustrated in Figure 6-3.<br />
While <strong>the</strong> existence of an innate BS can explain <strong>the</strong> ability of <strong>the</strong> neonates to engage <strong>the</strong><br />
appropriate part of <strong>the</strong>ir bodies, it is difficult to see how this BS alone can explain <strong>the</strong>ir<br />
ability to actually improve this performance. How do <strong>the</strong>y know which PI matches <strong>the</strong><br />
target PI that <strong>the</strong>y would receive when accurately copying <strong>the</strong> intended movement? 20 To<br />
be fair, it could be argued that at least a minimal Self-BI is also present for <strong>the</strong> neonates<br />
because of <strong>the</strong> existence of proprioceptive awareness (PA). Gallagher <strong>and</strong> Meltzoff<br />
(1996), for example, suggest that PA, as a tacit, pre-reflective awareness, constitutes <strong>the</strong><br />
very beginning of a primitive form of Self-BI. Thus, a primitive self-BI might also play<br />
an essential role, namely as <strong>the</strong> comparative goal-state to be attained by <strong>the</strong> innate BS.<br />
6.2.3 Gesturing by a deafferented subject (I)<br />
Is it possible to better differentiate between <strong>the</strong> contributions of <strong>the</strong> self-BI <strong>and</strong> PI for<br />
bodily coordination? We can gain some fur<strong>the</strong>r insights by considering <strong>the</strong> case of a<br />
deafferented subject, Ian Waterman (sometimes referred to as IW in <strong>the</strong> literature), who<br />
has lost all sense of touch <strong>and</strong> proprioception in his body below <strong>the</strong> neck. This acute<br />
sensory neuropathy developed in 1971, when Ian was 19 years old, because of an illness<br />
which damaged <strong>the</strong> large myelinated fibers below his neck. The case of Ian has been<br />
well document by his doctor, Jonathan Cole (1995). Ian‟s case is of interest here<br />
because one way to describe his condition is that of a lost body schema:<br />
20 Since some readers might be tempted to answer „mirror neurons‟ to this question (e.g. Rizzolatti, et al.<br />
2001; Gallese & Goldman 1998), it is fitting to very briefly consider why this response is fundamentally<br />
inadequate. First, <strong>the</strong>re are deep conceptual difficulties with <strong>the</strong> notion of „mirroring‟, including <strong>the</strong> lack<br />
of pretense at <strong>the</strong> level of neurons (Gallagher 2007). And even if <strong>the</strong>se difficulties could be resolved <strong>the</strong>re<br />
is experimental evidence that <strong>the</strong> activation of <strong>the</strong>se neurons is nothing specifically social, but ra<strong>the</strong>r a<br />
contingent outcome of associative learning that can even be reversed (Catmur, et al. 2007). Finally, <strong>the</strong><br />
reference to „mirror‟ system activation just shifts <strong>the</strong> explanatory problem to <strong>the</strong> neural domain: what is<br />
<strong>the</strong> mechanism which makes <strong>the</strong>se neurons fire when <strong>the</strong> infant sees <strong>the</strong> o<strong>the</strong>r‟s movement, <strong>and</strong> what<br />
mechanism can relate this firing activity to <strong>the</strong> infant‟s accurate copying of <strong>the</strong> target movement?<br />
87 | P a g e
At <strong>the</strong> earliest stage of his illness he had no control over his movements <strong>and</strong> was<br />
unable to put intention into action. There was, one might say, a disconnection of<br />
will from <strong>the</strong> specifics of movement. If Ian decided to move his arm in a certain<br />
direction, <strong>and</strong> <strong>the</strong>n tried to carry out <strong>the</strong> intended movement, <strong>the</strong> arm <strong>and</strong> o<strong>the</strong>r<br />
parts of his body would move in unpredictable ways. Without support, Ian was<br />
unable to maintain anything o<strong>the</strong>r than a prone posture. He had no knowledge of<br />
limb position unless he saw <strong>the</strong> limb. But even with vision, he had no control<br />
over his movement. Because of <strong>the</strong> absence of proprioceptive <strong>and</strong> tactile<br />
feedback his entire body schema system failed. (Gallagher 2005, p. 44)<br />
The case of Ian thus highlights <strong>the</strong> importance of proprioception for <strong>the</strong> basic<br />
maintenance of posture <strong>and</strong> <strong>the</strong> governance of movement. However, even though Ian<br />
never recovered from <strong>the</strong> original neuropathy he never<strong>the</strong>less recovered control over his<br />
movement <strong>and</strong> regained a close to normal <strong>life</strong> as a result of extreme effort. He has been<br />
able to address <strong>the</strong> motor problem on a cognitive <strong>and</strong> behavioral level by rebuilding a<br />
partial <strong>and</strong> very minimal body schema, in terms of nearly automated cognitive<br />
processes, <strong>and</strong> by using his well-developed body image to help consciously control<br />
movement:<br />
When we say that he has regained control over his posture <strong>and</strong> movement, we do<br />
not mean that he has recovered from <strong>the</strong> neuropathy that destroyed his sensory<br />
nerves. Ra<strong>the</strong>r his control of movement is based primarily on visual attention<br />
<strong>and</strong> cognitive effort (although some aspects of walking have become close to<br />
automatic due to consciously guided practice). In <strong>the</strong> dark, controlled movement<br />
is impossible since he has not visual access to current position of his limbs <strong>and</strong><br />
cannot tell where <strong>the</strong>y are in relation to one ano<strong>the</strong>r. (Cole, et al. 2002, p. 51)<br />
In o<strong>the</strong>r words, even after his behavioral recovery “Ian still does not know, without<br />
visual perception, where his limbs are or what posture he maintains. In order to maintain<br />
motor control he must conceptualize his movements <strong>and</strong> keep certain parts of his body<br />
in his visual field. His movement requires constant visual <strong>and</strong> mental concentration”<br />
(Gallagher 2005, p. 44). How has Ian‟s neurophysiological loss of a body schema <strong>and</strong><br />
subsequent recovery by means of a highly developed body image affected his ability to<br />
88 | P a g e
gesture? Cole, Gallagher <strong>and</strong> McNeill (2002a; 2002b) report on a series of experiments<br />
conducted with Ian that were designed to investigate this question. When Ian was asked<br />
to narrate a cartoon he had just seen <strong>the</strong>y found that “when vision of his h<strong>and</strong>s was<br />
available, he made numerous meaningful gestures well synchronized with his coexpressive<br />
speech, confirming that he had <strong>the</strong> ability to produce gestures. His gestural<br />
performance looked essentially identical to non-neuropathic performance, <strong>and</strong> fur<strong>the</strong>r<br />
computerized analysis of <strong>the</strong> video confirmed this” (Cole, et al. 2002a, p. 55).<br />
Ian reports that his relatively normal gestural performance is made possible by <strong>the</strong> fact<br />
that he consciously initiates <strong>the</strong> gestures, <strong>and</strong> that he controls <strong>the</strong>m just like he does for<br />
<strong>the</strong> case of instrumental movement, namely by continually checking his body by means<br />
of visual perception. This possible explanation is depicted in Figure 6-4.<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body schema (BS)<br />
Motor act<br />
Figure 6-4. The case of Ian Waterman presents us with <strong>the</strong> opposite situation of human neonates (at least<br />
to some extent). The neurophysiological loss of proprioception (PI <strong>and</strong> PA) <strong>and</strong> <strong>the</strong> subsequent behavioral<br />
compensation via visual perception have left him with a highly developed body image (Self-BI) but only<br />
a primitive body schema (BS), indicated by <strong>the</strong> non-solid box. Under <strong>the</strong>se conditions it is still possible to<br />
explain his successful gestural coordination in terms of <strong>the</strong> traditional framework, namely by appealing to<br />
a reflective (cognitive) comparison between <strong>the</strong> Self-BI <strong>and</strong> O<strong>the</strong>r-BI.<br />
Note that <strong>the</strong> case of Ian seemingly presents us with <strong>the</strong> opposite situation of human<br />
neonates. He can achieve bodily coordination in terms of his reliance on a visually<br />
guided Self-BI, <strong>and</strong> without proprioception (PI <strong>and</strong> PA). Accordingly, <strong>the</strong> cases of<br />
89 | P a g e
human neonate imitation <strong>and</strong> Ian‟s gestural competence have indicated that nei<strong>the</strong>r a<br />
Self-BI nor pre-reflective PI is a necessary condition for bodily coordination. On <strong>the</strong><br />
o<strong>the</strong>r h<strong>and</strong>, ei<strong>the</strong>r of <strong>the</strong>se factors in combination with a visually formed O<strong>the</strong>r-BI<br />
appears to be a sufficient condition to explain <strong>the</strong> existence of such bodily coordination.<br />
But is this kind of visual O<strong>the</strong>r-BI itself a necessary condition? How can we make sense<br />
of coordination between agents that cannot perceive each o<strong>the</strong>r as such?<br />
6.2.4 Perceptual crossing in a virtual space<br />
In order to answer this question we need to find a case in which a normal participant,<br />
i.e. one with a visually guided Self-BI <strong>and</strong> pre-reflective PI, can engage in mutual<br />
interactions with ano<strong>the</strong>r participant, but where that partner is not perceived in terms of<br />
an O<strong>the</strong>r-BI. One way to consider this situation is by means of Auvray, Lenay <strong>and</strong><br />
Stewart‟s (2009) psychological experiment in perceptual crossing. This study attempts<br />
to explore <strong>the</strong> necessary conditions for participants to locate each o<strong>the</strong>r through minimal<br />
technologically mediated interaction in a shared virtual space. A schematic of <strong>the</strong><br />
experimental setup is shown in Figure 6-5.<br />
Figure 6-5. The experimental setup of Auvray, Lenay <strong>and</strong> Stewart‟s (2009) study in perceptual crossing.<br />
Two participants face each o<strong>the</strong>r in a 600 unit long 1-D wrap-around environment. Note that each<br />
participant‟s receptor field (white) can encounter three different objects: a static object (gray), <strong>the</strong> o<strong>the</strong>r<br />
participant‟s receptor field (gray), <strong>and</strong> <strong>the</strong> o<strong>the</strong>r participant‟s „shadow‟ of that receptor field (gray).<br />
Since this experimental setup will be <strong>the</strong> basis for <strong>the</strong> models presented in Chapters 9<br />
<strong>and</strong> 10, it will be helpful to describe it in a bit more detail here. Two adult participants,<br />
acting under <strong>the</strong> same conditions, can move a mouse cursor left <strong>and</strong> right along a shared<br />
one-dimensional virtual tape that wraps around at <strong>the</strong> edges. They are asked to indicate<br />
<strong>the</strong> presence of <strong>the</strong> o<strong>the</strong>r partner by clicking <strong>the</strong>ir mouse pointer. The participants are<br />
90 | P a g e
lindfolded, in separate rooms, <strong>and</strong> all <strong>the</strong>y can sense are on/off tactile stimulations on a<br />
finger when <strong>the</strong>ir cursor crosses an entity on <strong>the</strong> tape. Apart from each o<strong>the</strong>r‟s receptor<br />
field, participants can encounter two o<strong>the</strong>r objects: a static object on <strong>the</strong> tape, <strong>and</strong> a<br />
displaced „shadow image‟ of <strong>the</strong> partner, which moves strictly in an identical manner to<br />
<strong>the</strong> partner‟s receptor (though it is not a source of stimulation for that partner). Thus,<br />
each participant can encounter three different types of object in <strong>the</strong> shared environment:<br />
(i) The four-unit wide sensory receptor field of <strong>the</strong> o<strong>the</strong>r participant. When any of<br />
<strong>the</strong> four sensors of an participant overlaps with any sensors of <strong>the</strong> o<strong>the</strong>r<br />
participant, both of <strong>the</strong>m receive sensory stimulation. This possibility of<br />
„perceptual interaction‟, or shared perception, represents <strong>the</strong> way in which an<br />
embodied perceiver cannot observe someone else without also at <strong>the</strong> same time<br />
being perceivable in some manner, at least potentially (e.g. mutual gaze or<br />
touch).<br />
(ii) A four-unit wide static object that is placed at a specific location. There are two<br />
static objects, one for participant „up‟ <strong>and</strong> one for participant „down‟, which are<br />
located between 148-152 units <strong>and</strong> 448-452 units, respectively. Each participant<br />
can only perceive its specific static object. The objects were chosen to be<br />
participant specific <strong>and</strong> placed in diametrically opposite locations of <strong>the</strong><br />
environment in order to encourage <strong>the</strong> participants‟ displacement within <strong>the</strong><br />
whole 1-D space.<br />
(iii) A four-unit wide ‘shadow’ object, whose position tracks that of <strong>the</strong> o<strong>the</strong>r<br />
participant at a fixed distance. The mobile shadow object reproduces <strong>the</strong> exact<br />
same movement as <strong>the</strong> receptor field of <strong>the</strong> o<strong>the</strong>r participant, but is displaced by<br />
48 units. Never<strong>the</strong>less, in contrast to an actual inter-individual encounter it does<br />
not give rise to <strong>the</strong> possibility of two-way or mutual perceptual interaction.<br />
Thus, if a participant encounters <strong>the</strong> o<strong>the</strong>r‟s shadow object, it will indeed<br />
receive <strong>the</strong> same sensory stimulation as if it had encountered <strong>the</strong> o<strong>the</strong>r<br />
participant. However, since <strong>the</strong> o<strong>the</strong>r participant does not receive any<br />
corresponding sensory stimulation, <strong>the</strong>re is only <strong>the</strong> possibility of „one-way‟<br />
interaction.<br />
91 | P a g e
In summary, <strong>the</strong>re are three distinct types of objects which can be encountered by a<br />
participant, one of which is placed at a fixed location <strong>and</strong> two of which are moving<br />
within <strong>the</strong> 1-D space. All objects are four units wide. The two mobile objects exhibit<br />
exactly <strong>the</strong> same movement, but only an overlap of <strong>the</strong> receptor fields of both<br />
participants gives rise to mutual sensory stimulation. Note that <strong>the</strong> difference between<br />
<strong>the</strong>se three types of objects cannot be directly provided by <strong>the</strong> sensors, which in all<br />
cases can only produce a binary, all-or-nothing response depending on whe<strong>the</strong>r<br />
something is overlapping <strong>the</strong>ir particular receptor field or not. There is no O<strong>the</strong>r-BI<br />
available; all entities produce <strong>the</strong> same immediate percept. Thus, if <strong>the</strong> participants are<br />
to be successful at distinguishing which of <strong>the</strong> encountered objects is <strong>the</strong> o<strong>the</strong>r agent (or<br />
more precisely, its receptor field), <strong>the</strong>y must accordingly rely on differences in <strong>the</strong> kinds<br />
of interactions that <strong>the</strong>se objects afford.<br />
The results of <strong>the</strong> psychological study show that, at least under <strong>the</strong> minimalist<br />
conditions of this experiment, <strong>the</strong> successful recognition of an ongoing interaction with<br />
ano<strong>the</strong>r person is impossible for individual participants. There is no significant<br />
difference in <strong>the</strong> probability of responding to an encounter with <strong>the</strong> o<strong>the</strong>r‟s receptor or<br />
with <strong>the</strong> „shadow image‟. Thus, overall success in this task cannot be due to individual<br />
capacities alone. It is also based on certain properties that are intrinsic to <strong>the</strong> joint<br />
perceptual activity itself. The important issue is that <strong>the</strong> scanning of an object<br />
encountered will only stabilize in <strong>the</strong> case that both partners are in contact with each<br />
o<strong>the</strong>r. If interaction is only one-way, between a participant <strong>and</strong> <strong>the</strong> o<strong>the</strong>r‟s shadow, <strong>the</strong><br />
shadow will eventually move away, because <strong>the</strong> participant it is shadowing (<strong>the</strong> partner)<br />
is still engaged in searching activity. Therefore, <strong>the</strong> solution to <strong>the</strong> task does not only<br />
rely on individuals performing <strong>the</strong> right kind of perceptual discrimination between<br />
different momentary sensory patterns, but also emerges from <strong>the</strong> mutual perceptual<br />
activity of <strong>the</strong> experimental subjects that is oriented towards each o<strong>the</strong>r. Moreover, this<br />
perceptual activity typically involves <strong>the</strong> spontaneous emergence of coordinated<br />
behaviours, namely both participants moving <strong>the</strong> mouse pointer so as to oscillate around<br />
92 | P a g e
each o<strong>the</strong>r. Bodily coordination in <strong>the</strong> form of two-way mutual scanning is, given <strong>the</strong><br />
task <strong>and</strong> experimental setup, <strong>the</strong> only long-term stable solution 21 .<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Proprioception<br />
(PI)<br />
Motor Act<br />
Figure 6-6. Under <strong>the</strong> minimalist conditions of perceptual crossing that have been investigated by<br />
Auvray, Lenay <strong>and</strong> Stewart (2009) bodily coordination emerges despite <strong>the</strong> lack of any evident O<strong>the</strong>r-BI.<br />
In sum, Auvray, Lenay <strong>and</strong> Stewart‟s study indicates that <strong>the</strong> factor of O<strong>the</strong>r-BI that has<br />
been appealed to in <strong>the</strong> previous cases is not a necessary condition in order to explain<br />
<strong>the</strong> existence of bodily coordination. The participants of <strong>the</strong> study are individually not<br />
able to distinguish between <strong>the</strong> o<strong>the</strong>r‟s avatar <strong>and</strong> <strong>the</strong> linked shadow, but can still<br />
collectively engage in coordinated behavior (cf. Figure 6-6). Accordingly, this study<br />
also indicates something much more important: <strong>the</strong>re is ano<strong>the</strong>r factor available that <strong>the</strong><br />
traditional form of explanation has not even properly considered yet, namely <strong>the</strong> selforganizing<br />
properties of <strong>the</strong> interaction process itself. In o<strong>the</strong>r words, in this case <strong>the</strong><br />
direct perception of <strong>the</strong> o<strong>the</strong>r participant (presumably necessary for <strong>the</strong> constitution of<br />
an O<strong>the</strong>r-BI) was not necessary. The overall experimental situation was organized in<br />
such a way that <strong>the</strong> appropriate individual behavior emerged spontaneously out of <strong>the</strong><br />
stabilities <strong>and</strong> instabilities of <strong>the</strong> interaction process (IP) within a multi-agent system.<br />
21 Strictly speaking, a solitary interaction with <strong>the</strong> static object is ano<strong>the</strong>r stable behavior, as long as <strong>the</strong><br />
participant does not realize her mistake <strong>and</strong> moves elsewhere. The possibility of getting stuck around<br />
static objects can <strong>the</strong>refore cause some problems. We will return to this issue in Chapter 10.<br />
93 | P a g e
6.2.5 Deafferented subject under a blind<br />
Ano<strong>the</strong>r way to develop a better appreciation of <strong>the</strong> role of <strong>the</strong> IP as a potential factor<br />
which can be appealed to in order to explain bodily coordination is to consider <strong>the</strong> more<br />
extreme cases. We have already seen that it is possible for a deafferented subject like<br />
Ian Waterman to be integrated within everyday <strong>life</strong>, even if he has to compensate for his<br />
lack of proprioception with a constantly visually updated Self-BI (cf. Section 6.2.3). We<br />
also know that he is incapable of performing instrumental actions without such visual<br />
feedback: “In reaching to grasp, for example, he has to see not only <strong>the</strong> target object,<br />
but also his h<strong>and</strong>. On <strong>the</strong> basis of what he sees, he needs to think about how to shape his<br />
h<strong>and</strong> in order to pick up <strong>the</strong> object” (Gallagher 2005, p. 50). Accordingly, it is to be<br />
expected that his ability to gesture would be severely impaired when he is not able to<br />
guide his h<strong>and</strong>s <strong>and</strong> arms by means of visual feedback.<br />
To discover whe<strong>the</strong>r Ian controlled his gestures using visual feedback, a blind was<br />
placed to block <strong>the</strong> view of his h<strong>and</strong>s. In this blind condition he is able to perform<br />
gestures similar to those seen without <strong>the</strong> blind, even though he is now lacking both a<br />
Self-BI <strong>and</strong> any sense of proprioception. This situation is depicted in Figure 6-7.<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Body Schema (BS)<br />
Motor Act<br />
Figure 6-7. When Ian gestures under <strong>the</strong> blind he lacks both visual <strong>and</strong> proprioceptive (PI <strong>and</strong> PA)<br />
feedback. But even under <strong>the</strong>se conditions he is able to coordinate his behavior meaningfully with his<br />
interlocutors. How can traditional accounts explain this situation?<br />
94 | P a g e
Gallagher (2005, p. 113) offers three factors to explain Ian‟s performance: (i) prosodic<br />
feedback in speech, (ii) semiotic feedback, i.e. <strong>the</strong> co-expressiveness of gesture <strong>and</strong><br />
speech, <strong>and</strong> (iii) pragmatic feedback relating to <strong>the</strong> communicative situation. It is factor<br />
(iii) that we are interested in here, which may include <strong>the</strong> meter or cadence of <strong>the</strong><br />
conversation, <strong>and</strong> response gestures made by one‟s interlocutor. These may offer some<br />
important cues for maintaining synchronization. Since Ian can perform gestures almost<br />
normally <strong>and</strong> fluently under conditions in which instrumental action would be difficult<br />
or impossible for him, namely without visual feedback, we can conclude that gestural<br />
<strong>and</strong> instrumental movement cannot be <strong>the</strong> same with regard to <strong>the</strong> mechanisms that<br />
generate <strong>the</strong>m. Gallagher explains:<br />
[G]esture is an action that helps to create <strong>the</strong> narrative space that is shared in <strong>the</strong><br />
communicative situation. This suggests that it is part of <strong>and</strong> is controlled by a<br />
linguistic/communicative system ra<strong>the</strong>r than a motor system. […] The fact that<br />
Ian‟s gestures can be decoupled from visual monitoring <strong>and</strong> can be performed<br />
without sensory feedback of any kind, <strong>and</strong> yet remain relatively accurate in time<br />
<strong>and</strong> form, suggests that gestural movements are controlled by a system that is in<br />
some measure independent from <strong>the</strong> system that controls <strong>the</strong> same muscles in<br />
instrumental actions. (Gallagher 2005, pp. 117-118)<br />
To be sure, due to Ian‟s condition it is not sufficient to appeal to <strong>the</strong> motor system in<br />
this case, but what does <strong>the</strong> additional control by this linguistic/communicative system<br />
precisely consist in? Unfortunately, due to Cole, Gallagher <strong>and</strong> McNeill‟s (2002a) focus<br />
on Ian‟s behavior alone, nowhere do we find a description of <strong>the</strong> interaction process or<br />
gesturing <strong>and</strong> general social presence of Ian‟s interlocutors. Why is nothing said about<br />
<strong>the</strong> pragmatic feedback relating to <strong>the</strong> communicative situation that is available?<br />
For our present purposes we can fortunately refer to two pictures of <strong>the</strong> experimental<br />
situation that are depicted in Gallagher‟s (2005) book on pages 112 <strong>and</strong> 115. The former<br />
shows Ian under a blind, facing a visible interlocutor <strong>and</strong> ano<strong>the</strong>r potentially visible<br />
participant sitting in a corner. The latter shows Ian wearing an eye-tracking device<br />
explaining, with <strong>the</strong> aid of maps <strong>and</strong> floor plans, how he had arrived at <strong>the</strong> lab that<br />
morning. At first sight it appears that he is alone in <strong>the</strong> room, but a closer look reveals<br />
95 | P a g e
that <strong>the</strong>re is someone sitting diagonally in front of him. She would be potentially visible<br />
to Ian if he only slightly turned his head.<br />
This makes it likely that Ian had access to an O<strong>the</strong>r-BI during <strong>the</strong> gesturing. But in what<br />
way was <strong>the</strong>re mutual interaction between him <strong>and</strong> <strong>the</strong> o<strong>the</strong>rs? Was <strong>the</strong>re coordination<br />
of gestures? The lack of detailed description on this point is surprising considering that<br />
Gallagher‟s integrative <strong>the</strong>ory of gesture portrays gesture as embodied (constrained <strong>and</strong><br />
enabled by motoric possibilities), communicative (pragmatically intersubjective), <strong>and</strong><br />
cognitive (contributing to thought). What is <strong>the</strong> role of pragmatic intersubjectivity for<br />
gesture here, o<strong>the</strong>r than perhaps to function as a source of inputs like it does in <strong>the</strong> AIM<br />
hypo<strong>the</strong>sis shown in Figure 6-2? Unfortunately, not much more is said on this matter,<br />
o<strong>the</strong>r than a statement to <strong>the</strong> effect that “gesture, as a movement concerned with <strong>the</strong><br />
construction of significance ra<strong>the</strong>r than with doing something, is organized primarly by<br />
<strong>the</strong> linguistic-communicative context” (Gallagher 2005, p. 120). But how precisely does<br />
<strong>the</strong> social context organize mere movement into gesture? And does <strong>the</strong> efficacy of this<br />
context require <strong>the</strong> existence of a visually formed O<strong>the</strong>r-BI?<br />
6.2.6 Gesturing by a deafferented subject (II)<br />
As a first step toward addressing <strong>the</strong>se questions we can ask: Can Ian gesture normally<br />
under conditions when (i) he has visual access to his own movements (Self-BI), but (ii)<br />
has no access to <strong>the</strong> movements of his interlocutors (O<strong>the</strong>r-BI)? This question has not<br />
yet been specifically investigated with Ian or ano<strong>the</strong>r deafferented subject, but we can<br />
try to piece a possible answer toge<strong>the</strong>r by considering what has been reported by Cole,<br />
Gallagher <strong>and</strong> McNeill (2002a; 2002b) <strong>and</strong> Gallagher (2005, pp. 107-129).<br />
We have already noted with respect to <strong>the</strong> previous case that, even though it is difficult<br />
to tell from <strong>the</strong> descriptions provided of <strong>the</strong> experimental situations, it appears that Ian‟s<br />
gesturing has always been observed in conditions where o<strong>the</strong>rs were visually present,<br />
<strong>the</strong>reby providing him with access to a visually formed O<strong>the</strong>r-BI to coordinate his<br />
gestural movements. This O<strong>the</strong>r-BI can <strong>the</strong>refore be appealed to by a traditional account<br />
in order to explain Ian‟s ability to gesture meaningfully under <strong>the</strong> blind where he cannot<br />
96 | P a g e
see his h<strong>and</strong>s or arms. However, a more enactive explanation that emphasizes <strong>the</strong> role of<br />
<strong>the</strong> interaction process is also possible.<br />
Consider, for example, <strong>the</strong> ability of participants to coordinate <strong>the</strong>ir movements during<br />
Auvray et al.‟s psychological study. In this setup a minimalist technological interface<br />
prevented <strong>the</strong> participants from directly perceiving <strong>the</strong> o<strong>the</strong>r as such, <strong>the</strong>reby leading to<br />
an individual failure of distinguishing that o<strong>the</strong>r‟s presence, but never<strong>the</strong>less resulted in<br />
collective coordination of perceptual crossing. It is <strong>the</strong>refore possible that in <strong>the</strong> case of<br />
Ian <strong>the</strong> interaction process could similarly be sufficient to organize his gestural activity<br />
appropriately, even without visual access to o<strong>the</strong>rs. It is evident, for example, that Ian‟s<br />
gesturing is shaped by <strong>the</strong> general context of <strong>the</strong> social situation, including its meaning,<br />
verbal structure <strong>and</strong> timing, even without conscious awareness (i.e. under <strong>the</strong> blind):<br />
IW‟s gestures reflected <strong>the</strong> meaning he was attempting to convey with a<br />
significant degree of morphokinetic precision, suggesting that meaning plays a<br />
part in how those gestures are shaped <strong>and</strong> how his h<strong>and</strong>s move in gesture. IW‟s<br />
gestures were also precisely synchronized with <strong>the</strong> verb phrases he used to<br />
describe various events. His gestures are normal both with respect to<br />
morphokinesis <strong>and</strong> with respect to timing. (Cole, et al. 2002a, p. 57) 22<br />
Moreover, it is clear that Ian also sometimes guides his gesturing by means of visual<br />
feedback, even while not directly observing his interlocutors at <strong>the</strong> time, for example<br />
during situations of joint attention. Such a situation occurred during <strong>the</strong> eye-tracking<br />
experiment when Ian was asked to describe how he had arrived at <strong>the</strong> lab that morning:<br />
“While recounting his journey <strong>and</strong> arrival most of his gestures involved pointing at <strong>the</strong><br />
map or floor plan, with his gaze directed at <strong>the</strong> same place as he was pointing”<br />
(Gallagher 2005, p. 115). In his book Gallagher complements this account with a picture<br />
22 Note that a lingering trace of methodological individualism appears in Cole, Gallagher <strong>and</strong> McNeill‟s<br />
approach to synchrony <strong>and</strong> timing. Instead of using <strong>the</strong>se terms in relation to <strong>the</strong> dynamics of <strong>the</strong> interindividual<br />
interaction process <strong>and</strong> <strong>the</strong> unfolding of bodily coordination between Ian <strong>and</strong> o<strong>the</strong>rs, <strong>the</strong>y are<br />
only used to indicate intra-individual coordination (e.g. <strong>the</strong> internal relation between h<strong>and</strong> movement <strong>and</strong><br />
speech production). In terms of this internalist focus <strong>the</strong>ir position is somewhat akin to <strong>the</strong> traditional<br />
framework of <strong>the</strong> „active intermodal mapping‟ hypo<strong>the</strong>sis of Meltzoff <strong>and</strong> Moore.<br />
97 | P a g e
of Ian looking at a floor plan while pointing to a particular location, with an onlooker<br />
observing this action. This kind of social situation, where nei<strong>the</strong>r O<strong>the</strong>r-BI nor PI is<br />
immediately available, is schematically depicted in Figure 6-8.<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Motor Act<br />
Figure 6-8. Experiments with Ian Waterman could present us with an example of gestural coordination<br />
that nei<strong>the</strong>r depends on <strong>the</strong> visual perception of bodily movements of <strong>the</strong> o<strong>the</strong>r interlocutor (O<strong>the</strong>r-BI),<br />
nor on any form of PI. The restricted availability of an O<strong>the</strong>r-BI, which may still be informed by <strong>the</strong><br />
overall social context, is indicated by <strong>the</strong> non-solidity of <strong>the</strong> circle.<br />
To be fair, such an intermittent episode during a joint attentional scene does not fully<br />
qualify for <strong>the</strong> case we are looking for, especially since Ian often visually faced <strong>the</strong><br />
o<strong>the</strong>r during <strong>the</strong> experiment: “Several times <strong>the</strong> gaze-tracking cross-hairs zeroed in on<br />
<strong>the</strong> interlocutor‟s face while Ian‟s gestures remained in perfect synchrony with his<br />
speech <strong>and</strong> were semantically appropriate for non-deictic comments – normal gestures<br />
in every way” (Gallagher 2005, p. 115). Never<strong>the</strong>less, we can derive a novel empirical<br />
prediction from <strong>the</strong>se considerations. If Ian is engaged in a mutually responsive social<br />
interaction during which he is able to control his movements by means of a visually<br />
informed Self-BI, but <strong>the</strong> o<strong>the</strong>r is not visually present (e.g. a conversation with someone<br />
in <strong>the</strong> next room), he will still be able to perform appropriate gestures. Moreover, if a<br />
Self-BI is sufficient for Ian‟s ability of bodily coordination even when only a severly<br />
restricted O<strong>the</strong>r-BI is available, we can hypo<strong>the</strong>size that for a normal experimental<br />
participant PI alone could be sufficient. These conjectures may appear to be forced, but<br />
<strong>the</strong> point is precisely to drive home <strong>the</strong> pervasive structuring presence of social context.<br />
98 | P a g e
6.2.7 Bodily coordination in a virtual space (I)<br />
Is <strong>the</strong>re a way to investigate whe<strong>the</strong>r bodily coordination is possible for a subject with<br />
PI, but where <strong>the</strong>re is nei<strong>the</strong>r a Self-BI nor any substantial O<strong>the</strong>r-BI available? Indeed,<br />
this made possible by means of ano<strong>the</strong>r minimalist psychological experiment devised by<br />
Charles Lenay <strong>and</strong> his group at <strong>the</strong> University of Compiègne in France. The study has<br />
not been published yet, so <strong>the</strong> description of <strong>the</strong> experimental setup <strong>and</strong> <strong>the</strong> results<br />
provided here are based on recent talks given by Lenay 23 .<br />
The general setup of <strong>the</strong> experiment is quite similar to that of <strong>the</strong> perceptual crossing<br />
study by Auvray, Lenay <strong>and</strong> Stewart (2009) that was described in Section 6.2.4 (cf. pp.<br />
90-94). Two participants face each o<strong>the</strong>r in a 1-D virtual environment that wraps around<br />
on itself after 400 units of space. Each participant can control <strong>the</strong>ir virtual embodiment,<br />
which consists of two aspects: (i) <strong>the</strong> ‘body as subject’, through which <strong>the</strong> environment<br />
is perceived, <strong>and</strong> (ii) <strong>the</strong> ‘body as object’, which can be perceived in <strong>the</strong> environment by<br />
o<strong>the</strong>rs. The subjective aspect of this virtual body is implemented as a basic receptor<br />
field that is activated (tactile stimulus) when <strong>the</strong> field overlaps with <strong>the</strong> o<strong>the</strong>r’s body<br />
object, o<strong>the</strong>rwise <strong>the</strong> receptor remains off (no tactile stimulus). The receptor field is<br />
moved horizontally by means of a mouse pointer, <strong>and</strong> it remains invisible to <strong>the</strong> o<strong>the</strong>r<br />
participant. The objective aspect is represented simply by an object which is attached by<br />
a variable-length connection to <strong>the</strong> receptor field. The receptor field <strong>and</strong> body object<br />
each occupy 8 units of space.<br />
The horizontal displacement of <strong>the</strong> body object from <strong>the</strong> receptor field is controlled by<br />
means of mouse clicks (i.e. left click for decrease <strong>and</strong> right click for increase), <strong>and</strong> it is<br />
<strong>the</strong> only object that is detectable by <strong>the</strong> o<strong>the</strong>r’s receptor field (i.e. only <strong>the</strong> body object<br />
gives rise to a tactile stimulation on contact). Importantly, participants are told <strong>the</strong><br />
direction of correlation between clicks <strong>and</strong> changes to body displacement at <strong>the</strong> start of<br />
<strong>the</strong> experiment (e.g. left-click for body size increase, etc.). This enables each participant<br />
23 One talk was given in August 2008, at <strong>the</strong> Workshop on Enactive Approaches to Social Cognition, held<br />
in Battle, UK, <strong>and</strong> <strong>the</strong> o<strong>the</strong>r was given in March 2009, as part of <strong>the</strong> Life <strong>and</strong> Mind seminars at <strong>the</strong><br />
University of Sussex, Brighton, UK. See: http://<strong>life</strong><strong>and</strong><strong>mind</strong>.wordpress.com<br />
99 | P a g e
to change <strong>the</strong> relative displacement of <strong>the</strong>ir body object according to <strong>the</strong> unfolding of<br />
<strong>the</strong> interaction process. This setup is illustrated schematically in Figure 6-9.<br />
Figure 6-9. A schematic of <strong>the</strong> experimental setup of <strong>the</strong> minimalist psychological study in body<br />
imitation designed by Lenay <strong>and</strong> colleagues at <strong>the</strong> University of Compiègne, France. Two participants<br />
face each o<strong>the</strong>r in a 400 unit long 1-D virtual environment that wraps around at <strong>the</strong> boundaries. Their<br />
virtual embodiment consists of two aspects: (i) a receptor field, <strong>and</strong> (ii) a body object, which is attached<br />
by a variable-length connection to <strong>the</strong> receptor field. Note that each participant‟s receptor field (white)<br />
can encounter only one object in <strong>the</strong> environment that gives rise to tactile stimulation: <strong>the</strong> o<strong>the</strong>r<br />
participant‟s body object (gray).<br />
The task for <strong>the</strong> participants is to engage in mutual perceptual crossing as best as<br />
possible. If <strong>the</strong>y are to achieve optimal performance <strong>the</strong>y will have to coordinate <strong>the</strong><br />
displacements of <strong>the</strong>ir body objects such that <strong>the</strong>y can be mutually close to each o<strong>the</strong>r‟s<br />
receptor field. O<strong>the</strong>rwise <strong>the</strong> situation will be inherently unstable: if one agent has to<br />
move its receptor field to locate <strong>the</strong> o<strong>the</strong>r‟s body object, its own body will follow this<br />
movement (since it is attached to <strong>the</strong> agent‟s receptor field with a certain displacement),<br />
<strong>the</strong>reby in turn forcing <strong>the</strong> o<strong>the</strong>r agent to move as well, <strong>and</strong> so forth.<br />
The task of coordinating displacements of body objects is made non-trivial due to <strong>the</strong><br />
following factors: (i) <strong>the</strong> receptor fields begin at r<strong>and</strong>om positions at <strong>the</strong> start of each<br />
trial (range [0, 400]), (ii) each participant‟s body object is attached to its receptor field<br />
by a r<strong>and</strong>om displacement at <strong>the</strong> start of each trial (range [-100, 100]), (iii) <strong>the</strong>y have no<br />
knowledge of <strong>the</strong> position of <strong>the</strong>ir receptor field nor <strong>the</strong> displacement of <strong>the</strong>ir body<br />
object, <strong>and</strong> (iv) <strong>the</strong>y have no knowledge of <strong>the</strong> position of <strong>the</strong> o<strong>the</strong>r‟s receptor field, nor<br />
<strong>the</strong> displacement of <strong>the</strong> o<strong>the</strong>r‟s body object. The only information directly available to<br />
<strong>the</strong> participants is <strong>the</strong> binary activation status of <strong>the</strong>ir receptor field, which indicates<br />
whe<strong>the</strong>r <strong>the</strong> field is currently overlapping with <strong>the</strong> o<strong>the</strong>r‟s body object or not.<br />
100 | P a g e
We have thus found an experimental setup to investigate this particular distribution of<br />
traditional explanatory factors since <strong>the</strong> participants (i) have some proprioceptive access<br />
to <strong>the</strong> movement of <strong>the</strong>ir receptor field (in terms of horizontal mouse movement) <strong>and</strong><br />
<strong>the</strong> changing displacement of <strong>the</strong>ir body object (in terms of mouse clicks), but (ii) have<br />
no perceptual access to <strong>the</strong> current displacement of <strong>the</strong>ir own or <strong>the</strong> o<strong>the</strong>r‟s body object,<br />
<strong>and</strong> (iii) never<strong>the</strong>less have to engage in movements <strong>and</strong> regulatory behavior that results<br />
in bodily coordination. This situation is depicted in Figure 6-10.<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Proprioception<br />
(PI)<br />
Motor Act<br />
Figure 6-10. In <strong>the</strong> minimalist psychological study of body imitation designed by Lenay <strong>and</strong> colleagues<br />
we find an experimental situation where participants (i) have some proprioceptive access to <strong>the</strong> relative<br />
changes of position <strong>and</strong> displacement of <strong>the</strong>ir body object, but (ii) have no perceptual access to <strong>the</strong><br />
current displacement of <strong>the</strong>ir own or <strong>the</strong> o<strong>the</strong>r‟s body object, <strong>and</strong> (iii) never<strong>the</strong>less have to engage in<br />
coordinated movements that are appropriate in a shared context of interaction.<br />
Some preliminary studies by Lenay <strong>and</strong> his colleagues with this experimental setup<br />
have indicated that it is possible for participants to regulate <strong>the</strong>ir movements <strong>and</strong> body<br />
size so as to engage in stable perceptual crossing, <strong>and</strong> that <strong>the</strong>y even manage to improve<br />
<strong>the</strong>ir performance during <strong>the</strong> interaction. It appears that <strong>the</strong>y can use <strong>the</strong> overall stability<br />
of perceptual crossing as an indicator to appropriately change <strong>the</strong>ir respective body<br />
objects, even though <strong>the</strong>y do not know <strong>the</strong> actual current displacements. In o<strong>the</strong>r words,<br />
during <strong>the</strong> interaction process <strong>the</strong>y quickly learn how to left- <strong>and</strong> right-click so as to<br />
appropriately regulate <strong>the</strong>ir relative body displacement in a coordinated fashion, namely<br />
in relation to improvement of <strong>the</strong> overall stability of <strong>the</strong> interaction.<br />
101 | P a g e
Here we have a case that would be difficult to explain in terms of <strong>the</strong> three traditional<br />
factors alone. The role of both <strong>the</strong> Self- <strong>and</strong> O<strong>the</strong>r-BI has been reduced to a minimum in<br />
<strong>the</strong> virtual environment, <strong>and</strong> even <strong>the</strong> role of proprioception is limited: active clicking<br />
can inform <strong>the</strong> participants about <strong>the</strong> direction of changes to <strong>the</strong> distance between <strong>the</strong>ir<br />
receptor field <strong>and</strong> body object, but it gives <strong>the</strong>m no sense of <strong>the</strong> actual magnitude of<br />
displacement. In o<strong>the</strong>r words, <strong>the</strong> results of this experimental study indicate that nei<strong>the</strong>r<br />
a perceptually informed Self-BI nor an O<strong>the</strong>r-BI is a necessary condition for <strong>the</strong><br />
emergence of bodily coordination during interaction in a multi-agent system. It appears<br />
that <strong>the</strong> autonomous dynamics of <strong>the</strong> interaction process, with a little help of PI, can<br />
organize <strong>the</strong> individual gestures appropriately. Might this remaining PI, in <strong>the</strong> sense of<br />
information about relative change, be a necessary condition? Considering <strong>the</strong> case when<br />
Ian was gesturing under <strong>the</strong> blind, we can expect that this is not <strong>the</strong> case.<br />
6.2.8 Bodily coordination in a virtual space (II)<br />
We have no reached <strong>the</strong> final case listed in Table 6-1 (p. 82), whereby <strong>the</strong> potential role<br />
of all three traditional explanatory factors can be called into question. The possibility of<br />
bodily coordination under this extreme condition has not yet been investigated<br />
explicitly. We can imagine such an extreme situation occurring when Ian was under <strong>the</strong><br />
blind, if he happened to talk to someone out of view. In this case he would not only lack<br />
proprioception <strong>and</strong> a visually formed Self-BI, but also a visually formed O<strong>the</strong>r-BI.<br />
Thus, considering Ian‟s spontaneous gesturing under <strong>the</strong> blind, we can hypo<strong>the</strong>size that<br />
normal communicative gestures will occur even during such a non-visual interaction.<br />
Given that this kind of experiment has not been conducted with a deafferented subject,<br />
is <strong>the</strong>re a way to at least approximate this case? One possibility is to effectively retain<br />
<strong>the</strong> experimental setup described for <strong>the</strong> body imitation experiment described in <strong>the</strong><br />
previous section, but fur<strong>the</strong>r reduce <strong>the</strong> possible role of PI. Fortunately, precisely this<br />
has been done by Lenay <strong>and</strong> colleagues in a second set of experiments, in which <strong>the</strong><br />
participants were uninformed about <strong>the</strong> relationship between <strong>the</strong>ir left/right clicking<br />
activity <strong>and</strong> <strong>the</strong> relative changes to <strong>the</strong> body object displacement. In o<strong>the</strong>r words, not<br />
only did <strong>the</strong>y have no perceptual access to <strong>the</strong>ir virtual embodiment (Self-BI) or that of<br />
102 | P a g e
<strong>the</strong> o<strong>the</strong>r participant (O<strong>the</strong>r-BI), but <strong>the</strong>y also had no sense of how <strong>the</strong>ir clicking<br />
activities changed <strong>the</strong> manner of this embodiment in terms of <strong>the</strong> displacement between<br />
<strong>the</strong>ir receptor field <strong>and</strong> body object (PI). This extremely constrained situation is<br />
depicted schematically in Figure 6-11.<br />
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Proprioception<br />
(PI/PA)<br />
Motor Act<br />
Figure 6-11. In a second version of <strong>the</strong> minimalist psychological study of body imitation designed by<br />
Lenay <strong>and</strong> colleagues, <strong>the</strong> participants (i) have no „proprioceptive‟ access to <strong>the</strong> displacement of <strong>the</strong>ir<br />
body object, nor its relative changes due to <strong>the</strong>ir clicking activity, <strong>and</strong> (ii) have no „perceptual‟ access to<br />
<strong>the</strong> current displacement of <strong>the</strong>ir own or <strong>the</strong> o<strong>the</strong>r‟s body object, <strong>and</strong> (iii) never<strong>the</strong>less have to engage in<br />
coordinated movements that are appropriate in a shared context of interaction.<br />
As we would expect by now, <strong>the</strong> fact that <strong>the</strong> participants were unaware of <strong>the</strong> current<br />
extent of <strong>the</strong>ir virtual embodiment, <strong>and</strong> that <strong>the</strong>y had no knowledge of how <strong>the</strong>ir<br />
clicking behavior changed this displacement, did not prevent <strong>the</strong>m from collectively<br />
accomplishing <strong>the</strong> task. The relative stability of <strong>the</strong> interaction process in <strong>the</strong> form of<br />
perceptual crossing effectively enabled <strong>the</strong>m to organize <strong>the</strong>ir movement <strong>and</strong> clicking<br />
behavior appropriately, such that bodily coordination, in terms of a complementary<br />
displacement of body objects, emerged out of regulatory behavior informed by <strong>the</strong><br />
stability of ongoing mutual interaction. It <strong>the</strong>refore seems that not even a single<br />
traditional explanatory factor is necessary to account for this result. It seems that <strong>the</strong><br />
most important explanatory factor is an appeal to <strong>the</strong> interaction dynamics.<br />
103 | P a g e
We have now considered all possible combinations of <strong>the</strong> three traditional factors used<br />
to explain bodily coordination. The cases have indicated that <strong>the</strong> popular approach of<br />
positing a Self-BI <strong>and</strong>/or PI alongside an O<strong>the</strong>r-BI, related by some internal comparator<br />
module, needs to be called into question. More precisely, while <strong>the</strong> sufficiency of some<br />
combinations appears to be supported by <strong>the</strong> experimental data, any claims of necessity<br />
must be rejected on grounds of <strong>the</strong> competing cases. But what kind of <strong>the</strong>oretical<br />
framework can account for all of <strong>the</strong>se eight cases, considering (i) that no combination<br />
of factors remains constant throughout, <strong>and</strong> (ii) that <strong>the</strong> case of Ian Waterman <strong>and</strong> <strong>the</strong><br />
o<strong>the</strong>r psychological experiments demonstrate that <strong>the</strong> traditional factors are largely<br />
dispensable without significantly impairing bodily coordination?<br />
To be sure, we have already indicated that <strong>the</strong>re actually might be a common factor to<br />
all of <strong>the</strong>se different cases, namely <strong>the</strong> self-organizing dynamics of <strong>the</strong> interaction<br />
process. However, in order to properly motivate a closer analysis of this potential fourth<br />
factor, let us first consider a recent attempt to make sense of all this experimental data in<br />
terms of <strong>the</strong> „integrative motor <strong>the</strong>ory‟ proposed by Gallagher <strong>and</strong> colleagues.<br />
6.3 An integrative motor <strong>the</strong>ory<br />
Cole, Gallagher <strong>and</strong> McNeill use <strong>the</strong> case of Ian Waterman to argue against what <strong>the</strong>y<br />
call a motor <strong>the</strong>ory of gesture, which holds that “gesturing is primarily a matter of<br />
movement, falling within <strong>the</strong> domain of sensory-motor behavior. Gesture is <strong>the</strong> same<br />
sort of movement as instrumental or locomotive movement” (Cole, et al. 2002a, p. 53).<br />
On this view, gesture comes under <strong>the</strong> control of <strong>the</strong> body schema, <strong>and</strong> we would<br />
<strong>the</strong>refore expect Ian to consciously control <strong>and</strong> monitor his gesturing, just like his<br />
instrumental <strong>and</strong> locomotive movements. But if this does not turn out to be <strong>the</strong> case,<br />
<strong>the</strong>n <strong>the</strong> motor <strong>the</strong>ory of gesture may not provide <strong>the</strong> whole story <strong>and</strong> <strong>the</strong> case of Ian<br />
would establish “a clear differentiation between expressive movement <strong>and</strong> o<strong>the</strong>r kinds<br />
of movement (e.g. reflex, locomotive, <strong>and</strong> instrumental)” (Cole, et al. 2002a, p. 55).<br />
To test <strong>the</strong> prediction of <strong>the</strong> motor <strong>the</strong>ory, i.e. that Ian controlled his gestures using<br />
visual feedback, Cole et al. placed a blind before him in such a way as to block view of<br />
his h<strong>and</strong>s. If he was indeed using visual feedback to control his gesturing, like Ian<br />
104 | P a g e
himself appears to believe (cf. Cole, et al. 2002a, p. 57), <strong>the</strong>n this setup should prevent<br />
him performing appropriate gestures under this condition. He was <strong>the</strong>n asked to narrate<br />
<strong>the</strong> plot of a cartoon he had just seen.<br />
Once IW allowed his gestures to get under way, <strong>the</strong>y seemed to have a <strong>mind</strong> of<br />
<strong>the</strong>ir own. That is, <strong>the</strong>y did not seem to be under IW‟s attentional control, <strong>and</strong><br />
<strong>the</strong>y were consistent with normal measurements in terms of timing <strong>and</strong> shape,<br />
relative to IW‟s speech acts. An especially striking illustration of this came later<br />
in <strong>the</strong> experiment when IW, no longer recounting <strong>the</strong> cartoon story, continued to<br />
converse while his h<strong>and</strong>s remained under <strong>the</strong> blind. His h<strong>and</strong>s began to form<br />
gestures but did so outside of awareness. During <strong>the</strong> first 20 seconds of <strong>the</strong><br />
conversation he performed a string of gestures (14 in all) <strong>and</strong> <strong>the</strong>n said,<br />
revealingly “… <strong>and</strong> I‟m starting to use my h<strong>and</strong>s now …” while continuing to<br />
gesture. His gestures were utterly non-exceptional in timing <strong>and</strong> shape during<br />
<strong>the</strong> critical 20 seconds when <strong>the</strong>y were outside of awareness. (Cole, et al. 2002a,<br />
p. 56)<br />
In contrast to <strong>the</strong> expectations of <strong>the</strong> motor <strong>the</strong>ory of gesture, which predicts significant<br />
impairment under this condition, Ian‟s gesturing in fact appears to be close to normal. It<br />
should be conceded, however, that <strong>the</strong>ir spatial organization lacks some topokinetic<br />
precision, which indicates an impaired ability to move to a target position in space. This<br />
topokinetic precision, in combination with morphokinetic precision (i.e. <strong>the</strong> ability to<br />
shape h<strong>and</strong>s appropriately), is usually required for <strong>the</strong> performance of instrumental<br />
movements. But topokinetic precision is less important for gesture than morphokinetic<br />
precision. In o<strong>the</strong>r words, it could be argued that Ian‟s ability to gesture under <strong>the</strong> blind<br />
is aided by <strong>the</strong> fact that gesturing requires less topokinetic precision than instrumental<br />
action. Never<strong>the</strong>less, this concession still does not explain why he is able to gesture at<br />
all outside of awareness, given that this is simply impossible in <strong>the</strong> case of instrumental<br />
movement. Accordingly, Cole et al. suggest that Ian‟s gesturing “is much more<br />
integrated with linguistic behavior, <strong>and</strong> controlled by factors that go beyond ordinary<br />
sensory-motor control” (Cole, et al. 2002a, p. 58). What precisely are <strong>the</strong>se additional<br />
factors that go beyond motor <strong>the</strong>ory?<br />
105 | P a g e
The self-organizing intentionality of instrumental <strong>and</strong> locomotive movement<br />
normally depends on <strong>the</strong> implicit workings of body schemas. For IW, as we<br />
have seen, self-organizing motility breaks down. Yet, <strong>the</strong> self-organizing<br />
intentionality of language remains intact, <strong>and</strong> gesture, temporarily disrupted by<br />
IW‟s illness, has been re-established to a higher degree than his capacity for<br />
instrumental or locomotive movement. On <strong>the</strong> communicative <strong>the</strong>ory of gesture<br />
<strong>the</strong> reason gesture can be re-established with such proficiency is that gesture, as<br />
a movement concerned with <strong>the</strong> construction of significance ra<strong>the</strong>r than doing<br />
something, is organized primarily by <strong>the</strong> linguistic-communicative context.<br />
(Cole, et al. 2002a, p. 61)<br />
And, as <strong>the</strong>se experiments indicate, whereas <strong>the</strong> operation of body schemas requires<br />
feedback via proprioceptive information, <strong>the</strong> efficacy of <strong>the</strong> linguistic-communicative<br />
context requires feedback related to <strong>the</strong> social situation. At first sight this position<br />
<strong>the</strong>refore appears to be compatible with <strong>the</strong> enactive approach to social cognition,<br />
especially since it also appeals to <strong>the</strong> construction of significance <strong>and</strong> <strong>the</strong> role of <strong>the</strong><br />
social context. However, this apparent similarity hides some essential differences.<br />
Most importantly, for Cole, Gallagher <strong>and</strong> McNeill <strong>the</strong> case of Ian provides us with an<br />
important distinction between (i) instrumental action, <strong>and</strong> (ii) communicative action, “or<br />
action with meaning mapped onto it” <strong>and</strong> as such gesture “comes under <strong>the</strong> control of<br />
linguistic/communicative systems ra<strong>the</strong>r than <strong>the</strong> instrumental motor system” (Cole, et<br />
al. 2002, p. 61). The basis of this distinction is deeply problematic, <strong>and</strong> leads to tensions<br />
in <strong>the</strong> rest of <strong>the</strong>ir account. Let us consider <strong>the</strong>se consequences in more detail.<br />
First of all, note that Cole, Gallagher <strong>and</strong> McNeill draw <strong>the</strong>ir distinction between<br />
instrumental <strong>and</strong> communicative action by asserting that only <strong>the</strong> latter form of action<br />
expresses meaning (that is somehow „mapped onto it‟). This way of carving up <strong>the</strong><br />
space of actions goes strictly against <strong>the</strong> enactive approach, which holds (i) that all<br />
action is a form of sense-making, <strong>and</strong> (ii) that this sense is intrinsic to <strong>the</strong> performance<br />
of <strong>the</strong> action. Moreover, this enaction of meaning is not an internal or private event. On<br />
<strong>the</strong> contrary, <strong>the</strong> meaning of embodied action, as an expression of dynamics in relation<br />
to <strong>the</strong> world, is typically directly perceivable by o<strong>the</strong>rs (cf. Gallagher 2008b). Action in<br />
106 | P a g e
<strong>the</strong> world is characterized by a pre-reflective expression of intention, mood, attitude <strong>and</strong><br />
general state of being that does not require any additional communicative intent to be<br />
perceived as such 24 . It follows that instrumental action, <strong>and</strong> even locomotive action, is<br />
always more than mere physical displacement. If we were ever faced by movement that<br />
was not also expressive of meaning in some manner, it is unlikely that we would even<br />
perceive it as an action, instrumental or o<strong>the</strong>rwise.<br />
Second, by contrasting communicative action as an „action with meaning mapped onto<br />
it‟ Cole, Gallagher <strong>and</strong> McNeill lose sight of what makes social interaction special,<br />
namely that it is a shared form of action. To be fair, <strong>the</strong>y occasionally do hint at <strong>the</strong> role<br />
of <strong>the</strong> intersubjective context in generating gestures. For example, <strong>the</strong>y admit that “it is<br />
possible that <strong>the</strong> semantic <strong>and</strong> communicative (pragmatic) aspects of gesture provide<br />
sufficient feedback to sustain control of gestural movement” (Cole, et al. 2002a, p. 64),<br />
<strong>and</strong> <strong>the</strong>y finish <strong>the</strong>ir paper by stating that “it is, of course, ano<strong>the</strong>r person like myself<br />
who moves, motivates, <strong>and</strong> mediates this process. To say that language moves my body<br />
is already to say that o<strong>the</strong>r people move me” (ibid., p. 65). However, despite this appeal<br />
to <strong>the</strong> intersubjective context we already noted that descriptions of <strong>the</strong> experiment with<br />
Ian lacked any consideration of <strong>the</strong> role of his interlocutors. Ian could presumably see<br />
<strong>the</strong>ir gestures, if any were made, but even this has not been made explicit. Did <strong>the</strong>y<br />
mirror, imitate, or support one ano<strong>the</strong>r? Unfortunately, Cole, Gallagher <strong>and</strong> McNeill are<br />
silent on <strong>the</strong>se matters of social interaction <strong>and</strong> intersubjectivity.<br />
Third, Cole, Gallagher <strong>and</strong> McNeill‟s lingering methodological individualism becomes<br />
even more evident in <strong>the</strong> explanation given for Ian‟s gestural ability. Instead of fur<strong>the</strong>r<br />
exploring <strong>the</strong> tantalizing idea that „o<strong>the</strong>r people move me‟, an idea which presumably<br />
24 Consider this phenomenological observation, for instance: When I pass a young man on his way to <strong>the</strong><br />
park, I experience his relaxed attitude in terms of <strong>the</strong> ease of his gait, <strong>and</strong> by <strong>the</strong> way he causally swings<br />
his gym bag over <strong>the</strong> shoulder; I perceive happiness in <strong>the</strong> melodies he is humming to himself, <strong>and</strong> in <strong>the</strong><br />
satisfied smile of a good day‟s end; I see his intent to play a game of football in his attire, <strong>and</strong> in <strong>the</strong><br />
direction of his chosen path. In o<strong>the</strong>r words, his whole bodily presence exudes a purposeful, emotional<br />
<strong>and</strong> intentional being-in-<strong>the</strong>-world, a manner of being which is intersubjectively accessible to o<strong>the</strong>rs. In<br />
addition, this meaningful presence of <strong>the</strong> o<strong>the</strong>r subsumes <strong>and</strong> is accentuated by every little bodily action,<br />
even if <strong>the</strong>y can ultimately be abstracted as being merely instrumental or locomotive by an observer.<br />
107 | P a g e
involves <strong>the</strong> embodiment of sociality in a self-world-o<strong>the</strong>r structure, <strong>the</strong>y hypo<strong>the</strong>size<br />
that <strong>the</strong> source of Ian‟s ability may be localized in his brain‟s activity: “As IW speaks<br />
<strong>and</strong> gestures brain regions responsible for <strong>the</strong> generation of language may be<br />
contributing to control of gestural movement by enabling access to motor programmes<br />
that underpin his gesture stream” (ibid., p. 60). Of course, <strong>the</strong> enactive approach to<br />
social cognition would certainly not deny that some form of brain activity is involved in<br />
Ian‟s gesturing. Never<strong>the</strong>less, it would strongly insist that this neural component can<br />
only ever be part of a more encompassing explanation. We should not forget that <strong>the</strong><br />
performance of gesture, like behavior in general, can only be described as arising out of<br />
an integrated brain-body-world systemic whole (cf. Chiel & Beer 1997).<br />
Fourth, we can note that Cole, Gallagher <strong>and</strong> McNeill‟s hypo<strong>the</strong>tical neuro-centric<br />
explanation throws up an apparent riddle that is difficult to resolve from within <strong>the</strong>ir<br />
perspective. The problem is that “under <strong>the</strong> blind, <strong>and</strong> without proprioception, IW has<br />
no explicit, conscious knowledge of <strong>the</strong> specifics of his gestural production. In this case,<br />
what does he gain from his gestures?” (Cole, et al. 2002a, p. 62). However, while this<br />
might be a puzzling problem for a functionalist <strong>the</strong>ory of <strong>mind</strong>, it is not a worry for <strong>the</strong><br />
enactive paradigm for several reasons. To begin with, <strong>the</strong> enactive approach can appeal<br />
to careful phenomenological observation which shows that <strong>the</strong> entire body‟s movement<br />
is involved in <strong>the</strong> expression of <strong>the</strong> lived meaning of a situation. Accordingly, from this<br />
holistic perspective it should not be surprising that a subject‟s involvement in a social<br />
situation becomes embodied in what we describe as gesturing, even if that movement is<br />
not actually visible to o<strong>the</strong>rs.<br />
Fifth, if we assume that ano<strong>the</strong>r way to underst<strong>and</strong> <strong>the</strong> question of „gain‟ is in terms of<br />
<strong>the</strong> movement‟s „value‟, <strong>the</strong>n <strong>the</strong> enactive approach can appeal to <strong>the</strong> notion of sensemaking.<br />
In o<strong>the</strong>r words, <strong>the</strong> gesture might not only be an embodied expression of <strong>the</strong><br />
subject‟s involvement in a social situation, it might even be an essential component of<br />
<strong>the</strong> actual realization of <strong>the</strong> social meaning of that situation. This interpretation is<br />
fur<strong>the</strong>r supported by recent psychological research which has revealed that people with<br />
Möbius Syndrome (<strong>the</strong> congenital absence of pre-reflective facial expression) not only<br />
difficulty in expressing emotion, as might be expected, but also suffer from an impaired<br />
experience of emotion itself (cf. Cole 2009). Or, conversely, we can note that even <strong>the</strong><br />
108 | P a g e
eflective positioning of facial muscles into a smile can lead to a more elevated mood<br />
(an experience which <strong>the</strong> reader is encouraged to try out). This inseparability of <strong>the</strong><br />
meaning of a movement from <strong>the</strong> movement itself is precisely what <strong>the</strong> enactive notion<br />
of embodied action as sense-making is trying to capture.<br />
In contrast to this interpretation from <strong>the</strong> enactive paradigm, Cole, Gallagher <strong>and</strong><br />
McNeill offer what could be construed as essentially a cognitivist response. The focus<br />
of <strong>the</strong> explanation is again on <strong>the</strong> individual‟s brain, <strong>and</strong> <strong>the</strong> „gain‟ is cashed out in<br />
terms of improved cognitive performance. They propose that gesture, as an aspect of<br />
language ra<strong>the</strong>r than mere h<strong>and</strong> movement, might assist in <strong>the</strong> accomplishment of Ian‟s<br />
thought. However, since Ian does not receive any feedback from movement in <strong>the</strong> blind<br />
condition, <strong>the</strong>y conclude that <strong>the</strong> mechanism by which an individual might receive this<br />
cognitive benefit must be sought in <strong>the</strong> brain: “His cognitive gain from gesture without<br />
visual feedback, we suggest, may be due to pre-motor preparatory processes involved in<br />
<strong>the</strong> generation of <strong>the</strong> gestural movement ra<strong>the</strong>r than from <strong>the</strong> gestural movement itself”<br />
(Cole, et al. 2002a, p. 62). The fundamental contrast between <strong>the</strong> interpretation offered<br />
by <strong>the</strong> enactive paradigm <strong>and</strong> this functionalist explanation, which separates <strong>the</strong> „gain‟<br />
from <strong>the</strong> actual realization of <strong>the</strong> movement <strong>and</strong> instead localizes <strong>the</strong> source of efficacy<br />
in <strong>the</strong> brain, should by now be sufficiently clear.<br />
Sixth, we can note that it still remains mysterious as to precisely why Ian‟s gesture<br />
should have been “re-established to a higher degree than his capacity for instrumental or<br />
locomotive movement” (Cole, et al. 2002a, p. 61, emphasis added). What causes this<br />
relative difference in performance? Cole, Gallagher, <strong>and</strong> McNeill hint at a possible<br />
answer in that “gesture is never a mere motor phenomenon; it draws <strong>the</strong> body into a<br />
communicative order defined by its own pragmatic rules” (ibid., p. 65). We have seen<br />
that one possibility is to attribute this „communicative order‟ to a special region of <strong>the</strong><br />
brain. However, might it not be possible, as indicated by <strong>the</strong> experiments conducted by<br />
Lenay <strong>and</strong> colleagues, that this order is realized by <strong>the</strong> distributed organization of <strong>the</strong><br />
social situation itself? An important advantage of explaining Ian‟s relative improvement<br />
in relation to <strong>the</strong> scaffolding provided by an appropriate interaction process is that it<br />
better generalizes to o<strong>the</strong>r neuropathological cases. For example, it might also account<br />
for <strong>the</strong> finding that patients with apraxia who, although <strong>the</strong>y are not paralyzed, are<br />
109 | P a g e
unable to execute learned movements even if <strong>the</strong>y want to, but can act almost normally<br />
when situated in social settings. One such patient, for instance, was unable to move a<br />
cup-like object from <strong>the</strong> table to her face area on comm<strong>and</strong> in an experimental setting,<br />
but was able to serve <strong>and</strong> drink tea in a social setting that involved expressions of<br />
hospitality (cf. Marcel 1992; Gallagher & Marcel 1999). Can <strong>the</strong> ability of this apraxia<br />
patient also be explained in terms of an appeal to <strong>the</strong> supposed cognitive gain of a<br />
communicative gesture that as an action has „meaning mapped onto it‟, as Cole, et al.<br />
suggest for <strong>the</strong> case of Ian? While gesturing is certainly an important aspect of this<br />
setting (e.g. expressions of hospitality), never<strong>the</strong>less <strong>the</strong> act of drinking tea as such, for<br />
example, appears to primarily be an instrumental action. It would <strong>the</strong>refore seem to lack<br />
<strong>the</strong> potential cognitive gain that <strong>the</strong>y associated with communicative gesturing, <strong>and</strong><br />
<strong>the</strong>refore could not bootstrap <strong>the</strong> ability of drinking in social circumstances.<br />
Accordingly, it appears that Cole, Gallagher <strong>and</strong> McNeill‟s position is confronted by a<br />
choice between two options: (i) <strong>the</strong>y concede that instrumental actions can also be<br />
meaningful <strong>and</strong> expressive, or (ii) <strong>the</strong>y persist in claiming that instrumental behavior is<br />
meaningless movement. But each of <strong>the</strong>se two options comes at a price. If <strong>the</strong>y choose<br />
to pursue option (ii) <strong>the</strong> internal consistency of <strong>the</strong> integrative <strong>the</strong>ory of gesture can be<br />
retained, but <strong>the</strong> applicability of <strong>the</strong> <strong>the</strong>ory has been limited such that it can account for<br />
<strong>the</strong> case of Ian, but not for <strong>the</strong> case of apraxia. However, if <strong>the</strong>y choose option (i) <strong>the</strong>y<br />
will have to give up <strong>the</strong> very distinction between instrumental <strong>and</strong> gestural behavior that<br />
formed <strong>the</strong> basis for <strong>the</strong>ir original explanation, <strong>the</strong>reby leaving <strong>the</strong>m in no position to<br />
account for ei<strong>the</strong>r <strong>the</strong> case of Ian or that of <strong>the</strong> apraxia patient 25 .<br />
The enactive approach to social cognition, on <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>, manages to avoid this<br />
dilemma by treating all embodied action as inherently meaningful, as captured by <strong>the</strong><br />
notion of sense-making. Moreover, this meaning generation, as a relational phenomenon<br />
pertaining to <strong>the</strong> interaction between an agent <strong>and</strong> its world (including o<strong>the</strong>r agents), is<br />
inherently expressive. There is no need to appeal to some internal process of mapping<br />
25 To be fair, <strong>the</strong>ir original distinction could perhaps be saved by an attempt to systematically distinguish<br />
between those bodily actions that are part of language <strong>and</strong> those that are not. However, given <strong>the</strong> holistic<br />
manner in which our social presence is embodied, <strong>the</strong> validity of this distinction could be questioned.<br />
110 | P a g e
meaning onto physical movement into order to explain <strong>the</strong> phenomenon of gesture. In<br />
addition, this starting point of bodily subjectivity enables <strong>the</strong> enactive approach to take<br />
better into account <strong>the</strong> shared nature of social situations in which <strong>the</strong> interactors enable<br />
<strong>and</strong> constrain each o<strong>the</strong>r‟s behavior. Of course, <strong>the</strong> individual actor with its abilities <strong>and</strong><br />
limitations remains a crucial element in this account, but as an embodied <strong>and</strong> situated<br />
being-in-<strong>the</strong>-world it is also essentially a being-with-o<strong>the</strong>rs, an integrated part of <strong>the</strong><br />
unfolding dynamics of social interaction (cf. Heidegger 1927). With this shift toward a<br />
consideration of <strong>the</strong> constitutive role of sociality for <strong>the</strong> kind of agents that we are, it<br />
becomes possible to offload an important part of <strong>the</strong> explanatory burden from <strong>the</strong><br />
individual level to <strong>the</strong> social context in which that individual is embedded.<br />
6.4 Summary<br />
The critical analysis of <strong>the</strong> various psychological case studies <strong>and</strong> <strong>the</strong>ir potential<br />
explanations in terms of <strong>the</strong> traditional framework of social psychology, as well as <strong>the</strong><br />
engagement with <strong>the</strong> more recently developed integrative <strong>the</strong>ory of gesture, have shown<br />
that it is nei<strong>the</strong>r necessary nor sufficient to explain Ian Waterman‟s gesturing in terms of<br />
special brain regions. In fact, <strong>the</strong> analysis of <strong>the</strong> case studies points <strong>the</strong> o<strong>the</strong>r way: since<br />
we cannot appeal to any of <strong>the</strong> three traditional factors to explain Ian‟s ability to gesture<br />
normally under <strong>the</strong> blind without any proprioceptive or visual feedback, where does this<br />
leave <strong>the</strong> hypo<strong>the</strong>tical „Self/O<strong>the</strong>r Comparator‟ module which is traditionally postulated<br />
as an internal mechanism to determine intersubjective equivalence on <strong>the</strong> basis of <strong>the</strong>se<br />
factors? Could this supposed brain module itself also be removed as a necessary<br />
condition given that we can appeal to <strong>the</strong> organizing dynamics of <strong>the</strong> interaction process<br />
instead? This additionally reduced situation is depicted in Figure 6-12.<br />
111 | P a g e
Perception of Target<br />
Acts (O<strong>the</strong>r-BI)<br />
Self/O<strong>the</strong>r<br />
Comparator<br />
Perception of Actual<br />
Acts (Self-BI)<br />
Body Schema (BS)<br />
Proprioception<br />
(PI/PA)<br />
Motor Act<br />
Figure 6-12. A fully reduced schematic of <strong>the</strong> traditional explanatory paradigm. The fact that Ian can<br />
gesture under <strong>the</strong> blind without any visual or proprioceptive feedback eliminates <strong>the</strong> necessity for <strong>the</strong><br />
three factors of Self-BI, O<strong>the</strong>r-BI <strong>and</strong> PI/PA. Without <strong>the</strong>se factors <strong>the</strong> need for an internally localized<br />
„Self/O<strong>the</strong>r Comparator‟ module has been called into question as well.<br />
Note that we now have almost rehabilitated <strong>the</strong> original motor <strong>the</strong>ory of gesture, which<br />
did not posit any difference between instrumental <strong>and</strong> gestural action. However, as<br />
should have become clear in <strong>the</strong> discussion of <strong>the</strong> integrative <strong>the</strong>ory of gesture, we have<br />
recovered this motor <strong>the</strong>ory with an added enactive twist. As shown in Figure 6-12, in<br />
terms of <strong>the</strong> individual actor <strong>the</strong> core of <strong>the</strong> explanation is centered on <strong>the</strong> existence of a<br />
body schema <strong>and</strong> motor action. But <strong>the</strong> enactive approach leaves <strong>the</strong> motor <strong>the</strong>ory (<strong>and</strong><br />
with it Figure 6-12) behind in two essential aspects: (i) it holds that all embodied action<br />
is inherently meaningful, <strong>and</strong> directly perceivable by o<strong>the</strong>rs as such, because it is bodily<br />
expression of <strong>the</strong> sense-making that is enacted by an autonomous agent, <strong>and</strong> (ii) it holds<br />
that explanations of individual agency need to take into account <strong>the</strong> constitutive role of<br />
<strong>the</strong> interaction process in <strong>the</strong> organization of behaviors.<br />
In order to get a better grasp of <strong>the</strong>se two essential aspects of <strong>the</strong> enactive approach to<br />
social cognition <strong>the</strong> next part of this <strong>the</strong>sis will proceed as follows. First, we will use an<br />
evolutionary robotics approach to clarify <strong>the</strong> constitutive role of <strong>the</strong> interaction process<br />
in dynamical terms (Chapters 7 to 10). In particular, we will demonstrate that <strong>the</strong>re is no<br />
need to posit a Self/O<strong>the</strong>r Comparator module to account for bodily coordination, <strong>and</strong><br />
that <strong>the</strong> interaction process can organize such bodily coordination appropriately even<br />
112 | P a g e
under minimal conditions that resemble <strong>the</strong> situation depicted in Figure 6-12. These<br />
models are designed to help us better underst<strong>and</strong> what precisely is <strong>the</strong> form of such<br />
interaction dynamics, <strong>and</strong> how can we use <strong>the</strong>m to explain behavioral capacities of<br />
individual agents in a non-individualistic <strong>and</strong> non-mysterious manner.<br />
It is <strong>the</strong>n argued that this systemic approach alone, however, is not sufficient to support<br />
<strong>the</strong> claim that all embodied action is inherently expressive, as it leaves out <strong>the</strong><br />
qualitative aspects of intersubjectivity (Chapter 11). The systemic approach will thus be<br />
complemented by more detailed phenomenological considerations of <strong>the</strong> presence of<br />
o<strong>the</strong>rs (Chapter 12). Finally, <strong>the</strong> systemic <strong>and</strong> phenomenological insights are combined<br />
to develop a novel perspective from which it is possible to begin to probe <strong>the</strong> current<br />
consensus on cumulative cultural development by taking perceived meaning <strong>and</strong> multiagent<br />
interaction as <strong>the</strong> starting point for explaining <strong>the</strong> empirical data (Chapter 13).<br />
113 | P a g e
7 Toward <strong>the</strong> syn<strong>the</strong>sis of minimally social behavior<br />
In response to <strong>the</strong> growing challenge of methodological individualism Boden (2006b, p.<br />
54) asks: “How can we drop it without descending into mysterianism about social<br />
forces, group <strong>mind</strong>s, or personal interactions?” She notes that <strong>the</strong> social behavior of<br />
agents has become a practical issue in <strong>the</strong> field of artificial intelligence (AI), <strong>and</strong> cites<br />
Di Paolo‟s (2000, 1999) work in evolutionary robotics (ER) modeling as an example of<br />
how <strong>the</strong> solipsistic attitude may be rejected in a non-mysterian way. His models<br />
demonstrate how <strong>the</strong> unfolding of social behavior can depend on <strong>the</strong> dynamical<br />
properties of <strong>the</strong> interaction process itself. This general result has since been replicated<br />
by several o<strong>the</strong>r modeling studies, some of which are based on actual psychological<br />
experiments <strong>and</strong> have in turn cast new light on <strong>the</strong> interpretation of empirical data (e.g.<br />
Rohde 2008).<br />
In this chapter we will explain <strong>the</strong> methodology of ER in more detail, <strong>and</strong> <strong>the</strong>n propose<br />
an integrated approach which links ER models to empirical science by means of<br />
hypo<strong>the</strong>sis generation <strong>and</strong> verification. Since <strong>the</strong> next three chapters will be based on<br />
ER models that have been syn<strong>the</strong>sized in a similar fashion, unnecessary repetition is<br />
avoided by providing <strong>the</strong> common implementation details here.<br />
7.1 Evolutionary robotics<br />
There was a time – <strong>and</strong> for many <strong>the</strong>re still is – when doing computer science was no<br />
different from doing cognitive science, a tradition that is captured well by <strong>the</strong> slightly<br />
nostalgic slogan „Good Old-Fashioned AI‟ (Haugel<strong>and</strong> 1985). The computer metaphor<br />
of <strong>mind</strong> ensured, to <strong>the</strong> delight of programmers <strong>and</strong> engineers, that when something was<br />
learned about computers <strong>and</strong> algorithms, ipso facto something was learned about <strong>mind</strong><br />
<strong>and</strong> cognition as well (cf. Harnish 2002). To be sure, much of this research continues<br />
today under <strong>the</strong> label of Artificial Intelligence, but a closer look reveals that as a field it<br />
is not much different from Applied Informatics (cf. Russell & Norvig 2002).<br />
However, even though <strong>the</strong> computer metaphor of <strong>mind</strong> has outlived its usefulness for<br />
many cognitive scientists, <strong>the</strong> use of computers itself has not. Ironically, <strong>the</strong>y have<br />
114 | P a g e
proven to be indispensable tools for <strong>the</strong> development of a variety of alternative<br />
approaches to <strong>the</strong> study of <strong>life</strong> <strong>and</strong> <strong>mind</strong>, most of which explicitly oppose <strong>the</strong><br />
computationalist tradition. The boundaries between <strong>the</strong>se alternative approaches is quite<br />
fluid, but at least two general trends can be distinguished. On <strong>the</strong> one h<strong>and</strong>, <strong>the</strong>re is a<br />
focus on questions related to basic processes of <strong>the</strong> living, especially <strong>the</strong>ir autonomy<br />
<strong>and</strong> metabolic realization (Bourgine & Varela 1992). This tradition has been around for<br />
a while (e.g. Varela, et al. 1974), but it really started to blossom in <strong>the</strong> late 1980s when<br />
<strong>the</strong> name of <strong>the</strong> field “Artificial Life” (A<strong>life</strong>) was coined by Langton (1989). Since <strong>the</strong>n<br />
A<strong>life</strong> has generated a wide variety of research programs which are united by <strong>the</strong>ir<br />
interest in <strong>the</strong> framework of non-linear dynamics <strong>and</strong> <strong>the</strong> concepts of self-organization<br />
<strong>and</strong> emergence (cf. Langton 1995; Boden 1996).<br />
Alongside <strong>the</strong> establishment of A<strong>life</strong> as a more or less well defined field of research<br />
<strong>the</strong>re was a related transformation in robotics toward a biologically inspired<br />
consideration of embodiment <strong>and</strong> situatedness (cf. Brooks 1991a). This new robotics<br />
movement shared much of <strong>the</strong> general conceptual framework of A<strong>life</strong>, but it focused its<br />
efforts on underst<strong>and</strong>ing <strong>the</strong> emergence of embodied cognition (Steels 1994). Today <strong>the</strong><br />
field of such biologically inspired robotics has established itself as a viable alternative<br />
to more traditional engineering approaches (Pfeifer, et al. 2007). A particularly<br />
important breakthrough occurred when this approach was combined with <strong>the</strong> insights<br />
gained from research in evolutionary algorithms (Holl<strong>and</strong> 1975; Goldberg 1989). Since<br />
<strong>the</strong> early 1990s this combined approach is referred to as “Evolutionary Robotics” (Cliff,<br />
et al. 1993), though it also often involves simulations of artificial agents ra<strong>the</strong>r than<br />
actual hardware implementations 26 . This approach explicitly rejects <strong>the</strong> computer<br />
26 There is an unresolved tension lurking here: a core assumption of A<strong>life</strong> is that <strong>life</strong> is a result of <strong>the</strong><br />
systemic organization of matter ra<strong>the</strong>r than something that inheres in matter itself (Langton 1989). But<br />
o<strong>the</strong>rs argue that a physical realization is necessary for <strong>the</strong> phenomenon to be „real‟ (Steels 1994), or that<br />
it is at least needed for establishing a serious dialogue with <strong>the</strong> empirical sciences (Webb 2001). There<br />
might indeed be something specific about materiality that cannot be replicated in an abstract domain of<br />
pure logic (Ruiz-Mirazo & Moreno 2004), but for <strong>the</strong> study of behavioral dynamics a well-designed<br />
model is often sufficient (Beer 2003). The continuing appeal of actual physical robots for scientific use of<br />
<strong>the</strong> syn<strong>the</strong>tic method can perhaps also be attributed to <strong>the</strong> fact that <strong>the</strong> physical domain is where GOFAI<br />
was first outcompeted by alternative approaches (Brooks 1991b), a historic moment for <strong>the</strong> field.<br />
115 | P a g e
metaphor, <strong>and</strong> views cognition in terms of complex agent-environment interactions <strong>and</strong><br />
non-linear dynamics instead (Harvey 1996; Beer 1995a). The basic idea behind<br />
evolutionary robotics (ER) can be summarized as follows:<br />
An initial population of different artificial chromosomes, each encoding <strong>the</strong><br />
control system (<strong>and</strong> sometimes <strong>the</strong> morphology) of a robot, are r<strong>and</strong>omly created<br />
<strong>and</strong> put in <strong>the</strong> environment. Each robot (physical or simulated) is <strong>the</strong>n let free to<br />
act (move, look around, manipulate) according to a genetically specified<br />
controller while its performance on various tasks is automatically evaluated. The<br />
fittest robots are allowed to reproduce (sexually or asexually) by generating<br />
copies of <strong>the</strong>ir genotypes with <strong>the</strong> addition of changes introduced by some<br />
genetic operators (e.g., mutations, crossover duplication). This process is<br />
repeated for a number of generations until an individual is born which satisfies<br />
<strong>the</strong> performance criterion (fitness function) set by <strong>the</strong> experimenter. (Nolfi &<br />
Floreano 2000, p. 1)<br />
Since we will be making extensive use of ER in <strong>the</strong> following chapters, it is worth<br />
describing <strong>the</strong> methodology in more detail. First, it takes a holistic approach to<br />
behavior. This is in contrast to much traditional AI, which largely studied <strong>the</strong> internal<br />
operations of abstract systems <strong>and</strong> isolated components (Dennett 1978). ER, on <strong>the</strong><br />
o<strong>the</strong>r h<strong>and</strong>, is concerned with how adaptive behavior emerges out of <strong>the</strong> non-linear<br />
interactions of a brain, body <strong>and</strong> world systemic whole (Beer 1997). One simple way to<br />
achieve this is to embed a control system within a sensory-motor loop <strong>and</strong> place it<br />
within an environment (Cliff 1991). The aim of this “syn<strong>the</strong>tic ethology” (MacLennan<br />
1992) is to combine <strong>the</strong> simplicity <strong>and</strong> control of behaviorist methods with <strong>the</strong><br />
ecological <strong>and</strong> contextual validity of empirical ethology.<br />
This holistic approach to behavior is complemented by an evolutionary approach to its<br />
optimization. In contrast to traditional AI, which mostly specified a system‟s cognitive<br />
functions programmatically, ER tries to take <strong>the</strong> human designer out of <strong>the</strong> loop as<br />
much as possible. To be sure, it is still necessary to specify what defines an agent, its<br />
environment <strong>and</strong> <strong>the</strong> desired behavior, but <strong>the</strong> particular way in which this behavior is<br />
realized depends on an evolutionary process. In this way we only have a minimal <strong>and</strong><br />
116 | P a g e
controllable impact of design assumptions, <strong>and</strong> it is easier to investigate <strong>the</strong> minimal<br />
conditions for a behavioral capacity (Harvey, et al. 2005). Often <strong>the</strong> evolutionary<br />
process leads to novel <strong>and</strong> surprising mechanisms that undermine our preconceptions<br />
about <strong>the</strong> necessary conditions for a certain behavior to emerge.<br />
Finally, <strong>the</strong> holistic approach to behavior <strong>and</strong> <strong>the</strong> evolutionary approach to its<br />
optimization are complemented by a dynamical approach to its realization (Beer 1997).<br />
Whereas <strong>the</strong> control systems for traditional AI are typically implemented in terms of<br />
symbolic representations that are specified by <strong>the</strong> designer, ER tries to minimize <strong>the</strong><br />
impact of prior assumptions about what internal operations might be necessary. The aim<br />
is to provide a generic substrate for <strong>the</strong> controller which can <strong>the</strong>n be shaped by <strong>the</strong><br />
selective pressures of <strong>the</strong> evolutionary algorithm. One popular way of implementing this<br />
generic substrate is in terms of continuous-time dynamical systems (Beer 1995b). An<br />
advantage of using such systems is that <strong>the</strong> autonomous system is no „back box‟; it is<br />
possible to use dynamical systems <strong>the</strong>ory to underst<strong>and</strong> <strong>and</strong> formalize <strong>the</strong> system‟s<br />
behavior (Beer 1997). Moreover, this <strong>the</strong>ory is especially attractive in relation to a<br />
holistic approach to behavior because it can deal with changes in behavior in a unified<br />
ma<strong>the</strong>matical manner that spans brain, body <strong>and</strong> world (Kelso 1995) as well as various<br />
temporal scales (Thelen & Smith 1994).<br />
We can also identify three broad contexts in which <strong>the</strong> ER methodology (<strong>and</strong> A<strong>life</strong><br />
more generally) is used. First, it is by <strong>and</strong> large <strong>the</strong> case that <strong>the</strong> sciences of <strong>the</strong> artificial<br />
are part of <strong>the</strong>oretical science. To be sure, many have succumbed to <strong>the</strong> functionalist<br />
temptation to view <strong>the</strong>ir artificial systems as actual empirical instances of <strong>the</strong><br />
phenomena <strong>the</strong>y are investigating, especially in <strong>the</strong> early years of <strong>the</strong> field (e.g. Langton<br />
1989). However, <strong>the</strong>re is a growing consensus that this confuses <strong>the</strong> model with what is<br />
being modeled. Note that this does not diminish <strong>the</strong> scientific value of <strong>the</strong> syn<strong>the</strong>tic<br />
approach, but shifts its emphasis toward creating “opaque thought experiments” by<br />
which it is possible to systematically explore <strong>the</strong> consequences of a <strong>the</strong>oretical position<br />
(Di Paolo, et al. 2000). The idea is that we use <strong>the</strong>ories about <strong>the</strong> empirical world to<br />
inform <strong>the</strong> design of ER models, <strong>and</strong> that <strong>the</strong>se models in turn constrain <strong>the</strong><br />
interpretation of <strong>the</strong> <strong>the</strong>ories (Moreno 2002). This can happen negatively, such as when<br />
<strong>the</strong> ER methodology is used as a subversive tool to undermine <strong>the</strong>oretical claims for<br />
117 | P a g e
necessity, but also positively, as when it is used to syn<strong>the</strong>size a model that serves as a<br />
proof of concept (Harvey, et al. 2005).<br />
What makes ER attractive to science, namely its capacity for <strong>the</strong> syn<strong>the</strong>sis of systematic<br />
thought experiments of indefinite complexity, also aligns it with <strong>the</strong> aims of analytic<br />
philosophy (Dennett 1994). Broadly speaking, we can capture this aspect of ER with <strong>the</strong><br />
slogan “philosophy of <strong>mind</strong> with a screwdriver” (Harvey 2000). It is not always <strong>the</strong> aim<br />
to explicitly model one‟s philosophical assumptions in <strong>the</strong> process of syn<strong>the</strong>sizing an<br />
artificial system, but in practice it is difficult – if not impossible – to avoid doing so at<br />
least implicitly. The fact that <strong>the</strong> systems we create embody our presuppositions has<br />
been exploited with great effect by <strong>Dr</strong>eyfus, who traces <strong>the</strong> limited success of traditional<br />
AI to its underlying Cartesian philosophy (<strong>Dr</strong>eyfus & <strong>Dr</strong>eyfus 1988). Moreover, it has<br />
been argued that <strong>the</strong> subsequent turn toward embodied-embedded AI coincides with a<br />
shift to a more Heideggerian philosophy (Wheeler 2005). The advantage of probing<br />
philosophical positions with ER ra<strong>the</strong>r than with traditional thought experiments are <strong>the</strong><br />
increased capacity to deal with complex systems, as well as to test <strong>the</strong>m in more<br />
realistic, but still fully controllable settings.<br />
Finally, an important but often underappreciated aspect is <strong>the</strong> syn<strong>the</strong>tic methodology‟s<br />
pedagogical value. It is quite a formative experience to spend countless hours in front of<br />
<strong>the</strong> computer trying to get an artificial agent to solve what should be a simple task, but<br />
getting nothing but senseless behavior (cf. Dennett 1984). It quickly becomes clear that<br />
<strong>the</strong>se systems do not know what <strong>the</strong>y are doing; <strong>the</strong>y have no underst<strong>and</strong>ing of <strong>the</strong>ir<br />
situation (Haugel<strong>and</strong> 1997) nor do <strong>the</strong>y care about <strong>the</strong> fact that <strong>the</strong>y don‟t (Di Paolo<br />
2003). Similarly, it is a humbling experience to consistently have your own <strong>and</strong> o<strong>the</strong>rs‟<br />
cherished presuppositions <strong>and</strong> expectations undermined by an opportunistic<br />
evolutionary algorithm. Over time this subversive process starts to affect <strong>the</strong> way in<br />
which you approach problems, exp<strong>and</strong>ing <strong>the</strong> range of possible explanations to be<br />
considered, while at <strong>the</strong> same time teaching you to be careful about positing necessary<br />
<strong>and</strong> sufficient conditions.<br />
In order to illustrate <strong>the</strong> effectiveness of <strong>the</strong> ER methodology it is helpful to mention a<br />
few concrete examples. While ER has been successful at syn<strong>the</strong>sizing artificial agents<br />
118 | P a g e
that are capable of engaging <strong>the</strong>ir environment in a robust, timely <strong>and</strong> adaptive manner,<br />
<strong>the</strong>re has been some debate about <strong>the</strong> internal mechanisms that are necessary for <strong>the</strong>se<br />
agents to switch between qualitatively different behaviors depending on situational<br />
changes. In o<strong>the</strong>r words, it has been demonstrated that <strong>the</strong> „intra-context frame problem‟<br />
can be resolved, but a solution to <strong>the</strong> „inter-context frame problem‟ arguably requires a<br />
different kind of mechanism (Wheeler 2008). In response to this debate it is possible to<br />
cite a recent ER study by Izquierdo <strong>and</strong> Buhrmann (2008), where a single dynamical<br />
system was optimized to perform two qualitatively different behaviors, chemotaxis <strong>and</strong><br />
legged locomotion, without providing a priori structural modules, explicit learning<br />
mechanisms, or an external signal for when to switch between <strong>the</strong>m. The agent‟s ability<br />
to switch its behavior appropriately when placed from one situation into ano<strong>the</strong>r is<br />
explained in terms of <strong>the</strong> interactions between <strong>the</strong> controller‟s dynamics, its body <strong>and</strong><br />
environment, <strong>the</strong>reby calling into question <strong>the</strong> internalist assumption that <strong>the</strong> necessary<br />
<strong>and</strong> sufficient conditions for context-switching behavior must reside in <strong>the</strong> individual<br />
alone.<br />
In fact, <strong>the</strong>re are many examples of ER models that teach us to be careful about what<br />
internal conditions we presuppose on <strong>the</strong> basis of observed behavior, <strong>and</strong> vice versa.<br />
Consider, for instance, <strong>the</strong> common assumption that some form of neural plasticity is a<br />
necessary condition for learning, an assumption which has come under attack by a<br />
number of ER studies. It has been shown that an embodied agent that is controlled by a<br />
continuous-time recurrent neural network without synaptic plasticity (i.e. connection<br />
weights remain fixed during a trial) nor any o<strong>the</strong>r a priori modular structures, can<br />
perform a continuous associative learning task (Izquierdo, et al. 2008). One simple<br />
solution to this problem is that <strong>the</strong> continuous to-be-remembered signal is simply<br />
associated with <strong>the</strong> activity of a network component that has a slower timescale.<br />
However, while this is an instance where it is possible to match a specific behavioral<br />
property to a localized internal component, such structural isomorphism is itself not a<br />
necessary condition. Buckley <strong>and</strong> colleagues (2008), for example, have shown that <strong>the</strong><br />
capacity to solve a task dem<strong>and</strong>ing multiple behavioral modes does not directly say<br />
anything about <strong>the</strong> complexity of <strong>the</strong> attractor structure of <strong>the</strong> internal dynamics.<br />
Contrary to common intuition, <strong>the</strong> agents were able to satisfy <strong>the</strong> task with only<br />
119 | P a g e
transient dynamics around a single fixed point attractor, <strong>the</strong>reby demonstrating that it is<br />
not necessary for a distinct behavioral mode to be associated with a distinct attractor. In<br />
fact, fur<strong>the</strong>r doubt has been cast on how much can be understood about <strong>the</strong> limitations<br />
of an agent‟s behavior from <strong>the</strong> limitations on its internal dynamics A nice illustration<br />
of this idea is an ER study which demonstrates that even a purely reactive (stateless)<br />
system, i.e. a system whose outputs are at each moment only determined by its current<br />
inputs, can engage in non-reactive behavior due to <strong>the</strong> ongoing history of interaction<br />
resulting from its situatedness (Izquierdo-Torres & Di Paolo 2005). It is <strong>the</strong>refore<br />
conceivable that a natural agent‟s behavior that appears to depend on some form of state<br />
may actually depend on a relational ra<strong>the</strong>r than an internal form of state. This work<br />
reinforces <strong>the</strong> idea that embodied behavior can exhibit properties that cannot be deduced<br />
directly from those of <strong>the</strong> individual‟s internal milieu itself.<br />
7.2 An integrative methodology<br />
Since its beginnings in <strong>the</strong> early 1990s ER has established itself as a viable<br />
methodology for syn<strong>the</strong>sizing models of what has become known as „minimally<br />
cognitive behavior‟, namely <strong>the</strong> simplest behavior that raises issues of genuine cognitive<br />
interest (Beer 1996). We have described some general illustrative examples of this<br />
methodology in <strong>the</strong> previous section. Within <strong>the</strong> context of this research framework<br />
<strong>the</strong>re has also been a growing interest in using this syn<strong>the</strong>tic method in order to<br />
investigate <strong>the</strong> interaction dynamics of social cognition (e.g. Williams, et al. 2008; Di<br />
Paolo, et al. 2008; <strong>Froese</strong> & Di Paolo 2008a; Ikegami & Iizuka 2007; Iizuka & Di Paolo<br />
2007b; Iizuka & Ikegami 2004a; Quinn 2001; Di Paolo 2000; 1999). As a specialization<br />
of <strong>the</strong> ER methodology, we can conceptualize this research as a <strong>the</strong>oretical investigation<br />
into <strong>the</strong> dynamics of „minimally social behavior‟ (<strong>Froese</strong> & Di Paolo in press-b).<br />
What is especially interesting about some of <strong>the</strong>se recent advances in ER is that <strong>the</strong><br />
syn<strong>the</strong>tic method has been used to create models which are explicitly inspired by<br />
psychological experiments. Moreover, some of <strong>the</strong>se models have been specifically<br />
designed to generate insights that have <strong>the</strong> potential to become <strong>the</strong> starting point for<br />
mutually informing collaborations between ER <strong>and</strong> <strong>the</strong> traditional empirical sciences,<br />
especially social psychology (cf. Di Paolo, et al. 2008; Rohde 2008). Here we will<br />
120 | P a g e
continue this effort to move ER into a more productive relationship with <strong>the</strong> rest of<br />
cognitive science. The crucial step of moving ER beyond mere technological wizardry<br />
or model collection <strong>and</strong> into a principled scientific research program is to link ER <strong>and</strong><br />
science toge<strong>the</strong>r in terms of hypo<strong>the</strong>sis generation <strong>and</strong> verification. This integrative<br />
methodology consists of four essential steps:<br />
(i) Syn<strong>the</strong>sis of model: The first step is generally <strong>the</strong> identification of an interesting<br />
empirical experiment whose <strong>the</strong>oretical interpretation could benefit from a modeling<br />
study. This might be <strong>the</strong> case for a variety of reasons. For example, it could be that<br />
<strong>the</strong> original interpretation of <strong>the</strong> experiment is incomplete <strong>and</strong> that a more detailed<br />
dynamical analysis is desirable, or that <strong>the</strong> given explanation posits some potentially<br />
unnecessary conditions of necessity that might have been introduced due to hidden<br />
philosophical presuppositions. Ano<strong>the</strong>r motivation could be to test <strong>the</strong> viability of a<br />
novel hypo<strong>the</strong>sis without having to replicate an entire empirical study.<br />
(ii) Emergence of behavior: How <strong>the</strong> target behavior is realized is not pre-specified by<br />
<strong>the</strong> experimenter. The behavior is an emergent phenomenon that depends on <strong>the</strong><br />
particular history of agent-environment interactions that is realized by <strong>the</strong> simulation<br />
syn<strong>the</strong>sized in step (i). As such, it cannot be found as a static element within <strong>the</strong><br />
program but must be observed as a pattern of activity.<br />
(iii) Analysis of behavior: The behavioral phenomena that emerge in step (ii) are<br />
essentially opaque, especially if <strong>the</strong> agent-environment system is characterized by<br />
complex non-linear dynamics. In o<strong>the</strong>r words, <strong>the</strong> observed behavior is typically in<br />
need of fur<strong>the</strong>r systematic analysis to determine its essential structures <strong>and</strong><br />
conditions of possibility. This typically takes <strong>the</strong> form of behavioral psychophysical<br />
tests, lesion studies, <strong>and</strong> formal dynamical analysis.<br />
(iv) Generation of hypo<strong>the</strong>ses: The insights gained in step (iii) form <strong>the</strong> basis for a<br />
<strong>the</strong>oretical response in relation to <strong>the</strong> study‟s original motivation. They also inform<br />
<strong>the</strong> process of generating novel hypo<strong>the</strong>ses, which <strong>the</strong>n become <strong>the</strong> basis for <strong>the</strong><br />
design of novel simulations for step (i).<br />
121 | P a g e
The relationship between <strong>the</strong> syn<strong>the</strong>tic, emergent, analytic, <strong>and</strong> generative aspects of <strong>the</strong><br />
ER methodology are illustrated in Figure 7-1.<br />
Generation of hypo<strong>the</strong>ses<br />
Analysis of behavior<br />
Computer<br />
Science<br />
Evolutionary<br />
Robotics<br />
Model<br />
Phenomenon<br />
Syn<strong>the</strong>sis of model<br />
Emergence of behavior<br />
Figure 7-1. Illustration of <strong>the</strong> key steps involved when using evolutionary robotics as a tool for cognitive<br />
science. The methodological circle typically starts with a <strong>the</strong>oretical, empirical or simply exploratory<br />
motivation that leads to (i) <strong>the</strong> syn<strong>the</strong>sis of a new simulation model, which when run gives rise to (ii) <strong>the</strong><br />
emergence of model behavior, whose complex non-linear realization necessitates (iii) a behavioral <strong>and</strong><br />
dynamical analysis of that behavioral phenomenon. Finally, <strong>the</strong> insights gained lead to (iv) <strong>the</strong> generation<br />
of novel hypo<strong>the</strong>sis, <strong>and</strong> <strong>the</strong> circle can start again.<br />
Note that steps (i)-(ii) <strong>and</strong> (ii)-(iii) already exist in current ER work (<strong>and</strong> in A<strong>life</strong> more<br />
generally). It is step (iv) which crucially turns <strong>the</strong>se disparate elements into a coherent<br />
scientific research program. Actually, a full-blown ER-based scientific study should<br />
ideally include an empirical element in this methodological circle so that it consists of<br />
two distinct phases: (i) An ER model, which can be novel or based on an existing<br />
empirical experiment, is used to generate a novel hypo<strong>the</strong>sis, <strong>and</strong> (ii) this hypo<strong>the</strong>sis<br />
<strong>the</strong>n becomes <strong>the</strong> basis for an empirical experiment to verify its validity. It is also<br />
possible for ER to be even more involved with empirical sciences: before <strong>the</strong><br />
psychological experiment it can be used as a means of testing experimental designs <strong>and</strong><br />
running simple pilot studies, <strong>and</strong> afterwards it is helpful for interpreting ambiguous<br />
empirical data (cf. Rohde & Di Paolo 2007; 2008). However, this ideal situation where<br />
122 | P a g e
ER <strong>and</strong> empirical science go h<strong>and</strong>-in-h<strong>and</strong> all <strong>the</strong> way is a dem<strong>and</strong>ing interdisciplinary<br />
endeavor that takes considerable effort to realize in practice. Never<strong>the</strong>less, one<br />
promising approach is to address work in psychology that already makes use of minimal<br />
technological interfaces as part of its experimental design (Rohde 2008).<br />
Previous work in ER on <strong>the</strong> topic of social cognition has shown that simple models can<br />
be used to highlight <strong>the</strong> important role of <strong>the</strong> interaction process itself for <strong>the</strong><br />
appropriate unfolding of inter-individual coordination behavior. These results support<br />
<strong>the</strong> “interaction <strong>the</strong>ory” of social cognition which holds that primary intersubjectivity<br />
(Trevar<strong>the</strong>n 1979), i.e. an embodied intersubjective interaction that is best understood in<br />
terms of enactive perception, is <strong>the</strong> foundation of our everyday social abilities (e.g.<br />
Gallagher 2008d; 2001). This explicit recognition of <strong>the</strong> essential role of <strong>the</strong> interaction<br />
process for social cognition provides a much needed challenge to <strong>the</strong> classical<br />
cognitivist „problem of o<strong>the</strong>r <strong>mind</strong>s,‟ which is traditionally solved by postulating some<br />
kind of „<strong>mind</strong>-reading‟ ability, ei<strong>the</strong>r in terms of <strong>the</strong>oretical inference <strong>and</strong> reasoning<br />
(e.g. Carru<strong>the</strong>rs 1996), or in <strong>the</strong> form of internal simulation (e.g. Gallese & Goldman<br />
1998).<br />
However, an explicit recognition of <strong>the</strong> role of <strong>the</strong> interaction process itself for social<br />
cognition should only be seen as <strong>the</strong> beginning, as pointing out a new problem space<br />
that dem<strong>and</strong>s to be fur<strong>the</strong>r investigated. Indeed, what is needed is a much more<br />
extensive reevaluation of <strong>the</strong> constitutive relationship between individual agency, social<br />
interactions <strong>and</strong> societal context (De Jaegher & <strong>Froese</strong> 2009). The enactive approach<br />
attempts to go beyond mere recognition of <strong>the</strong> importance of <strong>the</strong> interaction process to<br />
assessment of its constitutive role for <strong>the</strong> unfolding of social cognition (De Jaegher<br />
2009). In particular, it has been proposed that <strong>the</strong> interaction process itself can take on a<br />
form of constitutive autonomy, i.e. that multi-agent interactions can organize into a<br />
dynamical system that maintains its own identity (cf. Chapter 4, p. 58). It has also been<br />
argued that when <strong>the</strong> individual activities of cognitive agents become coupled in this<br />
kind of manner, previously inaccessible domains of co-regulated cognition can become<br />
available in <strong>the</strong> form of participatory sense-making (cf. Chapter 4, p. 64). And <strong>the</strong><br />
effects of <strong>the</strong> social interaction process do not remain limited to <strong>the</strong> cognitive domain,<br />
but constitute <strong>the</strong> embodied <strong>mind</strong> intersubjectively at even <strong>the</strong> most fundamental levels;<br />
123 | P a g e
a process of “self-o<strong>the</strong>r co-determination” (Thompson 2001). The ER experiments<br />
presented in Chapters 8-10 will illustrate <strong>the</strong>se conceptual <strong>and</strong> methodological ideas in<br />
more concrete terms.<br />
7.3 Implementation details<br />
After this general introduction to <strong>the</strong> ER methodology, <strong>and</strong> a brief assessment of its<br />
applicability to <strong>the</strong> problem of investigating <strong>the</strong> dynamics of social cognition, it is<br />
necessary to describe in more detail <strong>the</strong> way in which <strong>the</strong> models presented in <strong>the</strong><br />
following chapters have been implemented as computer simulations. These details do<br />
not vary significantly between <strong>the</strong> experiments so <strong>the</strong>y are only described once here. If<br />
<strong>the</strong>re are any specific differences pertaining to a particular experiment, <strong>the</strong>se will be<br />
noted explicitly in <strong>the</strong> text.<br />
All simulated agents are controlled by two structurally identical continuous-time<br />
recurrent neural networks (CTRNNs), as described by Beer (1995b). They were chosen<br />
to be clones because work by Iizuka <strong>and</strong> Ikegami (2004a) on a related task suggests that<br />
genetically similar agents are potentially better at coordination. The agents face each<br />
o<strong>the</strong>r in an unlimited continuous 1-D space (i.e. one agent faces „up‟ <strong>and</strong> one agent<br />
faces „down‟). Distance <strong>and</strong> time units are of an arbitrary scale. Each agent can only<br />
move horizontally, <strong>and</strong> sense only via a binary receptor field. The field is activated (set<br />
to 1) when <strong>the</strong> agents cross an object in <strong>the</strong>ir environment, o<strong>the</strong>rwise it is set to 0. The<br />
location of each agent is represented by a continuous variable, <strong>and</strong> <strong>the</strong> velocity is<br />
controlled by a fully inter-connected CTRNN with self-connections. No symmetry is<br />
imposed on <strong>the</strong> network structure. The time evolution of node activation y i is<br />
determined by Equation 7-1.<br />
Equation 7-1<br />
i<br />
i<br />
i<br />
N<br />
x bi<br />
y y w z ( y ) I , z ( x)<br />
1/(1 e )<br />
j 1<br />
ji<br />
j<br />
j<br />
i<br />
i<br />
In this equation y i represents <strong>the</strong> activation of node i, z i is <strong>the</strong> node output as calculated<br />
by <strong>the</strong> st<strong>and</strong>ard sigmoid function, τ i is its time constant, b i is a bias term, <strong>and</strong> w ji is <strong>the</strong><br />
124 | P a g e
strength of <strong>the</strong> connection from <strong>the</strong> node j to i. Each node i receives a weighted input I i<br />
from <strong>the</strong> agent‟s receptor field, which is calculated according to Equation 7-2.<br />
Equation 7-2 I w Sense(RF)<br />
i<br />
i<br />
The function Sense(RF) returns <strong>the</strong> current state of <strong>the</strong> agent‟s receptor field (0 or 1).<br />
This receptor state is connected to each CTRNN node i via a dedicated input weight<br />
matrix w I . While such a distributed network structure is more complex than having a<br />
dedicated „input‟ node, some initial exploratory evolutionary runs revealed that this<br />
more distributed way of perturbing <strong>the</strong> state of <strong>the</strong> system resulted in solutions that were<br />
more easily optimized (i.e. increased „evolvability‟). The ranges of <strong>the</strong>se parameters<br />
vary between experiments, <strong>and</strong> are described in <strong>the</strong> chapters.<br />
The behavior of <strong>the</strong> agents is optimized by using a genetic algorithm (GA) which is<br />
based on <strong>the</strong> “microbial GA”, a steady-state GA with tournament selection (cf. Harvey<br />
2001). Until some termination criterion is reached, two members of <strong>the</strong> population are<br />
chosen at r<strong>and</strong>om, both have <strong>the</strong>ir fitness evaluated, <strong>and</strong> while <strong>the</strong> „winner‟ of <strong>the</strong><br />
tournament remains unchanged in <strong>the</strong> population, <strong>the</strong> „loser‟ is replaced by a slightly<br />
mutated copy of <strong>the</strong> „winner‟. No crossover operator was used, especially as this might<br />
conflict with <strong>the</strong> increased genetic diversity entailed by niching (<strong>Froese</strong> & Spier 2008).<br />
Each member of <strong>the</strong> population is a clonal pair of agents whose overall performance<br />
will be tested in a given experimental setup. In this GA we define a generation as <strong>the</strong><br />
number of tournaments required to generate a number of offspring equal to <strong>the</strong><br />
population size. An evolutionary run finishes at some maximum number of generations,<br />
though it is sometimes manually terminated before <strong>the</strong> maximum is reached if solutions<br />
are sufficiently good.<br />
Niching. In order to enhance <strong>the</strong> evolvability of <strong>the</strong> CTRNNs, <strong>the</strong> st<strong>and</strong>ard microbial<br />
GA has been extended with a simple „geographical‟ method to allow different<br />
subpopulations to evolve semi-independently within <strong>the</strong> overall population (cf. Spector<br />
& Klein 2006; Izquierdo, et al. 2008). A minimalist, 1-D wrap-around „geography‟ was<br />
used. This was implemented by running 10 evolutionary runs in parallel, each with a<br />
population size of 10 solutions. Thus, for a particular evolutionary run ER i , after every<br />
125 | P a g e
generation, two solutions get selected at r<strong>and</strong>om to compete against <strong>the</strong> best, most<br />
recent solutions of <strong>the</strong> two neighboring subpopulations that are being evolved by runs<br />
ER i-1 <strong>and</strong> ER i+1 . The winner of each competition remains in <strong>the</strong> subpopulation of ER i for<br />
<strong>the</strong> next generation. The rest of <strong>the</strong> next generation is determined by tournament<br />
competition within ER i .<br />
Genetic encoding. All CTRNN parameters <strong>and</strong> gains are encoded by a real-valued<br />
vector (gene range [0, 1]). At <strong>the</strong> start of <strong>the</strong> GA <strong>the</strong> gene vector is initialized with<br />
r<strong>and</strong>om values drawn from a uniform distribution (range [0, 1]). The mutation operator<br />
changes each gene by a r<strong>and</strong>om value drawn from a Gaussian distribution (μ = 0; σ 2 =<br />
0.02) with reflection at <strong>the</strong> gene boundaries. Before every fitness evaluation, each gene<br />
is decoded linearly to <strong>the</strong> corresponding parameter range, except for <strong>the</strong> gains <strong>and</strong> time<br />
constants, which are exponentially scaled.<br />
Evaluation function. When <strong>the</strong> desirability of a solution is evaluated, it is tested for a<br />
number of trials, typically within <strong>the</strong> range of 15 <strong>and</strong> 100 (<strong>the</strong> precise number is<br />
specified in <strong>the</strong> chapters). A relatively large number of trials can be beneficial for <strong>the</strong><br />
evolutionary process because <strong>the</strong> behavior of <strong>the</strong> CTRNN solutions is highly susceptible<br />
to initial conditions, e.g. <strong>the</strong> respective starting positions of <strong>the</strong> agents. Each trial run<br />
consists of 800 units of time, unless specified o<strong>the</strong>rwise. At <strong>the</strong> start of each trial, <strong>the</strong><br />
agents have <strong>the</strong>ir internal node activations set to 0.<br />
Fitness weighting. Even with <strong>the</strong> increased genetic diversity due to „geographical‟<br />
niching, <strong>and</strong> <strong>the</strong> r<strong>and</strong>om spread of <strong>the</strong> initial conditions, it is still <strong>the</strong> case that<br />
evolutionary runs are highly susceptible to get stuck in local optima of <strong>the</strong> search space,<br />
especially because <strong>the</strong> difficulty associated with <strong>the</strong> starting positions is highly variable.<br />
For example, agents evolved to engage in perceptual crossing can easily locate each<br />
o<strong>the</strong>r when <strong>the</strong>y begin a trial in each o<strong>the</strong>r‟s proximity, but fail when <strong>the</strong>y start near<br />
<strong>the</strong>ir respective static objects. Under <strong>the</strong>se conditions a typical evolutionary run will<br />
simply start optimizing solutions for starting positions that are most easily optimized,<br />
because this will result at least in some improvement. However, eventually <strong>the</strong> solutions<br />
will become too specialized to be <strong>the</strong>n adapted to generalize over all starting positions.<br />
This problem was overcome by using a weighted measure of <strong>the</strong> overall success of a<br />
126 | P a g e
solution, whereby <strong>the</strong> contribution of a trial‟s score was inversely proportional to its<br />
ranking among all of <strong>the</strong> trials for that evaluation (cf. Husb<strong>and</strong>s, et al. 1998). This<br />
arrangement dynamically ensures that those starting positions which are difficult are<br />
given more weight.<br />
In <strong>the</strong> following chapters we will make use of simulations based on <strong>the</strong>se principles to<br />
investigate a range of phenomena. Chapter 8 provides an initial motivation for an<br />
integrative perspective that includes <strong>the</strong> interaction process as a crucial element of its<br />
explanatory framework. This is followed in Chapter 9 by a more detailed investigation<br />
into <strong>the</strong> capacity of <strong>the</strong> interaction process to organize <strong>the</strong> behavior of individual agents,<br />
including under impaired <strong>and</strong> unfavorable conditions. Chapter 10 builds on <strong>the</strong>se<br />
insights <strong>and</strong> develops <strong>the</strong>m fur<strong>the</strong>r by focusing on <strong>the</strong> minimal conditions for an<br />
interaction process to become <strong>the</strong> basis for social interaction, <strong>and</strong> how this affects <strong>the</strong><br />
behavioral repertoire of <strong>the</strong> agents.<br />
127 | P a g e
8 Investigating sensitivity to social contingency<br />
The overall aim of this <strong>the</strong>sis is to show that a consideration of sociality helps to address<br />
<strong>the</strong> cognitive gap faced by <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis by shifting a part of <strong>the</strong><br />
explanatory burden away from <strong>the</strong> individual agent to <strong>the</strong> enabling dynamics of <strong>the</strong><br />
interaction process, social interactions <strong>and</strong> cultural context. So far we have argued for<br />
this shift <strong>the</strong>oretically. As a first experimental step toward achieving this goal it is<br />
helpful to demonstrate <strong>and</strong> evaluate <strong>the</strong> possibility of a scientific perspective that<br />
includes <strong>the</strong> dynamics of interaction as an essential element of its explanations.<br />
One promising target for such an endeavor is Murray <strong>and</strong> Trevar<strong>the</strong>n‟s (1985) double<br />
TV monitor experiment. In this psychological study 2 month old infants were animated<br />
by <strong>the</strong>ir mo<strong>the</strong>rs to engage in coordination via a live double video link. However, when<br />
<strong>the</strong> live video of <strong>the</strong> mo<strong>the</strong>r was replaced with a video playback of her actions recorded<br />
previously, <strong>the</strong> infants became distressed or removed. These results, <strong>and</strong> those of a more<br />
rigorous follow-up study by Nadel <strong>and</strong> colleagues (1999), indicate that 2 month old<br />
infants are sensitive to social contingency, i.e. <strong>the</strong> mutual responsiveness during an<br />
interaction, <strong>and</strong> that this sensitivity plays a role in <strong>the</strong> unfolding of coordination.<br />
Figure 8-1. The double TV monitor experiment. The original illustration from Nadel, et al. (1999) has<br />
been modified so as to indicate <strong>the</strong> abstractions made for <strong>the</strong> simulation model. The dashed circles <strong>and</strong><br />
arrows represent <strong>the</strong> two interconnected agents, implemented as embodied dynamical systems.<br />
128 | P a g e
Traditional explanations of this sensitivity have focused on inborn factors. For example,<br />
Gergely <strong>and</strong> Watson (1999) have postulated <strong>the</strong> presence of an innate cognitive module<br />
which enables <strong>the</strong> detection of social contingency, <strong>and</strong> Russell (1996) hypo<strong>the</strong>sizes that<br />
infants have a native capacity to underst<strong>and</strong> intentionality <strong>and</strong> to process agency. Are<br />
<strong>the</strong>se postulations of innate capacities on <strong>the</strong> part of <strong>the</strong> infant necessary in order to<br />
explain <strong>the</strong> empirical results?<br />
Iizuka <strong>and</strong> Di Paolo (2007b) used an evolutionary robotics (ER) approach to test<br />
whe<strong>the</strong>r simpler solutions could also emerge from <strong>the</strong> dynamics of <strong>the</strong> interaction<br />
process itself. In <strong>the</strong>ir simulation model <strong>the</strong> evolved agents, which will be described in<br />
more detail later (cf. Figure 8-2), successfully acquired <strong>the</strong> capacity to discriminate<br />
between „live‟ (two-way) <strong>and</strong> „recorded‟ (one-way) interaction. Moreover, an analysis<br />
of <strong>the</strong> resulting dynamics suggests that <strong>the</strong> interaction process itself plays an important<br />
role in enabling this behavior. Similar results were also found by o<strong>the</strong>r related<br />
simulation studies (e.g. Di Paolo, et al. 2008; Ikegami & Iizuka 2007; Iizuka & Ikegami<br />
2004a; Di Paolo 2000; 1999).<br />
It could be argued that <strong>the</strong> result of Iizuka <strong>and</strong> Di Paolo‟s modeling study only<br />
represents a specific subset of <strong>the</strong> general solution space, in particular because <strong>the</strong>y used<br />
ER to explicitly generate agents that terminate <strong>the</strong> ongoing interaction when <strong>the</strong>re is a<br />
lack of social contingency. In o<strong>the</strong>r words, <strong>the</strong>y employed a fitness function which adds<br />
fitness scores to those solutions which lead to this avoidance behavior. We address this<br />
issue more indirectly by testing whe<strong>the</strong>r termination of interaction emerges under more<br />
general conditions. Answering this question is important if <strong>the</strong> argument is made that<br />
<strong>the</strong>se findings might apply more generally. Moreover, by changing <strong>the</strong> simulation setup<br />
in this manner we have moved <strong>the</strong> model closer to <strong>the</strong> original double TV monitor<br />
experiment: <strong>the</strong> infants presumably did not have <strong>the</strong> specific goal to detect whe<strong>the</strong>r <strong>the</strong>y<br />
were dealing with a live video or just a recording. It is more likely that <strong>the</strong>y were simply<br />
attempting to establish social coordination with <strong>the</strong>ir mo<strong>the</strong>rs but were unable to do so.<br />
In summary, we will use an ER methodology to generate pairs of simulated agents<br />
capable of reliably establishing <strong>and</strong> maintaining a coordination pattern under noisy<br />
conditions. Unlike previous related work, agents are only evolved for this ability <strong>and</strong><br />
129 | P a g e
not for <strong>the</strong>ir capacity to discriminate social contingency (i.e., a live responsive partner)<br />
from non-contingent engagements (i.e., a recording). However, as it turned out, when<br />
<strong>the</strong>y are made to interact with a recording of <strong>the</strong>ir partner made during a successful<br />
previous interaction, <strong>the</strong> coordination pattern cannot be established. An analysis of <strong>the</strong><br />
system‟s underlying dynamics reveals (i) that stability of <strong>the</strong> coordination pattern<br />
requires ongoing mutuality of interaction, <strong>and</strong> (ii) that <strong>the</strong> interaction process is not only<br />
constituted by, but also constitutive of, individual behavior. We suggest that this<br />
stability of coordination is a general property of a certain class of interactively coupled<br />
dynamical systems, <strong>and</strong> conclude that psychological explanations of an individual‟s<br />
sensitivity to social contingency need to take into account <strong>the</strong> role of <strong>the</strong> interaction<br />
process.<br />
8.1 Methods<br />
We implemented a minimal ER model analogous to Murray <strong>and</strong> Trevar<strong>the</strong>n‟s (1985)<br />
double TV monitor experiment by building on recent work by Iizuka <strong>and</strong> Di Paolo<br />
(2007b). The goal of <strong>the</strong> agents is to cross <strong>the</strong>ir sensors as far away from <strong>the</strong>ir starting<br />
positions as possible, a task which requires mutual localization, convergence on a target<br />
direction, <strong>and</strong> movement in that direction while not losing track of each o<strong>the</strong>r. This task<br />
is non-trivial since sensory stimulation only correlates with <strong>the</strong> overlapping of position<br />
(when <strong>the</strong> centers of <strong>the</strong> agents are less than 20 units of space apart); it does not convey<br />
<strong>the</strong> direction or speed of movement of <strong>the</strong> o<strong>the</strong>r agent. The agents are 40 units wide,<br />
have an on/off sensor at <strong>the</strong>ir center, <strong>and</strong> can only move left or right by controlling <strong>the</strong><br />
output of <strong>the</strong>ir left <strong>and</strong> right motor nodes (see Figure 8-2).<br />
Figure 8-2. A schematic view of <strong>the</strong> model adapted from Iizuka <strong>and</strong> Di Paolo (2007b). The two identical<br />
agents are 40 units wide, only able to move in a horizontal direction, <strong>and</strong> equipped with a single on/off<br />
sensor at <strong>the</strong>ir centre. They face each o<strong>the</strong>r in an unlimited continuous 1-D space.<br />
130 | P a g e
Agents are controlled by a CTRNN consisting of 3 fully-connected nodes with selfconnections<br />
27 . Similar settings have already been successfully used by Iizuka <strong>and</strong> Di<br />
Paolo (2007b). The main differences are that (i) <strong>the</strong> agents of <strong>the</strong> current study only<br />
have 3 nodes, (ii) <strong>the</strong> input is fed to all nodes instead of one dedicated sensory node,<br />
(iii) <strong>and</strong> each actuator node has its own gain parameter. The first difference was chosen<br />
to fur<strong>the</strong>r minimize <strong>the</strong> conditions of <strong>the</strong> model <strong>and</strong> facilitate analysis; differences (ii)<br />
<strong>and</strong> (iii) were implemented because <strong>the</strong>y were found to increase <strong>the</strong> evolvability of <strong>the</strong><br />
solutions.<br />
Noise is introduced into <strong>the</strong> simulation for two main reasons: (i) since <strong>the</strong> agents are<br />
identical <strong>the</strong>y will need to make use of noise in order to break <strong>the</strong> symmetry of <strong>the</strong>ir<br />
movements <strong>and</strong> converge on a common target direction, <strong>and</strong> (ii) robustness against<br />
noise increases <strong>the</strong> ability of „live‟ agents to cope with playback situations (Iizuka &<br />
Ikegami 2004a). Accordingly, at each Euler time step <strong>the</strong>re is a 5% probability that <strong>the</strong><br />
current sensory state is flipped into its opposite state. We add a small perturbation to <strong>the</strong><br />
motor outputs at each time step drawn from a Gaussian distribution (μ = 0; σ 2 = 0.05).<br />
The noise is applied to <strong>the</strong> outputs before <strong>the</strong> application of motor gains. In order to<br />
fur<strong>the</strong>r increase <strong>the</strong> robustness of <strong>the</strong> behavioral strategies, <strong>the</strong> initial relative<br />
displacement between <strong>the</strong> agents varies (range [-25, 25]). Starting from any of <strong>the</strong>se<br />
possible relative positions, <strong>the</strong> task for <strong>the</strong> agents is to coordinate <strong>the</strong>ir behavior such<br />
that <strong>the</strong>y cross each o<strong>the</strong>r as far away from position 0 as possible. Since <strong>the</strong> agents are<br />
started in opposite orientation („up‟ vs. „down‟), it is not possible for <strong>the</strong> evolutionary<br />
algorithm to hard code any trivial solution (e.g. „always move left‟).<br />
In terms of <strong>the</strong> evolutionary algorithm, <strong>the</strong> population size is 40 <strong>and</strong> <strong>the</strong> algorithm<br />
terminates at 5000 generations. No niching was used. During each fitness evaluation an<br />
27 For each of <strong>the</strong> three nodes of <strong>the</strong> CTRNN <strong>the</strong> parameter ranges are as follows: time constant τ i has [1,<br />
100], bias b i has range [-3, 3]), <strong>and</strong> weights w ji have range [-8, 8]. All nodes receive <strong>the</strong> same sensory<br />
input, namely <strong>the</strong> sensor state multiplied by an input gain with range [1, 100]. The overall agent velocity<br />
is calculated as <strong>the</strong> difference between <strong>the</strong> left <strong>and</strong> right motor nodes. The velocity of each motor node is<br />
calculated by mapping its output onto <strong>the</strong> range [-1, 1] <strong>and</strong> <strong>the</strong>n multiplying it by an output gain<br />
parameter (range [1, 50]). The time evolution of each agent‟s controller is calculated by using Euler<br />
integration with a time step of 0.1.<br />
131 | P a g e
agent is tested in 15 trials; to increase <strong>the</strong> robustness of <strong>the</strong> evolving solutions to noise<br />
<strong>and</strong> variations in initial conditions only <strong>the</strong> lowest score achieved in any of <strong>the</strong> trials is<br />
chosen as <strong>the</strong> overall score. Each trial run consists of 50 units of time (500 Euler time<br />
steps). At <strong>the</strong> start of each trial agents have <strong>the</strong>ir internal node activations set to small<br />
r<strong>and</strong>om values drawn from a st<strong>and</strong>ard uniform Gaussian distribution (μ = 0; σ 2 = 1). The<br />
initial distance between <strong>the</strong> agents varies; agent ‘down’ always gets placed at position 0,<br />
while agent ‘up’ starts at a different position for each trial (15 different positions evenly<br />
distributed across range [-25, 25]).<br />
The fitness score of a trial run is calculated on <strong>the</strong> basis of a single factor, namely <strong>the</strong><br />
absolute value of <strong>the</strong> final crossing position of <strong>the</strong> two agents (divided by a factor of<br />
10). Thus, in contrast to <strong>the</strong> work done by Iizuka <strong>and</strong> Di Paolo (2007b), <strong>the</strong>se agents<br />
were not explicitly evolved to break off <strong>the</strong> interaction pattern when detecting a lack of<br />
social contingency. Instead, we simply aimed to generate a model that under normal<br />
(but noisy) circumstances results in highly fit coordination behavior. Presumably, <strong>and</strong><br />
this is our null hypo<strong>the</strong>sis, <strong>the</strong> agents capable of such robust behavior should be able to<br />
sustain an interaction even when faced with <strong>the</strong> ‘playback’ condition.<br />
8.2 Results<br />
The GA was run 4 times. The fittest agent, with a score of 244.8, was produced during<br />
<strong>the</strong> 4 th run in generation 3477. This solution was <strong>the</strong>n tested extensively; agent „down‟<br />
was always placed at position 0, while agent „up‟ starts at a different position for each<br />
trial (101 positions evenly distributed across range [-50, 50]). Each trial is repeated 150<br />
times. The mean score across this range of initial conditions is plotted in Figure 8-3<br />
(left). The agents are able to generalize <strong>the</strong>ir behavior well beyond <strong>the</strong> range that <strong>the</strong>y<br />
were originally evolved to cope with. On average <strong>the</strong> best initial position for agent „up‟<br />
turned out to be at 11 (mean score: 292.9).<br />
132 | P a g e
Mean fitness<br />
Mean fitness<br />
Mean fitness<br />
Mean fitness<br />
400<br />
400<br />
300<br />
300<br />
200<br />
200<br />
100<br />
100<br />
0<br />
-50 -25 0 25 50<br />
Relative displacement<br />
0<br />
-50 -25 0 25 50<br />
Relative displacement<br />
Figure 8-3. Left: Mean score achieved by <strong>the</strong> fittest agent starting from various initial positions, with<br />
st<strong>and</strong>ard deviation. Right: Mean score by <strong>the</strong> fittest agent but this time interacting with non-responsive,<br />
recorded movements obtained from <strong>the</strong> original trials.<br />
In order to demonstrate <strong>the</strong> general robustness of <strong>the</strong> evolved agents under this initial<br />
condition, we ran ano<strong>the</strong>r set of trials from this initial position, while varying noise<br />
levels (see Figure 8-4). The motor noise was varied while <strong>the</strong> sensor noise remained<br />
constant at evolutionary strength (5%), <strong>and</strong> sensor noise was varied while motor noise<br />
remained constant (σ 2 = 0.05). At each noise level we tested <strong>the</strong> agents for 150 trials.<br />
400<br />
400<br />
300<br />
300<br />
200<br />
200<br />
100<br />
100<br />
0<br />
0 0.5 1 1.5 2 2.5<br />
Motor noise level<br />
0<br />
0 10 20 30 40 50<br />
Sensor noise level<br />
Figure 8-4. Robustness to noise: mean fitness score achieved over 150 trials by <strong>the</strong> fittest evolved agent<br />
starting from position 11 for a range of noise levels, with st<strong>and</strong>ard deviation. Original noise strength<br />
during evolution is 0.05 for motor (left) <strong>and</strong> 5% for sensor noise (right).<br />
The agents are able to cope with a wide range of perturbations. Indeed, <strong>the</strong>ir overall<br />
performance degrades gracefully until <strong>the</strong> sensor <strong>and</strong> motor signals are completely<br />
swamped by noise. In <strong>the</strong> case of sensor noise, for example, average performance only<br />
approaches 0 just before reaching <strong>the</strong> 50% mark (at which point sensory activation<br />
becomes completely arbitrary). This demonstrates that <strong>the</strong> agents are able to produce<br />
highly robust coordination behavior.<br />
133 | P a g e
Finally, ano<strong>the</strong>r 150 trials were conducted with agent „up‟ at position 11 (under normal<br />
noise conditions). The movement of agent „down‟ during <strong>the</strong> best trial (score: 321) was<br />
recorded for playback. Ano<strong>the</strong>r 150 trials were <strong>the</strong>n run under playback conditions: <strong>the</strong><br />
initial conditions reflect those of <strong>the</strong> recorded best trial run (agent „up‟ always starts at<br />
position 11 <strong>and</strong> with <strong>the</strong> same initial internal activation), <strong>and</strong> <strong>the</strong> movement of agent<br />
„down‟ replicate those which it produced during <strong>the</strong> recording. While <strong>the</strong> sensorimotor<br />
noise for agent „up‟ was different during each of <strong>the</strong>se trials, no additional noise was<br />
added to <strong>the</strong> recorded movement of agent „down‟.<br />
The results are striking: whereas <strong>the</strong> original 150 trials of mutual (two-way) interaction<br />
were highly successful (mean score: 268), <strong>the</strong> 150 trials of playback (one-way)<br />
interaction were a drastic failure (mean score: 19). Effectively, agent „up‟ was not able<br />
to sustain an interaction with <strong>the</strong> playback movements of agent „down‟. The severity of<br />
this failure is especially surprising since under normal conditions <strong>the</strong> active agent is<br />
robust against various forms of noise, <strong>and</strong> able to cope effectively with a wide range of<br />
initial conditions. Moreover, during <strong>the</strong> playback condition its „partner‟ performs what<br />
had previously been a highly fit behavioral repertoire. Still, <strong>the</strong> active agent is unable to<br />
adapt to <strong>the</strong> situation of interacting with a non-responsive „partner‟. It could be argued<br />
that this result is unique to <strong>the</strong> chosen situation. However, this is not <strong>the</strong> case: when<br />
testing agent „up‟ with each of <strong>the</strong> original trials we get <strong>the</strong> same result (see Figure 8-3,<br />
right).<br />
8.3 Behavioral analysis<br />
In order to explain <strong>the</strong>se results we will first analyze <strong>the</strong> behavior of <strong>the</strong> agents. The<br />
behavior under normal conditions can be broken down conceptually into three important<br />
aspects: (i) localization, (ii) alignment, <strong>and</strong> (iii) coordination. We will briefly discuss<br />
<strong>the</strong> first two aspects <strong>and</strong> <strong>the</strong>n focus on <strong>the</strong> third. The activity during <strong>the</strong> first time steps<br />
of <strong>the</strong> best trial run is shown in Figure 8-5.<br />
134 | P a g e
Figure 8-5. Initial activity of <strong>the</strong> two agents during <strong>the</strong> best trial run. From top to bottom <strong>the</strong> traces show<br />
<strong>the</strong> evolution over time of (i) <strong>the</strong>ir relative displacement, (ii) <strong>the</strong>ir noisy input signal <strong>and</strong> actual sensory<br />
contact, (iii) <strong>the</strong>ir velocity, <strong>and</strong> (iv) <strong>the</strong> CTRNN node outputs of agent „up‟.<br />
Initially <strong>the</strong> agents have no knowledge of how <strong>the</strong>ir own position relates to that of <strong>the</strong>ir<br />
partner. Moreover, <strong>the</strong>y have no way of gaining that information except when changing<br />
<strong>the</strong>ir sensory input by engaging in movement. However, it turns out that one<br />
stereotypical behavioral pattern is sufficient to solve <strong>the</strong> non-trivial problem of reliable<br />
localization. First, each agent moves rightwards for a few units of time, <strong>and</strong> <strong>the</strong>n starts<br />
moving leftwards. This sweeping behavior usually takes up to 5 units of time <strong>and</strong> under<br />
evolved conditions always enables <strong>the</strong> agents to locate each o<strong>the</strong>r. In <strong>the</strong> case of<br />
negative initial displacement <strong>the</strong>y will encounter each o<strong>the</strong>r during <strong>the</strong>ir rightward<br />
sweep; o<strong>the</strong>rwise <strong>the</strong>y will cross <strong>the</strong>ir positions during <strong>the</strong>ir leftward return.<br />
Interestingly, <strong>the</strong> agents always end up with positive relative displacement after <strong>the</strong>ir<br />
initial localization. With this clever maneuver <strong>the</strong> agents have significantly reduced <strong>the</strong><br />
complexity of <strong>the</strong>ir coordination task: while sensory input is ambiguous (<strong>the</strong>re is no<br />
indication about <strong>the</strong> direction or speed of <strong>the</strong> o<strong>the</strong>r agent‟s movement), it has now been<br />
135 | P a g e
co-arranged as a „touching on <strong>the</strong> left‟ indicator! This change of <strong>the</strong> sensory meaning is<br />
possible because <strong>the</strong> CTRNN controllers are not symmetric.<br />
How does <strong>the</strong> final oscillatory coordination pattern emerge out of <strong>the</strong> relative<br />
movements of <strong>the</strong> agents? Before analyzing <strong>the</strong> behavior of <strong>the</strong> agents in more detail it<br />
is necessary to briefly describe <strong>the</strong> evolved CTRNN controller, as shown in Figure 8-6.<br />
Most importantly, <strong>the</strong> sensory input excites all of <strong>the</strong> nodes with a gain of S = 10.9, <strong>and</strong><br />
<strong>the</strong> right output gain (44.5) is almost twice as high as <strong>the</strong> left output gain (24.9). The<br />
two motor nodes are inhibited by <strong>the</strong> non-motor node <strong>and</strong> <strong>the</strong>y also inhibit each o<strong>the</strong>r<br />
while hardly affecting <strong>the</strong> non-motor node.<br />
Figure 8-6. The best evolved CTRNN controller. The circles represent <strong>the</strong> nodes with <strong>the</strong>ir time constants<br />
<strong>and</strong> biases. The arrows represent connections, with <strong>the</strong> size indicating <strong>the</strong> weight strength. Dotted arrows<br />
represent inhibitory connections. All nodes receive identical sensory input.<br />
As an example, we can see in Figure 8-5 that <strong>the</strong> output of <strong>the</strong> right motor node (z 3 ) of<br />
agent „up‟ starts to slightly decrease just before time t = 8, due to lack of sensory<br />
stimulation. This shift in velocity entails that agent „down‟ catches up with agent „up‟<br />
<strong>and</strong> <strong>the</strong>y remain in contact (I i = 1) until just before t = 9. During this contact agent „up‟<br />
regains its previous rightward velocity due to sensory stimulation. After separating<br />
again (I i = 0) <strong>the</strong> firing of <strong>the</strong> left motor node goes down followed by <strong>the</strong> right motor<br />
node which eventually leads to <strong>the</strong> behavioral pattern being reinitiated. This behavior<br />
136 | P a g e
appears to be an individual achievement, <strong>and</strong> we <strong>the</strong>refore might expect that agent „up‟<br />
should be able to engage with a playback recording.<br />
Figure 8-7. Initial activity during a playback trial run in which <strong>the</strong> movements of agent „down‟ are <strong>the</strong><br />
same as in Figure 8-5. From top to bottom <strong>the</strong> traces show <strong>the</strong> evolution over time of (i) <strong>the</strong> relative<br />
displacement between <strong>the</strong> agents, (ii) <strong>the</strong>ir noisy input signal <strong>and</strong> <strong>the</strong> actual moments of sensory contact,<br />
(iii) <strong>the</strong>ir velocities, <strong>and</strong> (iv) <strong>the</strong> CTRNN node outputs of agent „up‟.<br />
The activity during <strong>the</strong> playback trial run is shown in Figure 8-7. At first <strong>the</strong> „live‟ agent<br />
aligns itself with <strong>the</strong> „playback‟ agent as it did in <strong>the</strong> original situation. During mutual<br />
(two-way) interaction agent „down‟ would always respond to contact by moving away<br />
slightly; however, in <strong>the</strong> playback situation this co-regulation is prevented from<br />
occurring. Accordingly, every noise-displaced encounter results in a slight decrease of<br />
relative displacement between <strong>the</strong> two agents, <strong>the</strong>reby in turn making it more likely that<br />
<strong>the</strong>re will be ano<strong>the</strong>r sensory stimulation. Up to about t = 3, agent „up‟ is still able to<br />
partially regulate this displacement on its own by adjusting <strong>the</strong> output of its right motor<br />
node. However, from that point onwards <strong>the</strong> right motor node saturates at z 3 = 1, <strong>and</strong><br />
137 | P a g e
<strong>the</strong>reafter remains unaffected by fur<strong>the</strong>r sensory stimulation. Finally, at around t = 6 <strong>the</strong><br />
positive feedback loop between increasing sensory stimulation <strong>and</strong> mounting leftward<br />
velocity becomes unstable in such a way that recovery from breakdown is impossible.<br />
The live agent falls behind <strong>the</strong> playback agent, <strong>and</strong> heads in <strong>the</strong> opposite direction.<br />
Why does this breakdown of coordination not occur when both agents engage in „live‟<br />
interaction? The simple answer provided by this model is that <strong>the</strong> stability of ongoing<br />
coordination requires mutuality of interaction. After <strong>the</strong> initial alignment we find that<br />
coordinated movement in one direction consists of continuous co-regulated oscillatory<br />
behavior. Agents control <strong>the</strong>ir respective velocities such that <strong>the</strong>y cross <strong>the</strong>ir sensors at<br />
relatively regular intervals. This iterative reaction chain constitutes an ongoing pattern<br />
of turn-taking; noise perturbations get amplified in a way that requires continuous coregulated<br />
re-establishment of <strong>the</strong> interaction (cf. Ikegami & Iizuka 2007).<br />
8.4 Dynamical analysis<br />
Can we account for <strong>the</strong> oscillating pattern in dynamical terms? Since <strong>the</strong> output of <strong>the</strong><br />
non-motor node z 1 (y 1 ) is saturated at 1 during coordination, it can be treated as a fixed<br />
parameter. The rest of <strong>the</strong> system consists only of <strong>the</strong> two motor nodes 28 . If agents are<br />
not in contact (I i = 0), <strong>the</strong>re is a stable equilibrium point at (-3.4, -7.5). This state slows<br />
down rightward velocity, <strong>and</strong> <strong>the</strong> agents make contact. When I i = 1 <strong>the</strong> equilibrium<br />
point is shifted to (0.3, 1.9). This speeds up <strong>the</strong> rightward velocity of <strong>the</strong> agent. The<br />
vector field of this autonomous dynamical system is shown in Figure 8-8.<br />
Interestingly, under normal conditions <strong>the</strong> dynamical system never reaches ei<strong>the</strong>r of<br />
<strong>the</strong>se two equilibrium points, because <strong>the</strong>ir existence is made transitory through <strong>the</strong><br />
ongoing interaction. This is illustrated in Figure 8-9 (left) in terms of <strong>the</strong> motor node<br />
firing rates for agent „up‟ over a whole run (50 units of time). The trajectory settles<br />
down into an oscillatory pattern that traces <strong>the</strong> corner near point (0, 1), in <strong>the</strong> middle of<br />
<strong>the</strong> two equilibrium points (located at (0.95, 1) when I i = 1, <strong>and</strong> at (0.30, 0) when I i = 0).<br />
28 The parameters for <strong>the</strong>se two nodes are τ 2 = 1.6, b 2 = 2.6, w 12 = -3.7, w 22 = 1.0, w 32 = -7.9, <strong>and</strong> τ 3 = 1.1,<br />
b 3 = 2.9, w 13 = -5.8, w 23 = -5.8, w 33 = 2.3 (values rounded to one decimal place).<br />
138 | P a g e
The state trajectory for <strong>the</strong> playback situation of <strong>the</strong> same run is displayed in Figure 8-9<br />
(right). At first <strong>the</strong> trajectory moves into <strong>the</strong> same region of state space but <strong>the</strong>n, during<br />
<strong>the</strong> period of prolonged contact, <strong>the</strong> left motor node gets saturated while <strong>the</strong> right motor<br />
node remains at 1. This continues until <strong>the</strong> system almost reaches <strong>the</strong> equilibrium point<br />
at (0.95, 1), but it eventually causes agent „up‟ to slow down too much <strong>the</strong>reby breaking<br />
out of <strong>the</strong> coordination pattern.<br />
Figure 8-8. The vector fields of <strong>the</strong> autonomous dynamical system consisting of <strong>the</strong> left <strong>and</strong> right motor<br />
nodes only (<strong>the</strong> remaining node is saturated at z 1 (y 1 ) = 1). Left: sensory input I = 0, <strong>and</strong> <strong>the</strong>re is a globally<br />
attracting stable equilibrium point at (-3.4, -7.5). Right: input I = 1, <strong>and</strong> <strong>the</strong> equilibrium point is (0.3, 1.9).<br />
Figure 8-9. State trajectory of <strong>the</strong> outputs for <strong>the</strong> 2 motor nodes of agent „up‟ during 50 units of time. The<br />
trace starts at <strong>the</strong> top right corner of each graph. The gray <strong>and</strong> black dot represent <strong>the</strong> globally attracting<br />
stable equilibrium point when sensory input I = 0 <strong>and</strong> I = 1, respectively. Left: mutual (two-way)<br />
interaction. Right: playback (one-way) interaction.<br />
After agent „up‟ slows down enough such that <strong>the</strong> playback movement of agent „down‟<br />
overtakes it, its input I i stays at 0. This causes its motor system to settle near <strong>the</strong><br />
equilibrium point at (0.30, 0), from where it is occasionally perturbed by sensor noise.<br />
139 | P a g e
Thus, without <strong>the</strong> responsive help of <strong>the</strong> o<strong>the</strong>r agent, agent „up‟ is unable to regulate its<br />
behavior such as to avoid falling into this attractor, an event which limits its fur<strong>the</strong>r<br />
behavior to mere leftward movement. The ability to oscillate is co-determined by <strong>the</strong><br />
agents through <strong>the</strong>ir interaction.<br />
So far we have said nothing about <strong>the</strong> kind of dynamics that underlie <strong>the</strong> process by<br />
which <strong>the</strong> agents can co-arrange <strong>the</strong>ir behavior so as to coordinate <strong>the</strong>ir oscillatory<br />
interactions in ei<strong>the</strong>r <strong>the</strong> left- or right-h<strong>and</strong> direction. Indeed, <strong>the</strong> state trajectory that is<br />
shown in Figure 8-9 is focused on one stable regime of behavior <strong>and</strong> its breakdown,<br />
namely when agent „up‟ is coordinating its oscillatory behavior rightwards. There exists<br />
a complementary mode of behavior for <strong>the</strong> leftward direction. How are <strong>the</strong>se two modes<br />
of behavior separated dynamically in <strong>the</strong> state space of <strong>the</strong> CTRNN? If we look closely<br />
at Figure 8-8 we see that <strong>the</strong> transient trajectories in <strong>the</strong> area between <strong>the</strong> two attractors<br />
all run in parallel directions, namely alongside a hypo<strong>the</strong>tical line that would cross both<br />
attractors if <strong>the</strong>y co-existed in one state space. Thus, as <strong>the</strong> system switches back <strong>and</strong><br />
forth between <strong>the</strong> two attractors, it is possible for transient dynamics to oscillate back<br />
<strong>and</strong> forth ei<strong>the</strong>r on <strong>the</strong>ir left- or right-h<strong>and</strong> side. It is likely <strong>the</strong>refore that <strong>the</strong> collective<br />
decision of direction that emerges through <strong>the</strong> interaction of <strong>the</strong> two agents during <strong>the</strong><br />
beginning of a trial is <strong>the</strong> result of a coordinated bifurcation into one or <strong>the</strong> o<strong>the</strong>r of<br />
<strong>the</strong>se two transitory regions of state space 29 .<br />
8.5 Summary<br />
With our simulation study it was found that stable <strong>and</strong> robust coordination can be<br />
reliably established between simulated agents. While <strong>the</strong> agents were only selected on<br />
<strong>the</strong> basis of this coordination ability (ra<strong>the</strong>r than <strong>the</strong>ir capacity to detect social<br />
contingency), coordination still breaks down when a „live‟ agent is forced to interact<br />
with a playback of movements from a previous, successful trial. Agents interacting with<br />
such a non-responsive „partner‟ do not have <strong>the</strong> capacity to generate <strong>and</strong> sustain <strong>the</strong><br />
29 Fur<strong>the</strong>r work is required to determine <strong>the</strong> precise dynamics underlying <strong>the</strong> agents‟ coordinated behavior<br />
at <strong>the</strong> decision point. Collaborations with Jose Fern<strong>and</strong>ez-Leon, who has replicated <strong>and</strong> extended this<br />
simulation study, are underway in order to better underst<strong>and</strong> <strong>the</strong> operations of <strong>the</strong> system.<br />
140 | P a g e
kind of oscillatory behavior necessary for coordination. Thus, what at first appears to be<br />
a behavioral capacity of <strong>the</strong> individual agent is shown to actually emerge out of a<br />
combination of <strong>the</strong> internal dynamics as well as <strong>the</strong> interaction process.<br />
We are thus faced with a peculiar situation in which <strong>the</strong> behavior of <strong>the</strong> individual<br />
agents brings forth <strong>the</strong> interaction process, <strong>and</strong> that interaction process enables <strong>the</strong><br />
behavior of <strong>the</strong> individual agents (cf. Figure 8-10). This makes a reduction of <strong>the</strong><br />
coordination breakdown to an individual agent‟s capacity to detect social contingency<br />
impossible. Moreover, it points to <strong>the</strong> autonomy of <strong>the</strong> interaction process, as postulated<br />
by <strong>the</strong> enactive approach to social cognition (cf. Chapter 4). A more detailed analysis of<br />
<strong>the</strong> dynamics of <strong>the</strong> interaction process in this context is desirable, especially in terms of<br />
an artificial <strong>life</strong> investigation into <strong>the</strong> systemic basis of constitutive autonomy (cf.<br />
<strong>Froese</strong> & Di Paolo 2008b). Finally, this focus on <strong>the</strong> efficacy of <strong>the</strong> interaction process<br />
also has practical relevance for <strong>the</strong> design of multi-agent systems, especially in cases<br />
where an agent‟s „role‟ in a formation is not an intrinsic property of that individual but<br />
emerges from <strong>the</strong> mutual interactions between multiple agents (cf. Quinn, et al. 2003).<br />
Interaction Process<br />
Enables/Constrains<br />
Individual Behavior<br />
Enables/Constrains<br />
Figure 8-10. An illustration of <strong>the</strong> reciprocal relationship between <strong>the</strong> coordinated interaction process <strong>and</strong><br />
<strong>the</strong> individual behavior of <strong>the</strong> agents. The mutual enabling/constraining makes a reduction of <strong>the</strong><br />
coordination breakdown to an individual agent‟s capacity to detect social contingency impossible.<br />
It is worth emphasizing that we do not claim that our model instantiates <strong>the</strong> behavioral<br />
phenomenon which we are investigating, nor that <strong>the</strong> baby-mo<strong>the</strong>r interaction studied<br />
by Murray <strong>and</strong> Trevar<strong>the</strong>n (1985) is reducible to such a simple system. The model is<br />
purely conceptual in that it shows at work a possible explanation that may later be<br />
considered <strong>and</strong> tested in specific empirical cases. Thus, by generating simple models<br />
which do not presuppose <strong>the</strong> methodological individualism which prevails in social<br />
cognitive science <strong>and</strong> psychology, we can re-conceptualize <strong>the</strong> space of possible<br />
explanations (Di Paolo, et al. 2008). In particular, <strong>the</strong> model presented in this paper<br />
141 | P a g e
suggests that <strong>the</strong> capacity for social behavior is strongly dependent on <strong>the</strong> existence of<br />
an appropriate social context, one whose stability is in turn dependent on <strong>the</strong> active <strong>and</strong><br />
responsive engagement of <strong>the</strong> participants.<br />
On this basis we propose that an explanation for <strong>the</strong> infants‟ distressed reaction, which<br />
is observed when confronting <strong>the</strong>m with a video recording ra<strong>the</strong>r than a live stream of<br />
<strong>the</strong>ir mo<strong>the</strong>r, also needs to take into account <strong>the</strong> role of <strong>the</strong> interaction process. Of<br />
course, this does not mean that <strong>the</strong> infants cannot by <strong>the</strong>mselves alone detect social<br />
contingency or that <strong>the</strong>y cannot develop <strong>and</strong> internalize this ability. But this model does<br />
open up <strong>the</strong> possibility for explanations that do not suppose any necessity for inborn<br />
behavioral capabilities <strong>and</strong>/or a complex perceptual strategy on <strong>the</strong> part of <strong>the</strong> infant.<br />
Never<strong>the</strong>less, it might still be argued that what <strong>the</strong> results show is only <strong>the</strong> possibility of<br />
<strong>the</strong> constitutive role of <strong>the</strong> interaction process, but that it says nothing about what is <strong>the</strong><br />
explanation for <strong>the</strong> empirical cases. After all, we adults are able to perceive <strong>the</strong> presence<br />
of o<strong>the</strong>rs without having to directly interact with <strong>the</strong>m at <strong>the</strong> time, for instance when I<br />
perceive someone walking in <strong>the</strong> distance with his back turned to me. The possibility of<br />
this detached o<strong>the</strong>r-perception clearly shows that more research needs to be done. It is<br />
worth emphasizing again, however, that we have not excluded <strong>the</strong> possibility that<br />
internal mechanisms are playing a role in <strong>the</strong> full explanation, but have ra<strong>the</strong>r advocated<br />
a more inclusive explanatory strategy that makes available a more encompassing<br />
scientific perspective. For example, it may well be that sensitivity to social contingency<br />
is initially a socially mediated phenomenon, but that this external mediation becomes<br />
internalized during development (cf. Vygotsky 1934). However, even in adult <strong>life</strong> <strong>the</strong><br />
primary basis of social underst<strong>and</strong>ing might still be interaction (Gallagher 2001). That<br />
this is indeed a possibility has been supported by <strong>the</strong> psychological study by Auvray<br />
<strong>and</strong> colleagues (2009), which showed that <strong>the</strong> behavior of adult participants can be<br />
effectively organized in <strong>the</strong> presence of an appropriate interaction process (cf. Chapter<br />
6, p. 90). The aim of <strong>the</strong> modeling experiments presented in <strong>the</strong> next two chapters is to<br />
get a better underst<strong>and</strong>ing of this finding.<br />
142 | P a g e
9 Investigating <strong>the</strong> interaction process<br />
In this chapter we continue <strong>the</strong> investigation into <strong>the</strong> organizing dynamics of <strong>the</strong><br />
interaction process by means of ano<strong>the</strong>r set of modeling experiments. These are based<br />
on <strong>the</strong> minimalist psychological study of perceptual crossing by Auvray, Lenay <strong>and</strong><br />
Stewart (2009), which has been described as a detailed case study in Chapter 6 (p. 90).<br />
The value of modeling this experiment has already been shown by Di Paolo, Rohde <strong>and</strong><br />
Iizuka (2008), who used evolutionary robotics to generate a simulation model which<br />
successfully replicated <strong>the</strong> main empirical results while at <strong>the</strong> same time gaining some<br />
additional insights into <strong>the</strong> dynamics of <strong>the</strong> interaction process. For example, <strong>the</strong><br />
problems that <strong>the</strong> model agents had with avoiding interactions with <strong>the</strong>ir respective<br />
static objects led <strong>the</strong>m to predict similar difficulties for human participants. This<br />
prediction was already supported by <strong>the</strong> empirical data presented by Auvray <strong>and</strong><br />
colleagues, but previously went unnoticed. Moreover, <strong>the</strong>y found it practically<br />
impossible to artificially evolve a robust behavioral strategy without introducing<br />
temporal delays into <strong>the</strong> simulation, <strong>the</strong>reby leading to <strong>the</strong> additional hypo<strong>the</strong>sis that<br />
<strong>the</strong>re is a crucial role of timing between external stimulation <strong>and</strong> <strong>the</strong> participants‟<br />
behavior. In <strong>the</strong> cognitive sciences this combination of empirical <strong>and</strong> modeling work on<br />
minimal perceptual crossing has already been used to support <strong>the</strong> development of <strong>the</strong><br />
interactionist approach to social cognition (Gallagher 2008c, pp. 162-166), as well as<br />
<strong>the</strong> enactive approach to social interaction (De Jaegher & <strong>Froese</strong> 2009).<br />
In this chapter we will continue this modeling research with <strong>the</strong> aim of gaining a better<br />
appreciation of <strong>the</strong> fur<strong>the</strong>r potential of this general experimental setup <strong>and</strong>, at <strong>the</strong> same<br />
time, of improving our underst<strong>and</strong>ing of <strong>the</strong> constitutive role of <strong>the</strong> interaction process.<br />
We begin by using a similar modeling setup as that used by Di Paolo <strong>and</strong> colleagues<br />
(2008), <strong>and</strong> provide a comprehensive analysis of <strong>the</strong> evolved behavioral strategy by<br />
means of a set of psycho-physical tests. Their results are successfully replicated. The<br />
novel aspect of this re-implementation is <strong>the</strong> great simplicity of <strong>the</strong> evolved agents,<br />
which enables a detailed dynamical underst<strong>and</strong>ing of <strong>the</strong>ir behavior. The original task is<br />
also modified in two ways in order to fur<strong>the</strong>r test <strong>the</strong> extent to which successful<br />
behavior depends on <strong>the</strong> dynamics of <strong>the</strong> interaction process. In a first variation, <strong>the</strong><br />
outputs of <strong>the</strong> receptor fields are switched between <strong>the</strong> agents, a modification that<br />
143 | P a g e
cripples <strong>the</strong>ir ability to make sense of sensory-motor correlations. Never<strong>the</strong>less, it is<br />
found that even under this impaired condition stable perceptual crossing reliably<br />
emerges from <strong>the</strong> inter-agent interactions. In a second variation, we changed <strong>the</strong> task so<br />
as to introduce a conflict between individual behavior <strong>and</strong> global stability, namely by<br />
evolving agents to locate <strong>the</strong> mobile object which is not <strong>the</strong> o<strong>the</strong>r agent. It is found that<br />
agents can temporarily succeed at this task, but only by regularly falling back into stable<br />
patterns of perceptual crossing. These variations lead to novel hypo<strong>the</strong>ses about human<br />
behavior that are open to verification by additional psychological experiments 30 .<br />
Finally, we use <strong>the</strong> psycho-physical studies of <strong>the</strong> evolved agents to derive a traditional<br />
hypo<strong>the</strong>sis about <strong>the</strong> sub-personal processes which give rise to <strong>the</strong>ir behavior. However,<br />
a detailed analysis of <strong>the</strong> internal dynamics of <strong>the</strong> agents refutes this hypo<strong>the</strong>sis in favor<br />
of one which instead focuses on temporality <strong>and</strong> <strong>the</strong> interaction process. Our inability to<br />
predict <strong>the</strong> operations of even such simple dynamical systems serves as a warning<br />
against similar attempts to underst<strong>and</strong> sub-personal mechanisms on <strong>the</strong> basis of<br />
behavioral observations.<br />
9.1 Methods<br />
The simulation model includes two agents which, following Auvray, et al. (2009), face<br />
each o<strong>the</strong>r in a 1-D environment (cf. Figure 6-5, p. 90). The 1-D environment wraps<br />
around on itself after 600 units of space (i.e. <strong>the</strong> environment is a circle with a<br />
circumference of 600 units). In <strong>the</strong> simulation all distance <strong>and</strong> time units are of an<br />
arbitrary scale. Each model agent can control <strong>the</strong> horizontal movement of its „body‟, i.e.<br />
<strong>the</strong> position of its receptor field that occupies a total of four units of space. The sensory<br />
input of an agent is activated (set to 1) when its receptor field overlaps with ano<strong>the</strong>r<br />
object in <strong>the</strong> 1-D space, o<strong>the</strong>rwise <strong>the</strong> input remains off (set to 0). The position of each<br />
agent is represented by a continuous variable, <strong>and</strong> <strong>the</strong> velocity is determined by a fully<br />
30 In fact, <strong>the</strong> switched receptor field condition has already been <strong>the</strong> target of a recent pilot study by Di<br />
Paolo <strong>and</strong> De Jaegher (personal communication) at <strong>the</strong> University of Sussex. It appears that participants<br />
are able to deal with this condition effectively.<br />
144 | P a g e
inter-connected CTRNN with self-connections 31 . No noise was applied to any part of<br />
<strong>the</strong> simulation.<br />
In terms of <strong>the</strong> evolutionary algorithm, <strong>the</strong> population size was set to 100 <strong>and</strong> an<br />
evolutionary run finished at a maximum of 5000 generations, though it was sometimes<br />
manually terminated beforeh<strong>and</strong> if solutions were already sufficiently good. When <strong>the</strong><br />
desirability of a solution is evaluated, it is tested for a total of 100 trials that are evenly<br />
spread out across <strong>the</strong> set of possible initial conditions: 10 * 10 trials over 600 * 600<br />
different possible starting positions, where <strong>the</strong> difference in positions is determined by a<br />
step-size of 60 (= 600 / 10) units of space. Moreover, to prevent <strong>the</strong> CTRNNs from<br />
simply learning how to deal with an arbitrary set of starting positions, for each<br />
evaluation, <strong>the</strong> whole set of trials is adjusted by a general position offset drawn from a<br />
uniform r<strong>and</strong>om distribution (offset range [0, 60]), <strong>and</strong> each particular position is also<br />
displaced by a r<strong>and</strong>om value drawn from a Gaussian distribution (μ = 0; σ 2 = 30). Each<br />
trial run consists of 800 units of time. At <strong>the</strong> start of each trial, both agents have <strong>the</strong>ir<br />
internal node activations set to 0.<br />
It is important to note that Di Paolo, Rohde <strong>and</strong> Iizuka (2008) introduced a time delay<br />
between <strong>the</strong> activation of an agent‟s receptor field, i.e. due to an encounter in <strong>the</strong> 1-D<br />
environment, <strong>and</strong> <strong>the</strong> perturbation of <strong>the</strong> agent‟s CTRNN. This was apparently needed<br />
in order to evolve more robust <strong>and</strong> dynamic solutions for this particular task. The need<br />
for an explicit time delay is peculiar because a CTRNN, as a universal function<br />
approximator (cf. Funahashi & Nakamura 1993), should in principle be capable of<br />
incorporating such a delay within its own network structure. Accordingly, we have<br />
spent a considerable amount of effort trying to evolve similarly robust <strong>and</strong> dynamic<br />
solutions without <strong>the</strong> external imposition of such a delay. However, this effort supported<br />
<strong>the</strong> findings of Di Paolo <strong>and</strong> colleagues that in practice such an approach does not seem<br />
to generate solutions that robustly generalize over all initial conditions. Consequently,<br />
31 For each of <strong>the</strong> three nodes of <strong>the</strong> CTRNN <strong>the</strong> parameter ranges are: time constant τ i has [1, 200], bias<br />
b i has range [-8, 8]), <strong>and</strong> weights w ji have range [-8, 8]. All nodes receive <strong>the</strong> same sensory input, namely<br />
<strong>the</strong> sensor state multiplied by an input gain with range [1, 100]. The overall agent velocity is calculated as<br />
<strong>the</strong> difference between <strong>the</strong> left <strong>and</strong> right motor nodes (range [-1, 1]). No output gains were used. The time<br />
evolution of each controller is calculated by means of Euler integration with a time step of 0.1.<br />
145 | P a g e
we introduced a delay of 25 units of time into <strong>the</strong> simulation, which finally made <strong>the</strong><br />
evolution of more robust solutions possible. This delay is an integral part of <strong>the</strong>se<br />
solutions. Preliminary tests with shorter delays showed that <strong>the</strong> evolved agents are<br />
capable of dealing with small alterations to some extent, but shortening <strong>the</strong> delay below<br />
20 units makes <strong>the</strong>m completely incapable of distinguishing between an encounter with<br />
a static object <strong>and</strong> <strong>the</strong> o<strong>the</strong>r agent.<br />
The reason for why <strong>the</strong> evolutionary process is unable to incorporate a time delay into<br />
<strong>the</strong> CTRNNs used in this experiment, <strong>and</strong> why <strong>the</strong> inclusion of such a time delay is<br />
practically needed for robust solutions at all, deserves fur<strong>the</strong>r study in <strong>the</strong> future. It is<br />
likely that a CTRNN, as a purely formal system where each node immediately affects<br />
all <strong>the</strong> nodes to which it is connected, is inherently unsuitable for giving rise to delayed<br />
activity. In material systems, on <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>, delays are widespread. Moreover, <strong>the</strong>y<br />
are an important property of biological neural systems, where synaptic <strong>and</strong> conduction<br />
delays depend on <strong>the</strong> length of <strong>the</strong> synaptic path. From this biological perspective it<br />
might appear that time delays are merely an unintended side-product of <strong>the</strong> material<br />
underpinnings of <strong>the</strong> nervous system. Moreover, since delays increasingly disrupt <strong>the</strong><br />
possibility of synchrony in large networks with long-distance connectivity, it seems that<br />
<strong>the</strong>ir effects are mainly deleterious <strong>and</strong> in need of compensation, for example through<br />
inter-neuron „shortcut‟ connections (cf. Buzsáki 2006, p. 78). This is in contrast with<br />
our finding that <strong>the</strong> extension of <strong>the</strong> CTRNN controller with a delay structure appears to<br />
be an integral part of <strong>the</strong> evolved solutions to <strong>the</strong> task.<br />
What might <strong>the</strong> role of <strong>the</strong> delay consist of? Fur<strong>the</strong>r study of this aspect of <strong>the</strong> model is<br />
still needed, but we can already propose a hypo<strong>the</strong>sis. In this particular experiment, <strong>the</strong><br />
delay in <strong>the</strong> sensory-motor loop is constitutive of a loose system-environment coupling,<br />
i.e. delay provides a source of relative decoupling. At first sight <strong>the</strong> need for decoupling<br />
might appear counterintuitive, especially from <strong>the</strong> point of view of embodied-embedded<br />
robotics. After all, a decisive point of that approach was precisely to get away from <strong>the</strong><br />
highly decoupled systems of GOFAI. Instead, <strong>the</strong> focus has shifted to robotic systems<br />
that are tightly embedded in <strong>the</strong>ir environment via continuous <strong>and</strong> immediate sensorymotor<br />
interaction. In o<strong>the</strong>r words, from this perspective we would expect that <strong>the</strong><br />
presence of decoupling will interfere with an embodied-embedded system‟s ability to<br />
146 | P a g e
espond to environmental changes in a robust <strong>and</strong> timely manner, <strong>and</strong> <strong>the</strong>refore is a<br />
factor that should be eliminated if possible.<br />
However, <strong>the</strong>re is a growing amount of research in robotics which demonstrates that<br />
small amounts of decoupling might not only be desirable but actually essential for a<br />
variety of behaviors. In effect, decoupling provides a form of mediacy between a system<br />
<strong>and</strong> its environment. And, as Jonas (1966) has argued at length, mediacy <strong>and</strong> autonomy<br />
are complements of each o<strong>the</strong>r. They give rise to a kind of dialectic at <strong>the</strong> core of <strong>life</strong>, a<br />
tension which is most clearly expressed in developmental changes <strong>and</strong> throughout <strong>the</strong><br />
major transitions of evolution. While it is beyond <strong>the</strong> scope of this chapter to present<br />
Jonas‟ philosophy in more detail, it is important to note that <strong>the</strong>re are already a number<br />
of studies in embodied-embedded robotics which have begun to explore <strong>the</strong>se ideas. For<br />
example, <strong>the</strong> essential role of sensory-motor decoupling for active perception has been<br />
investigated in terms of a dedicated „gating‟ neuron (Iizuka & Ikegami 2004b), a<br />
homeostatic neural mechanism (Di Paolo & Iizuka 2008), <strong>and</strong> externally imposed time<br />
delays (Rohde & Di Paolo 2008). Moreover, even control engineering can practically<br />
benefit from considering <strong>the</strong> role of relative decoupling. For instance, it has been shown<br />
that adding slippery soles to a quadruped robot increases <strong>the</strong> stability of locomotion<br />
while saving motor energy (Iida & Pfeifer 2004). In general, <strong>the</strong>se studies show how <strong>the</strong><br />
incorporation of mediacy can increase a system‟s robustness, because with less rigid<br />
coupling it is less at <strong>the</strong> whim of environmental perturbations, <strong>and</strong> flexibility, because<br />
relative decoupling creates a gap that can be filled by active perception strategies.<br />
Given <strong>the</strong>se advantages for <strong>the</strong> design of embodied-embedded robotics, it is likely that<br />
<strong>the</strong> role of mediacy in biological systems will become an important area of research in<br />
<strong>the</strong> future. Fur<strong>the</strong>rmore, it is possible that such practical concerns will lead robotics to<br />
align itself more closely with <strong>the</strong> <strong>the</strong>oretical framework of <strong>the</strong> enactive paradigm, in<br />
which <strong>the</strong> bio-philosophy of Jonas plays a central role (cf. Di Paolo 2003). While <strong>the</strong>re<br />
clearly remains much more to be done in this area, here we will simply follow Di Paolo,<br />
Rohde <strong>and</strong> Iizuka‟s (2008) approach by using an externally defined time delay in order<br />
to bootstrap <strong>the</strong> artificial evolution of more active behavioral strategies.<br />
147 | P a g e
9.2 Experiments<br />
In this section we describe three sets of experiments. First, <strong>the</strong> performance of <strong>the</strong><br />
simulated agents was evaluated in terms of <strong>the</strong> experimental setup of <strong>the</strong> original<br />
psychological study. We <strong>the</strong>n introduce two novel studies that are specifically aimed at<br />
investigating <strong>the</strong> extent of <strong>the</strong> self-organizing properties of <strong>the</strong> inter-individual<br />
interaction process. In <strong>the</strong> first instance we tested <strong>the</strong> robustness of <strong>the</strong> solution that was<br />
evolved for <strong>the</strong> st<strong>and</strong>ard task by exchanging <strong>the</strong> input signals of <strong>the</strong> receptor fields<br />
between <strong>the</strong> two agents, as this significantly limits <strong>the</strong>ir ability for engaging in<br />
individual sensory-motor exploration. Then we changed <strong>the</strong> experimental setup to one<br />
in which <strong>the</strong> „intentions‟ of <strong>the</strong> individual agents <strong>and</strong> <strong>the</strong> overall interaction dynamics<br />
are in conflict with each o<strong>the</strong>r, namely by evolving agents that were rewarded to interact<br />
with each o<strong>the</strong>r‟s „shadow‟ object, i.e. an object that is active (mobile) but not interactive<br />
(non-contingent).<br />
9.2.1 Experimental setup 1: Original setup<br />
As a first step we tried to replicate <strong>the</strong> modeling work that has already been done by Di<br />
Paolo, Rohde <strong>and</strong> Iizuka (2008). Their simulation model is relatively faithful to <strong>the</strong><br />
details of Auvray, Lenay <strong>and</strong> Stewart‟s (2009) experimental setup, with one significant<br />
difference: while <strong>the</strong> original task for <strong>the</strong> participants was to click a mouse whenever<br />
<strong>the</strong>y encountered each o<strong>the</strong>r, <strong>the</strong> task for <strong>the</strong> model agents is to locate <strong>the</strong> partner agent<br />
<strong>and</strong> spend as much time as possible as close to each o<strong>the</strong>r as possible. Implicit in this<br />
task is <strong>the</strong> requirement for <strong>the</strong> agents not to become „trapped‟ by <strong>the</strong> static object or<br />
„shadow‟ object of <strong>the</strong> o<strong>the</strong>r agent. Accordingly, in <strong>the</strong>ir model <strong>the</strong> fitness score F of<br />
each trial is calculated to be inversely proportional to <strong>the</strong> average distance between <strong>the</strong><br />
two agents (it is <strong>the</strong>refore <strong>the</strong> same for both). The score F for a particular trial is<br />
determined by Equation 9-1.<br />
Equation 9-1<br />
F<br />
1<br />
T<br />
T<br />
d(<br />
t)<br />
1<br />
0 300<br />
In this equation T is <strong>the</strong> total number of time steps per trial, 300 is <strong>the</strong> maximum spatial<br />
distance between <strong>the</strong> agents (since <strong>the</strong> 1-D environment wraps around between 0 <strong>and</strong><br />
148 | P a g e
600 units), <strong>and</strong> d(t) is <strong>the</strong> spatial distance between <strong>the</strong> agents at time step t. Fitness<br />
scores vary continuously (fitness range [0, 1]), with 0 being <strong>the</strong> worst.<br />
The advantage of this evaluation function is that it simplifies <strong>the</strong> task for <strong>the</strong> model<br />
agents, since <strong>the</strong>y do not have to engage in any additional explicit classification<br />
response (i.e. some form of „clicking‟). Moreover, because <strong>the</strong> score is a continuous<br />
measure this increases <strong>the</strong> evolvability of <strong>the</strong> solutions. In contrast to a discrete<br />
evaluation measure of a number of successful clicks, here every approaching behavior is<br />
rewarded in a proportional manner, even if <strong>the</strong> agents do not happen to find each o<strong>the</strong>r<br />
(i.e. <strong>the</strong>y do not actually need to cross).<br />
However, <strong>the</strong>re are also drawbacks in using <strong>the</strong> average distance between <strong>the</strong> agents as<br />
a measure of <strong>the</strong> desirability of <strong>the</strong> evolved solutions. Most importantly, this measure<br />
fails to properly distinguish between (i) an agent‟s general exploratory movements <strong>and</strong><br />
interactions, <strong>and</strong> (ii) <strong>the</strong> explicit distinction of an ongoing interaction as an encounter<br />
with <strong>the</strong> receptor field of <strong>the</strong> o<strong>the</strong>r agent. In contrast, it is possible for a human<br />
participant in <strong>the</strong> original psychological experiments to spend a considerable amount of<br />
time engaging in interactions with <strong>the</strong> static <strong>and</strong>/or shadow object of <strong>the</strong> o<strong>the</strong>r, which is<br />
often <strong>the</strong> case, but <strong>the</strong>n still decide against a mouse click response. Never<strong>the</strong>less, due to<br />
<strong>the</strong> fact that this fitness measure turned out to be much more evolvable than equivalent<br />
measures based on „clicking‟ accuracy, we chose to retain it for this study. Future work<br />
could investigate whe<strong>the</strong>r <strong>the</strong> addition of an explicit „clicking‟ ability changes <strong>the</strong><br />
general behavioral strategies of <strong>the</strong> agents, or perhaps even <strong>the</strong> inclusion of a model of<br />
arm morphology (cf. Rohde & Di Paolo 2008).<br />
We explored a range of CTRNN network sizes, starting with 11 nodes (<strong>the</strong> maximum<br />
size used by Di Paolo <strong>and</strong> colleagues), but were able to find robust solutions with as<br />
little as 4 nodes. We <strong>the</strong>n chose <strong>the</strong> highest scoring solution out of <strong>the</strong> population which<br />
had achieved <strong>the</strong> highest average fitness out of <strong>the</strong> 10 different evolutionary runs for<br />
fur<strong>the</strong>r testing. During testing <strong>the</strong> duration of each trial was doubled to 1600 units of<br />
time in order to better assess <strong>the</strong> general robustness of <strong>the</strong> selected solution. In addition,<br />
we were able to manually prune several connections in <strong>the</strong> evolved CTRNN without<br />
significantly affecting its performance. Indeed, this pruning eventually allowed us to<br />
149 | P a g e
educe <strong>the</strong> solution to a 3-node CTRNN network, which is significantly smaller than <strong>the</strong><br />
11-node network obtained by Di Paolo <strong>and</strong> colleagues. This pruned CTRNN was <strong>the</strong>n<br />
fur<strong>the</strong>r optimized for 800 generations. The resulting CTRNN is shown in Figure 9-1.<br />
Figure 9-1. The CTRNN controller used for experimental setup 1. Legend: Circles represent CTRNN<br />
nodes with time constants τ i , block-tailed arrows represent bias connections b i , diamond-tailed arrows<br />
represent weighted input connections Iw i , normal arrows represent motor outputs z i (left motor: L-M;<br />
right motor: R-M), circle-headed arrows represent weighted inter-node connections w ij (including selfconnections).<br />
Negative connections are depicted as dashed lines, while positive connections are solid. The<br />
size of <strong>the</strong> arrows is roughly proportional to <strong>the</strong> strength of <strong>the</strong> connection, while <strong>the</strong> size of <strong>the</strong> circles is<br />
roughly proportional to <strong>the</strong> speed of <strong>the</strong> CTRNN node.<br />
In order to get an initial rough idea of what <strong>the</strong> different fitness scores mean in terms of<br />
agent behavior for this solution, it is helpful to consider <strong>the</strong> following illustrative cases:<br />
(i) A fitness score of 0 is an absolute limit point that is only ever attained when<br />
evolving solutions without delay, which often produces agents that suddenly stop<br />
moving when <strong>the</strong>ir receptor field becomes activated. Thus, when <strong>the</strong>se agents start<br />
<strong>the</strong> trial on <strong>the</strong>ir respective static objects, in this case <strong>the</strong>y will not move for <strong>the</strong> rest<br />
of <strong>the</strong> trial, <strong>and</strong> remain maximally distant from each o<strong>the</strong>r (<strong>the</strong> static objects are<br />
located 300 units apart).<br />
(ii) A fitness score of 0.5 is obtained, for example, when permanently turning off <strong>the</strong><br />
receptor field of <strong>the</strong> evolved agents, which <strong>the</strong>n continually circle <strong>the</strong> 1-D<br />
environment until <strong>the</strong> end of <strong>the</strong> trial. The behavior is also often displayed by<br />
150 | P a g e
<strong>and</strong>omly initialized CTRNNs that do not yet respond to perturbations. In <strong>the</strong>se<br />
cases <strong>the</strong> agents move continuously in opposite directions around <strong>the</strong> 1-D<br />
environment, <strong>and</strong> <strong>the</strong>refore spend equal amounts of time near <strong>the</strong> maximally-distant<br />
<strong>and</strong> maximally-close positions.<br />
(iii) A fitness score of 1 is ano<strong>the</strong>r absolute limit point that is only ever attained by<br />
agents that have been evolved without delay, <strong>and</strong> which immediately stop moving<br />
when <strong>the</strong>ir receptor field gets activated. Thus, when <strong>the</strong>se agents start <strong>the</strong> trial on<br />
top of each o<strong>the</strong>r, <strong>the</strong>y will not move for <strong>the</strong> rest of <strong>the</strong> trial, <strong>and</strong> remain maximally<br />
close to each o<strong>the</strong>r.<br />
Accordingly, <strong>the</strong> most interesting behavior for this CTRNN will be found by examining<br />
trials that have fitness scores of around 0.75. In <strong>the</strong>se cases <strong>the</strong> agents have successfully<br />
engaged in perceptual crossing for at least some amount of time during <strong>the</strong> trial, but<br />
<strong>the</strong>y must have also faced some difficulties. This usually means that an agent was<br />
delayed by an encounter with its static object or <strong>the</strong> shadow of <strong>the</strong> o<strong>the</strong>r agent during<br />
<strong>the</strong> beginning of <strong>the</strong> trial, or that <strong>the</strong>re was some o<strong>the</strong>r kind of interference which<br />
caused <strong>the</strong> perceptual crossing to break down temporarily before being reestablished.<br />
In order to obtain a comprehensive overview of this solution‟s performance we tested it<br />
for each possible combination of starting positions of <strong>the</strong> two agents (600 x 600 trials).<br />
The results of <strong>the</strong>se tests are depicted graphically in Figure 9-2.<br />
151 | P a g e
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
1<br />
0.9<br />
0.8<br />
0.7<br />
Region 1<br />
Region 2<br />
Region 3<br />
Region 4<br />
Region 5<br />
Region 6<br />
100 200 300 400 500 600<br />
0.6<br />
Figure 9-2. Graphical representation of fitness scores achieved at each possible combination of starting<br />
positions for agent „up‟ (x-axis) <strong>and</strong> agent „down‟ (y-axis). Note that <strong>the</strong> axes wrap around due to <strong>the</strong> 1-D<br />
circular shape of <strong>the</strong> virtual environment. Fitness scores range from 0.60 to 0.96 with an average of 0.87.<br />
See text for an explanation of <strong>the</strong> different regions.<br />
We can identify several salient regions in Figure 9-2 which enable us to make some<br />
general comments about <strong>the</strong> evolved solution. First, it should be noted how well <strong>the</strong><br />
model agents manage to generalize over all initial conditions; no trial resulted in a<br />
fitness score of less than 0.60. Second, <strong>the</strong> evolved solution is not strictly symmetrical.<br />
However, this is not surprising since we did not enforce any structural symmetry on <strong>the</strong><br />
CTRNN controller. Second, <strong>the</strong>re are some regions of clear qualitative changes<br />
indicated by sharp grayscale differences. These regions have been marked as 1 to 6 in<br />
Figure 9-2 for ease of reference. A brief description of <strong>the</strong> behavior of <strong>the</strong> agents when<br />
initially starting from <strong>the</strong>se different regions gives an overview of <strong>the</strong>ir general<br />
behavioral domain:<br />
1) The fitness in this region is relatively lower because both agents get perturbed<br />
by <strong>the</strong>ir respective static objects during <strong>the</strong> beginning of <strong>the</strong> trials. This will<br />
cause <strong>the</strong>m to briefly oscillate around <strong>the</strong> object. After a few contacts <strong>the</strong> agents<br />
proceed to break out of <strong>the</strong> oscillation, continue to move on, <strong>and</strong> eventually<br />
come into contact. From this point onward <strong>the</strong>y engage in perceptual crossing<br />
until <strong>the</strong> end of <strong>the</strong> trial.<br />
152 | P a g e
2) The low fitness diagonal region going on <strong>the</strong> right-h<strong>and</strong> side from starting<br />
position (0, 0) to (600, 600) can be explained as follows: agent „up‟ (x-axis)<br />
moves rightwards without any input, while agent „down‟ (y-axis) moves<br />
leftwards. This means that when agent „up‟ starts within 52 units of space to <strong>the</strong><br />
right of agent „down‟, it will encounter <strong>the</strong> o<strong>the</strong>r agent‟s shadow object. This<br />
will trigger its receptor field <strong>and</strong> cause it to briefly turn back. However, agent<br />
„down‟ has not been perturbed by this encounter <strong>and</strong> continues moving<br />
leftwards. Agent „up‟ will thus not make any fur<strong>the</strong>r contact <strong>and</strong> continue<br />
moving rightwards. In sum, what happens in this region is that agent „up‟ gets<br />
delayed, <strong>and</strong> <strong>the</strong> agents thus spend more time apart from each o<strong>the</strong>r. This ceases<br />
to be a problem when agent „up‟ starts fur<strong>the</strong>r along <strong>the</strong> x-axis compared to <strong>the</strong><br />
starting position of agent „down‟.<br />
3) This whole region is relatively flawless. The agents move unperturbed past each<br />
o<strong>the</strong>r‟s static object, eventually come into contact, <strong>and</strong> engage in perceptual<br />
crossing until <strong>the</strong> end of <strong>the</strong> trial. The increasing fitness gradient toward <strong>the</strong><br />
corner (600, 0) reflects <strong>the</strong> fact that <strong>the</strong> agents have to travel less distance to<br />
meet each o<strong>the</strong>r.<br />
4) This streak of uneven fitness distribution reflects <strong>the</strong> fact that <strong>the</strong> agents, when<br />
<strong>the</strong>y start from this region, sometimes encounter each o<strong>the</strong>r in <strong>the</strong> vicinity of <strong>the</strong><br />
static object of agent „up‟ (located at x = 148). This interference causes <strong>the</strong><br />
agents to break off perceptual crossing eventually, at least under some<br />
conditions.<br />
5) This streak of uneven fitness distribution reflects <strong>the</strong> fact that <strong>the</strong> agents, when<br />
<strong>the</strong>y start from this region, sometimes encounter each o<strong>the</strong>r in <strong>the</strong> vicinity of <strong>the</strong><br />
static object of agent „down‟ (located at y = 448). This interference causes <strong>the</strong><br />
agents to break off perceptual crossing eventually, at least under some<br />
conditions.<br />
6) This solid line of relative low fitness represents cases where <strong>the</strong> agents<br />
encounter each o<strong>the</strong>r at <strong>the</strong> beginning of <strong>the</strong> trial, but eventually disengage,<br />
153 | P a g e
move on, <strong>and</strong> finally establish proper perceptual crossing until <strong>the</strong> end of <strong>the</strong><br />
run. These cases are <strong>the</strong> only trials in which <strong>the</strong> two agents break off perceptual<br />
crossing even though <strong>the</strong>re is no external interference present. Such spontaneous<br />
disengagement did not occur with <strong>the</strong> original 4-node solution; <strong>the</strong> 3-node<br />
solution is thus slightly less robust. The behavior appears to be related to <strong>the</strong><br />
way in which <strong>the</strong> agents initially engage each o<strong>the</strong>r. In certain cases <strong>the</strong>y are<br />
unable to stabilize <strong>the</strong> interaction appropriately.<br />
We now have a rough underst<strong>and</strong>ing of <strong>the</strong> behavioral domain of <strong>the</strong> agents. However,<br />
so far we have not gained any detailed underst<strong>and</strong>ing of what allows <strong>the</strong> agents to be<br />
sensitive to <strong>the</strong> social contingency of <strong>the</strong>ir interactions. In o<strong>the</strong>r words, how do <strong>the</strong><br />
agents distinguish an interaction with <strong>the</strong>ir partner from an interaction with <strong>the</strong> static or<br />
shadow object? To avoid prolonging any interaction with <strong>the</strong> shadow object is actually<br />
relatively straightforward, since <strong>the</strong> situation itself is unstable. The o<strong>the</strong>r agent will not<br />
be perturbed by <strong>the</strong> encounter with its shadow <strong>and</strong> <strong>the</strong>refore move away, trailing its<br />
shadow object behind it, <strong>and</strong> thus eventually terminate <strong>the</strong> one-sided interaction<br />
process. However, if an agent happens to encounter its static object <strong>the</strong>n things are more<br />
complicated, since oscillating around this object is based on a stable environmental<br />
situation. Never<strong>the</strong>less, <strong>the</strong> evolved agents do manage to disengage from interactions<br />
with <strong>the</strong>ir static objects eventually. How is this possible?<br />
In order to get a better underst<strong>and</strong>ing of this discriminatory capacity it is helpful to<br />
study a particular interaction in more depth. We chose a representative trial starting<br />
from position (100, 500), which is illustrated in Figure 9-3. The fitness score was 0.74.<br />
During this trial <strong>the</strong> agents first encounter <strong>the</strong>ir respective static objects (before 1000<br />
time steps), continue to oscillate around this object (from 1000 to 3000 steps), <strong>the</strong>n<br />
disengage <strong>and</strong> continue searching (from 3000 to 4500 steps), <strong>the</strong>n finally locate each<br />
o<strong>the</strong>r (after ca. 4500 steps), <strong>and</strong> establish perceptual crossing until <strong>the</strong> end (16000<br />
steps). On <strong>the</strong> basis of this trial it is possible to investigate why <strong>the</strong> agents disengage<br />
from <strong>the</strong> interactions with <strong>the</strong>ir static objects, but continue to engage in perceptual<br />
crossing once <strong>the</strong>y make contact with each o<strong>the</strong>r.<br />
154 | P a g e
Figure 9-3. Illustration of <strong>the</strong> behavior of <strong>the</strong> agents during a representative trial starting from point (100,<br />
500) with a score of 0.74. They first encounter <strong>the</strong>ir respective static objects, <strong>the</strong>n continue searching, <strong>and</strong><br />
finally locate each o<strong>the</strong>r <strong>and</strong> establish perceptual crossing until <strong>the</strong> end of <strong>the</strong> run (16000 time steps). Top:<br />
<strong>the</strong> position of <strong>the</strong> agents <strong>and</strong> objects in <strong>the</strong> 1D environment over time. Middle: <strong>the</strong> node outputs of agent<br />
„up‟ over time. Bottom: <strong>the</strong> status of <strong>the</strong> receptor field <strong>and</strong> <strong>the</strong> velocity of agent „up‟ over time. Note that<br />
a change of receptor field status reaches <strong>the</strong> agent‟s controller only after a delay of 25 units of time (250<br />
steps).<br />
How does <strong>the</strong> sensitivity to social contingency emerge in this system? The first thing to<br />
notice is that both agents disengage from <strong>the</strong>ir respective static objects after <strong>the</strong> third<br />
time that <strong>the</strong>ir receptor field has become activated. We will <strong>the</strong>refore initially focus our<br />
analysis on this particular moment in time. We know from Di Paolo <strong>and</strong> colleagues<br />
(2008) that it is possible for <strong>the</strong> agents to base <strong>the</strong>ir behavior on <strong>the</strong> duration of<br />
stimulation afforded by an encounter. Similarly, in this trial <strong>the</strong> third encounter between<br />
agent „up‟ <strong>and</strong> its static object lasts 116 steps, while <strong>the</strong> third encounter with <strong>the</strong> o<strong>the</strong>r<br />
agent only lasts 56 steps. The static object <strong>the</strong>refore perturbs <strong>the</strong> agent almost exactly<br />
155 | P a g e
twice as long as <strong>the</strong> o<strong>the</strong>r agent. The reason for this, of course, is that <strong>the</strong> o<strong>the</strong>r agent<br />
moves with an opposite velocity, while <strong>the</strong> static object remains stationary. Do <strong>the</strong><br />
agents make use of this difference in stimulus duration in order to distinguish between<br />
types of interaction?<br />
In order to determine if this is indeed <strong>the</strong> case we performed some psycho-physical tests<br />
on <strong>the</strong> model agents. By altering <strong>the</strong> size of <strong>the</strong> objects within <strong>the</strong> 1D environment, it is<br />
possible to systematically vary <strong>the</strong> length of stimulation encountered by <strong>the</strong> agents.<br />
Similarly, we can also alter <strong>the</strong> size of <strong>the</strong> body of <strong>the</strong> agents. Thus, if <strong>the</strong> sensitivity of<br />
<strong>the</strong> agents relies on this temporal factor (duration of contact), <strong>the</strong>n it should be possible<br />
to alter <strong>the</strong>ir behavior accordingly. To explore this hypo<strong>the</strong>sis we conducted two tests:<br />
1) We start <strong>the</strong> trial from <strong>the</strong> same initial conditions as before. However, just before<br />
<strong>the</strong> 3 rd interaction with <strong>the</strong> static object (at 2100 steps) we decrease <strong>the</strong> size of <strong>the</strong><br />
static object to 3 units of space (from <strong>the</strong> usual 4). In this case <strong>the</strong> performance of<br />
<strong>the</strong> agents is drastically altered (score 0.06); both agents fail to disengage from <strong>the</strong>ir<br />
static objects <strong>and</strong> continue to oscillate around <strong>the</strong>m until <strong>the</strong> end of <strong>the</strong> trial. As<br />
expected, <strong>the</strong> decrease in static object size entailed a shorter stimulation during <strong>the</strong><br />
3 rd contact (101 ra<strong>the</strong>r than 116 steps).<br />
2) We start <strong>the</strong> trial from <strong>the</strong> same initial conditions as <strong>the</strong> original setup. However,<br />
just before <strong>the</strong> 3 rd perceptual crossing (at 5500 steps) we increase <strong>the</strong> size of <strong>the</strong><br />
agents to 10 units of space (from <strong>the</strong> usual 4). In this case <strong>the</strong> performance of <strong>the</strong><br />
agents drops significantly (score 0.59). After making <strong>the</strong> 3 rd contact <strong>the</strong>y drift apart.<br />
As expected, <strong>the</strong> increase in agent size entailed a longer stimulation during <strong>the</strong> 3 rd<br />
encounter (108 ra<strong>the</strong>r than 56 steps).<br />
These tests appear to demonstrate that an agent‟s sensitivity to social contingency<br />
largely depends on <strong>the</strong> duration of <strong>the</strong> 3 rd contact. It <strong>the</strong>refore seems that we can explain<br />
this sensitivity in terms of two thresholds related to <strong>the</strong> duration of contact, t 1 <strong>and</strong> t 2 .<br />
These determine <strong>the</strong> cut-off points between two distinct behaviors, namely continued<br />
oscillation <strong>and</strong> linear exploration. More precisely, we hypo<strong>the</strong>size to find a mechanism<br />
such that: 0 < t 1 < „continue to oscillate around stimulus‟ < t 2 < „continue to explore <strong>the</strong><br />
156 | P a g e
est of environment‟. Is it possible to deduce <strong>the</strong> values for <strong>the</strong>se thresholds from fur<strong>the</strong>r<br />
psycho-physical tests? We will return to this issue during <strong>the</strong> more detailed analysis<br />
presented in Section 9.3.<br />
9.2.2 Experimental setup 2: Switched receptor fields<br />
It could be argued on <strong>the</strong> basis of experimental setup 1 that <strong>the</strong> agents actually employ a<br />
solitary strategy to solve <strong>the</strong> task, namely by internally distinguishing between two<br />
different lengths of stimulation. To be sure, part of <strong>the</strong> basis of this distinction, namely<br />
<strong>the</strong> duration of contact between agents, is co-determined by <strong>the</strong>ir velocities. However,<br />
this co-determination does not necessarily require coordination. As long as both agents<br />
are moving at different velocities <strong>the</strong> possibility of making this distinction effectively<br />
remains <strong>the</strong> same, <strong>and</strong> no mutual interaction is necessary to establish this difference in<br />
<strong>the</strong> first place. Moreover, <strong>the</strong> experimental setup ensures that interactions between an<br />
agent <strong>and</strong> <strong>the</strong> o<strong>the</strong>r‟s shadow are inherently unstable, <strong>the</strong>reby removing <strong>the</strong> shadow as a<br />
possibility for fur<strong>the</strong>r entrainment. On this view, <strong>the</strong> distinction between a static object<br />
<strong>and</strong> <strong>the</strong> o<strong>the</strong>r agent would be largely an individual accomplishment without any need<br />
for <strong>the</strong> autonomous dynamics of an interaction process 32 .<br />
In order to test this null hypo<strong>the</strong>sis we modified <strong>the</strong> experimental setup slightly, namely<br />
by switching <strong>the</strong> inputs to <strong>the</strong> receptor fields between <strong>the</strong> two agents. Under this setup<br />
both agents still control <strong>the</strong> location of <strong>the</strong>ir respective receptor fields, but <strong>the</strong>ir CTRNN<br />
controllers receive <strong>the</strong> perturbations due to <strong>the</strong> o<strong>the</strong>r agent‟s field. In this manner we<br />
have severely disrupted <strong>the</strong> individual sensory-motor correlations, but crucially <strong>the</strong><br />
possibility for engaging in mutual interactions leading to <strong>the</strong> establishment of perceptual<br />
32 Note that this discussion of <strong>the</strong> model raises difficult questions about how to best define <strong>the</strong> concept of<br />
„interaction process‟. When do we mark its beginning, when its end? Does <strong>the</strong> interaction process have to<br />
involve mutual perturbation of internal state or, more abstractly, does it require mutual inter-dependence<br />
of conditions for <strong>the</strong> sustainment of behaviors? In this <strong>the</strong>sis we make use of <strong>the</strong> former, more intuitive<br />
interpretation, but <strong>the</strong> latter might in <strong>the</strong> end turn out to be more accurate. After all, it is possible that one<br />
agent‟s erratic searching behavior partly constitutes, via <strong>the</strong> rigid agent-shadow link, <strong>the</strong> condition for <strong>the</strong><br />
o<strong>the</strong>r agent to continue interacting with <strong>the</strong> seemingly responsive shadow object, a behavior which in turn<br />
partly constitutes <strong>the</strong> condition for <strong>the</strong> first agent, namely that it remains without stimulation from its<br />
partner <strong>and</strong> continues its solitary search, which potentially keeps <strong>the</strong> o<strong>the</strong>r fur<strong>the</strong>r entrained, etc.<br />
157 | P a g e
crossing has remained unchanged from <strong>the</strong> original experiment. Consequently, if <strong>the</strong><br />
agents in addition to <strong>the</strong>ir individual capacities also rely on mutually responsive<br />
interaction in order to distinguish between a static object <strong>and</strong> each o<strong>the</strong>r, <strong>the</strong>y should<br />
still be able to perform <strong>the</strong> task even if <strong>the</strong>y are incapable of using reliable sensorymotor<br />
correlations at <strong>the</strong> individual level. The results of a comprehensive test of this<br />
experimental setup are shown in Figure 9-4.<br />
Figure 9-4. Graphical representation of fitness scores achieved at each possible combination of starting<br />
positions for agent „up‟ (x-axis) <strong>and</strong> agent „down‟ (y-axis). Note that <strong>the</strong> axes wrap around due to <strong>the</strong> 1-D<br />
circular shape of <strong>the</strong> virtual environment. Fitness scores range from 0.40 to 0.965 with an average score<br />
of 0.87.<br />
As predicted, <strong>the</strong> average fitness of this modified condition (0.87) is not different from<br />
that of <strong>the</strong> normal condition (0.87). Moreover, even <strong>the</strong> fitness distribution of <strong>the</strong><br />
comprehensive results of this modified condition look strikingly similar to those of <strong>the</strong><br />
normal condition shown in Figure 9-2. To be sure, some of <strong>the</strong> salient patterns <strong>and</strong><br />
asymmetries have shifted slightly, but <strong>the</strong>se general differences might be expected,<br />
especially considering that we have effectively cross-wired <strong>the</strong> sensory-motor loops of<br />
<strong>the</strong> two agents. How are <strong>the</strong> agents able to cope with <strong>the</strong> distortion of <strong>the</strong>ir sensorymotor<br />
coupling? As a comparison it is helpful to examine again <strong>the</strong> conditions for trial<br />
shown in Figure 9-3, but with switched input parameters. A trace of <strong>the</strong> movement of<br />
<strong>the</strong> agents is shown in Figure 9-5.<br />
158 | P a g e
Figure 9-5. Illustration of <strong>the</strong> behavior of <strong>the</strong> agents during a representative trial starting from point (100,<br />
500). Same experimental setup as shown in Figure 9-3, except that <strong>the</strong> input to <strong>the</strong> receptor fields of <strong>the</strong><br />
agents has been exchanged between <strong>the</strong>m. Fitness score: 0.71<br />
During <strong>the</strong> beginning of <strong>the</strong> trial <strong>the</strong> agents encounter similar situations, <strong>the</strong>reby<br />
providing each o<strong>the</strong>r with relatively matching simulation. At 3000 time steps, however,<br />
agent „down‟ moves across its static object but, due to <strong>the</strong> input switching, remains<br />
unperturbed, while agent „up‟ is stimulated <strong>and</strong> turns back, does not find anything, <strong>and</strong><br />
continues moving rightwards. Finally, <strong>the</strong> agents encounter each o<strong>the</strong>r <strong>and</strong> engage in<br />
perceptual crossing until <strong>the</strong> end of <strong>the</strong> trial. The switched inputs might thus even be<br />
helpful in some circumstances because most of <strong>the</strong> time <strong>the</strong> agents do not cross <strong>the</strong>ir<br />
respective static objects at <strong>the</strong> same time. Thus, when an agent turns back after being<br />
perturbed by <strong>the</strong> o<strong>the</strong>r agent who has just passed its own static object, it does not find<br />
anything <strong>the</strong>re <strong>and</strong> does not get held back any fur<strong>the</strong>r. Similarly, <strong>the</strong> o<strong>the</strong>r agent will<br />
have remained oblivious to this occurrence as well. Moreover, <strong>the</strong> only time when <strong>the</strong>re<br />
will be no interference from <strong>the</strong> swapped receptor fields at all is when <strong>the</strong> agents engage<br />
in mutual interaction, as this interaction results in identical (matching) receptor<br />
activations.<br />
In sum, by modifying <strong>the</strong> original experimental setup in <strong>the</strong> current manner we have<br />
thus demonstrated that <strong>the</strong> interaction process not only makes interaction with <strong>the</strong><br />
shadow object unstable, <strong>the</strong>reby removing it as a possibility for fur<strong>the</strong>r entrainment, but<br />
that it also plays a role in making perceptual crossing a stable possibility. Even without<br />
any consistent sensory-motor correlations as a basis for individual behavior, <strong>the</strong> agents<br />
essentially negate this lack by means of mutually responsive interactions. In this manner<br />
it is possible for successful perceptual crossing to self-organize in terms of <strong>the</strong> relative<br />
stabilities of <strong>the</strong> interaction process. On this basis we can hypo<strong>the</strong>size that if we<br />
159 | P a g e
changed <strong>the</strong> psychological study accordingly, human participants would similarly be<br />
able to continue to accomplish <strong>the</strong> task successfully.<br />
9.2.3 Experimental setup 3: Conflicting behaviors<br />
It could be argued that <strong>the</strong> only reason why perceptual crossing emerges under <strong>the</strong><br />
modified conditions of experimental setup 2 is because <strong>the</strong> two agents actively establish<br />
<strong>the</strong> interaction process. After all, this is precisely what <strong>the</strong>y were originally evolved for,<br />
<strong>and</strong> <strong>the</strong> success of <strong>the</strong>ir strategy might <strong>the</strong>refore still be better accounted for by<br />
appealing to <strong>the</strong>ir individual behavioral efforts ra<strong>the</strong>r than to <strong>the</strong> self-organizing<br />
dynamics of <strong>the</strong> interaction process. How can we separate out <strong>the</strong> contribution of <strong>the</strong>se<br />
two factors?<br />
It is certainly <strong>the</strong> case that in <strong>the</strong> original experimental setup of Auvray, Lenay <strong>and</strong><br />
Stewart‟s psychological study, both <strong>the</strong> individual interactors <strong>and</strong> <strong>the</strong> interaction<br />
process are essentially „cooperating‟ toge<strong>the</strong>r in <strong>the</strong> successful completion of <strong>the</strong> task:<br />
(i) if one individual finds <strong>the</strong> o<strong>the</strong>r‟s shadow, <strong>the</strong>n <strong>the</strong> o<strong>the</strong>r will still be looking <strong>and</strong> <strong>the</strong><br />
shadow will move away, <strong>the</strong>reby preventing fur<strong>the</strong>r interactions, <strong>and</strong> (ii) if one<br />
individual finds <strong>the</strong> receptor field of <strong>the</strong> o<strong>the</strong>r, <strong>the</strong> o<strong>the</strong>r has effectively found that<br />
individual, too, <strong>the</strong>reby entailing fur<strong>the</strong>r interactions. It follows that <strong>the</strong> stability of <strong>the</strong><br />
interaction process in <strong>the</strong> experimental situation <strong>and</strong> <strong>the</strong> intentions of <strong>the</strong> individual<br />
interactors are reciprocally reinforcing. But just how important is <strong>the</strong> organization of <strong>the</strong><br />
interaction process for its own stability? Is its existence as a process largely supported<br />
by <strong>the</strong> behavior of <strong>the</strong> individual agents or does it also possess some self-organizing<br />
efficacy of its own?<br />
Fortunately, it is also possible to investigate a „competitive‟ situation in which <strong>the</strong><br />
interaction process self-sustains even despite <strong>the</strong> individual intentions of <strong>the</strong> interactors.<br />
Consider, for example, <strong>the</strong> case of self-perpetuating arguments in which all participants<br />
actually want to stop arguing. Such a case could provide strong support for <strong>the</strong>ories<br />
which propose that social interactions can be characterized by <strong>the</strong>ir autonomous<br />
dynamics (cf. De Jaegher & Di Paolo 2007). Note that <strong>the</strong> notion of a „competitive‟<br />
situation, as it is used here, refers to <strong>the</strong> competing goal-directedness of an individual<br />
160 | P a g e
interactor <strong>and</strong> <strong>the</strong> stability of <strong>the</strong> collective interaction process. Of course, this does not<br />
exclude <strong>the</strong> possibility that <strong>the</strong>se interactors <strong>the</strong>mselves also have conflicting intentions,<br />
but such inter-individual conflict is not necessary for an individual to be in conflict with<br />
<strong>the</strong> stability of <strong>the</strong> interaction process itself.<br />
In order to fur<strong>the</strong>r investigate <strong>the</strong> autonomous role of <strong>the</strong> interaction process within this<br />
particular experimental situation, we <strong>the</strong>refore need to change <strong>the</strong> basic setup of Auvray<br />
<strong>and</strong> colleagues to give rise to this kind of „competitive‟ situation. The task of detecting<br />
social contingency remains <strong>the</strong> same as before: <strong>the</strong> individuals must distinguish between<br />
those interactions that occur with <strong>the</strong> o<strong>the</strong>r‟s receptor field, <strong>and</strong> those that result from<br />
<strong>the</strong> mobile shadow object, as well as avoid any interaction with <strong>the</strong> static object.<br />
However, in contrast to <strong>the</strong> original psychological study, here <strong>the</strong> agents are required to<br />
stay with <strong>the</strong>ir partner‟s shadow object, ra<strong>the</strong>r than staying with <strong>the</strong> receptor field of<br />
<strong>the</strong>ir actual partner. The task is <strong>the</strong>refore to detect a certain kind of mobile object that<br />
gives rise to non-contingent interactions, a task that can only be achieved by detecting<br />
<strong>and</strong> avoid interactions with contingently responsive mobile objects.<br />
Due to <strong>the</strong> asymmetry inherent in this setup (i.e. agents face in opposite directions, but<br />
<strong>the</strong>ir shadows are displaced in <strong>the</strong> same direction), it is impossible for both participants<br />
to be interacting with each o<strong>the</strong>r‟s shadow at <strong>the</strong> same time. Therefore, in order to<br />
complete <strong>the</strong> task it is now necessary for <strong>the</strong> participants to avoid engaging in interindividual<br />
interaction with each o<strong>the</strong>r. This will not be easy because (i) engaging in<br />
perceptual crossing is still a relatively stable behavior, at least for as long as both<br />
interactors remain convinced that <strong>the</strong>y are interacting with <strong>the</strong> o<strong>the</strong>r‟s shadow, <strong>and</strong> (ii)<br />
crossing with <strong>the</strong> o<strong>the</strong>r‟s shadow remains inherently unstable, since that o<strong>the</strong>r<br />
participant will keep on looking for <strong>the</strong> shadow of its partner. In this manner we have<br />
created an experimental setup in which <strong>the</strong> intentions of <strong>the</strong> individuals <strong>and</strong> <strong>the</strong><br />
dynamics of <strong>the</strong> inter-individual interaction process are in direct conflict.<br />
In order to implement this setup in terms of a simulation model we used <strong>the</strong> same<br />
parameters as for experimental setup 1, but with <strong>the</strong> essential difference that <strong>the</strong><br />
function to evaluate an agent‟s fitness (cf. Equation 9-1) now measures <strong>the</strong> average<br />
distance to <strong>the</strong> o<strong>the</strong>r agent‟s shadow. Since <strong>the</strong> asymmetry of this setup means that <strong>the</strong><br />
161 | P a g e
evaluation function can now give different values for <strong>the</strong> two agents (i.e. <strong>the</strong>y cannot be<br />
in contact with each o<strong>the</strong>r‟s shadow at <strong>the</strong> same time), we decided to have each agent<br />
controlled by a different CTRNN. In o<strong>the</strong>r words, whereas in <strong>the</strong> previous experiments<br />
we evaluated a solution by having two identical CTRNNs cooperating on <strong>the</strong> task, here<br />
we are evaluating two solutions by having <strong>the</strong>m compete with each o<strong>the</strong>r during <strong>the</strong><br />
trials. More specifically, <strong>the</strong> evolutionary algorithm has been changed so that two<br />
r<strong>and</strong>omly selected parents are removed from <strong>the</strong> current generation, <strong>and</strong> tested against<br />
each o<strong>the</strong>r for 100 trials of evaluation. The losing solution is discarded. The winning<br />
solution <strong>and</strong> a mutated copy of its genome are added to <strong>the</strong> population of <strong>the</strong> next<br />
generation. This tournament is repeated until <strong>the</strong>re are no more parents in <strong>the</strong> current<br />
generation, at which point <strong>the</strong> process is repeated with <strong>the</strong> newly created population.<br />
It could be argued that this approach still does not represent proper competition between<br />
<strong>the</strong> solutions because of <strong>the</strong> likelihood of genetic convergence of <strong>the</strong> population (cf.<br />
<strong>Froese</strong> & Spier 2008). Genetic convergence could make <strong>the</strong> solutions almost identical,<br />
<strong>and</strong> <strong>the</strong>refore effectively turn this situation into a „cooperative‟ one once again, at least<br />
at <strong>the</strong> genetic level. Never<strong>the</strong>less, this worry is unnecessary as several attempts to<br />
evolve agents to solve this task under <strong>the</strong>se conditions did not succeed. The target<br />
solution appears to be too unstable to make such convergence possible, <strong>and</strong> even after<br />
thous<strong>and</strong>s of generations <strong>the</strong> evolved behavior is nowhere near as successful in terms of<br />
fitness as that evolved for experimental setup 1. These difficulties indicate that <strong>the</strong>re are<br />
limits to <strong>the</strong> ability of <strong>the</strong> interaction process to entrain <strong>the</strong> behavior of <strong>the</strong> individual<br />
agents so as to sustain appropriate patterns of interaction. Suitable initial conditions<br />
must be present for <strong>the</strong> emergence of a self-organizing interaction process.<br />
In response to <strong>the</strong>se difficulties of evolving a competitive scenario from scratch we<br />
chose a slightly more advantageous starting point for <strong>the</strong> evolutionary process, namely<br />
by seeding <strong>the</strong> populations with <strong>the</strong> best evolved agent from experimental setup 1. In<br />
this case we still have two contrasting influences on <strong>the</strong> behavior of <strong>the</strong> agents, i.e. a<br />
competitive fitness evaluation forcing <strong>the</strong> agents to disengage, <strong>and</strong> <strong>the</strong> entraining<br />
stability of <strong>the</strong> interaction process in <strong>the</strong> form of perceptual crossing. The results of<br />
comprehensive tests of <strong>the</strong> best agent taken from an evolutionary run that lasted almost<br />
5000 generations are shown in Figure 9-6a. The results for <strong>the</strong> same agent, but with<br />
162 | P a g e
swapped receptor fields, are shown in Figure 9-6b. Note that for <strong>the</strong> purposes of <strong>the</strong>se<br />
trials <strong>the</strong> agent competes against a clone of itself.<br />
(a)<br />
(b)<br />
Figure 9-6. Graphical representation of fitness scores achieved at each possible combination of starting<br />
positions for agent „up‟ (x-axis) <strong>and</strong> agent „down‟ (y-axis). Note that <strong>the</strong> axes wrap around due to <strong>the</strong> 1-D<br />
circular shape of <strong>the</strong> virtual environment. (a) Normal condition. Fitness scores range from 0.09 to 0.94<br />
with an average of 0.80. (b) Swapped receptor fields. Fitness scores range from 0.06 to 0.94 with an<br />
average of 0.80.<br />
The comprehensive results shown in Figure 9-6 indicate that in terms of fitness scores<br />
<strong>the</strong> best evolved solution is in many respects qualitatively similar to <strong>the</strong> solutions found<br />
for <strong>the</strong> previous experiments, though <strong>the</strong>re is a slight reduction in overall robustness.<br />
For example, <strong>the</strong>re is a noteworthy exception that can be seen as a small black square in<br />
<strong>the</strong> top left corner of Figure 9-6a, an area where this solution received almost no score<br />
because both agents got stuck on <strong>the</strong>ir respective static objects. However, this particular<br />
problematic situation is largely resolved when we modify this setup to <strong>the</strong> „swapped<br />
receptor field‟ condition, as is shown in Figure 9-6b. This result is consistent with what<br />
we have learned on <strong>the</strong> basis of experimental setup 2. Of course, some points of low<br />
fitness remain, but this is to be expected considering <strong>the</strong> conflicting influences shaping<br />
agent behavior.<br />
How are <strong>the</strong>se two competing factors reflected in <strong>the</strong> behaviors <strong>and</strong> mutual interactions<br />
of <strong>the</strong> agents? We find that <strong>the</strong> original perceptual crossing evolved for experimental<br />
setup 1 is retained across generations, although <strong>the</strong> precise strategy of <strong>the</strong> agents has<br />
163 | P a g e
adapted slightly to <strong>the</strong> new constraints. Figure 9-7 shows a sample trial run selected<br />
from Figure 9-6a.<br />
Figure 9-7. Illustration of <strong>the</strong> behavior of <strong>the</strong> agents during a representative trial starting from point (100,<br />
500) with a score of 0.72. They first encounter <strong>the</strong>ir respective static objects, <strong>the</strong>n continue searching, <strong>and</strong><br />
finally locate each o<strong>the</strong>r <strong>and</strong> establish perceptual crossing until <strong>the</strong> end of <strong>the</strong> run (16000 time steps). Top:<br />
<strong>the</strong> position of <strong>the</strong> agents <strong>and</strong> objects in <strong>the</strong> 1-D environment over time. Middle: <strong>the</strong> node outputs of agent<br />
„up‟ over time. Bottom: <strong>the</strong> status of <strong>the</strong> receptor field <strong>and</strong> <strong>the</strong> velocity of agent „up‟ over time. Note that<br />
a change of receptor field status reaches <strong>the</strong> agent‟s controller only after a delay of 25 units of time (250<br />
steps).<br />
It is revealing to compare this trial to <strong>the</strong> one illustrated in Figure 9-3. The scores of <strong>the</strong><br />
two trials are almost identical, but <strong>the</strong>re are some qualitative differences in behavior.<br />
The first thing to notice is that in this modified setup <strong>the</strong> agents make much less contact<br />
with <strong>the</strong> objects in <strong>the</strong>ir environment. For example, <strong>the</strong>y only make contact with <strong>the</strong>ir<br />
respective static objects once (ra<strong>the</strong>r than three times) before moving on. Similarly,<br />
perceptual crossing is established with only two contacts (ra<strong>the</strong>r than periods of three).<br />
These shorter periods of interaction might have evolved to make it easier for <strong>the</strong> agents<br />
164 | P a g e
to break away from <strong>the</strong> stable, but contingent interaction before becoming too entrained.<br />
The second thing to notice is that <strong>the</strong> perceptual crossing is now characterized by a<br />
certain spatial drift, whereas before it was localized to one area of <strong>the</strong> environment. This<br />
is due to <strong>the</strong> asymmetry introduced by <strong>the</strong> competitive fitness function (both agents<br />
have <strong>the</strong>ir shadows displaced in <strong>the</strong> same direction). For example, after engaging in<br />
basic perceptual crossing in between time steps 4000 <strong>and</strong> 8000, agent „down‟ traces<br />
along <strong>the</strong> shadow object of agent „up‟ until about 11000 time steps. At this point <strong>the</strong><br />
o<strong>the</strong>r agent, not having made any contact with agent „down‟ for a while, starts to speed<br />
up its rightward exploratory movement, <strong>and</strong> <strong>the</strong> agents are eventually forced back into<br />
establishing perceptual crossing.<br />
This inherent trade-off between (i) staying in contact with <strong>the</strong> o<strong>the</strong>r‟s shadow <strong>and</strong> (ii)<br />
<strong>the</strong> self-organizing stability of <strong>the</strong> interaction process is especially noticeable when<br />
agent „up‟ starts <strong>the</strong> trial located between agent „down‟ <strong>and</strong> its shadow (trial not<br />
depicted). In this case agent „up‟ moves rightwards, makes contact with <strong>the</strong> o<strong>the</strong>r‟s<br />
shadow object, <strong>and</strong> traces its movement leftwards for a short period. However, since<br />
agent „down‟ has not been perturbed during this activity it keeps increasing its leftward<br />
velocity until it eventually breaks away. In o<strong>the</strong>r words, lack of interaction makes this<br />
an unstable strategy <strong>and</strong> prevents <strong>the</strong> self-organizing dynamics of <strong>the</strong> interaction<br />
process to emerge. However, when <strong>the</strong> agents do make contact <strong>the</strong>y generally engage in<br />
a pattern of interaction similar to that shown in Figure 9-7, where short bursts of<br />
perceptual crossing are interspersed with periods of distancing. Here we thus have a<br />
situation in which <strong>the</strong> interaction process sustains itself, even though <strong>the</strong> individual<br />
agents have been specifically evolved to break this mutual interaction in favor of<br />
localizing <strong>the</strong> o<strong>the</strong>r‟s shadow.<br />
Of course, individual behavior still plays an enabling role in this situation. It is <strong>the</strong><br />
agents who make contact with each o<strong>the</strong>r, <strong>and</strong> thus allow <strong>the</strong> interaction process to take<br />
hold in <strong>the</strong> first place. And it is <strong>the</strong> agents who at times manage to break away from this<br />
entrainment as well. But never<strong>the</strong>less it is <strong>the</strong> interaction process which constrains <strong>the</strong>ir<br />
behavior most of <strong>the</strong> time, even despite <strong>the</strong>ir individual goals. On this basis we can<br />
hypo<strong>the</strong>size that if we changed <strong>the</strong> psychological study accordingly, human participants<br />
would encounter great difficulty in accomplishing <strong>the</strong> task successfully. To some extent<br />
165 | P a g e
this hypo<strong>the</strong>sis is already supported by <strong>the</strong> empirical data presented by Auvray <strong>and</strong><br />
colleagues who found in <strong>the</strong>ir original study that <strong>the</strong> probability of a participant‟s click<br />
after stimulation by ei<strong>the</strong>r <strong>the</strong> o<strong>the</strong>r‟s receptor field or his shadow object was not<br />
significantly different (cf. Auvray, et al. 2009, p. 39).<br />
9.3 Dynamical analysis<br />
The results of <strong>the</strong>se simulated experimental setups <strong>and</strong> <strong>the</strong> original empirical data<br />
presented by Auvray <strong>and</strong> colleagues undermine any attempt to attribute sensitivity to<br />
social contingency to <strong>the</strong> individual agents under <strong>the</strong>se conditions. It is <strong>the</strong> collective<br />
interaction process, itself enabled by simple exploratory <strong>and</strong> discriminatory individual<br />
behavior, which in turn enables <strong>and</strong> constrains appropriate „social‟ behavior on <strong>the</strong><br />
individual level so as to self-sustain its entraining presence. We know, for example, that<br />
human participants of <strong>the</strong> study cannot tell from <strong>the</strong>ir perspective alone whe<strong>the</strong>r <strong>the</strong>y<br />
are actually interacting with ano<strong>the</strong>r responsive partner or merely with an unresponsive<br />
copy. Yet <strong>the</strong>ir individual behavior is clearly influenced by <strong>the</strong> presence of social<br />
contingency as such, since on average <strong>the</strong>y spend more time interacting in contingent<br />
situations. How can we explain this interaction between <strong>the</strong> individual <strong>and</strong> collective<br />
levels of dynamics?<br />
In order to get a better underst<strong>and</strong>ing of how <strong>the</strong> individual <strong>and</strong> interaction levels relate<br />
to each o<strong>the</strong>r, let us for a moment entertain <strong>the</strong> traditional perspective of methodological<br />
individualism, i.e. we adopt <strong>the</strong> common perspective that <strong>the</strong> individual agent is <strong>the</strong><br />
only correct unit of analysis. The two behavioral tests described in Section 9.2.1 support<br />
<strong>the</strong> idea that an agent‟s discriminatory ability depends on <strong>the</strong> duration of <strong>the</strong> 3 rd contact.<br />
If this contact is too long it is a static object, if it is short enough <strong>the</strong>n it is <strong>the</strong> o<strong>the</strong>r or<br />
<strong>the</strong> o<strong>the</strong>r‟s shadow. We have also seen that this ambiguity in <strong>the</strong> latter half of <strong>the</strong><br />
discrimination is not a problem. Note, however, that already here we have to appeal to<br />
mutual responsiveness in order to discount <strong>the</strong> shadow as a serious possibility, since it<br />
only affords unstable interactions. Never<strong>the</strong>less, it still appears possible that <strong>the</strong> agent‟s<br />
controller is implementing a simple thresholding circuit that determines which of <strong>the</strong><br />
two behaviors, i.e. oscillation or exploration, should become active depending on <strong>the</strong><br />
166 | P a g e
length of stimulation. A hypo<strong>the</strong>tical circuit of this traditional internalist explanation is<br />
illustrated in Figure 9-8.<br />
Input I<br />
D1<br />
„body‟<br />
„brain‟<br />
K 1<br />
∫ I dt<br />
0<br />
T1<br />
T2<br />
G1<br />
G2<br />
left motor<br />
-<br />
+<br />
2<br />
right motor<br />
D2<br />
Velocity v<br />
„world‟<br />
Figure 9-8. Diagram of a hypo<strong>the</strong>tical circuit that could explain <strong>the</strong> discriminatory behavior of <strong>the</strong> agents.<br />
D1 <strong>and</strong> D2 are delays (total delay: 25 units of time), ∫ I dt is <strong>the</strong> integral of <strong>the</strong> input signal I for K units of<br />
time, T1 <strong>and</strong> T2 are thresholds, <strong>and</strong> G1 <strong>and</strong> G2 are output gains.<br />
We know from <strong>the</strong> implementation of <strong>the</strong> agents that <strong>the</strong>ir controller is separated from<br />
<strong>the</strong>ir environment by a delay, which can be represented as part of <strong>the</strong>ir body (D1 <strong>and</strong><br />
D2). We also know that <strong>the</strong> velocity v of <strong>the</strong> agents is <strong>the</strong> difference between left <strong>and</strong><br />
right motor activation. In addition, we have established that <strong>the</strong> crucial factor is duration<br />
of stimulation. Accordingly, we posit an integrator element, which takes <strong>the</strong> delayed<br />
input signal I <strong>and</strong> outputs <strong>the</strong> integral, calculated over K units of time. This integral is<br />
<strong>the</strong>n passed through a threshold unit which becomes active (i.e. it outputs 1, ra<strong>the</strong>r than<br />
0) when <strong>the</strong> integral falls between two thresholds T1 <strong>and</strong> T2. The initial behavioral tests<br />
described in Section 9.2.1 allow us to set T2 at about 105 time steps. The output of <strong>the</strong><br />
threshold unit is <strong>the</strong>n multiplied by <strong>the</strong> left motor gain G1 <strong>and</strong> subtracted from <strong>the</strong><br />
agent‟s overall velocity. At <strong>the</strong> same time we know that when <strong>the</strong> agent receives no<br />
input (I = 0) it continues to move rightwards. Accordingly, we posit ano<strong>the</strong>r connection<br />
from <strong>the</strong> delayed input signal that passes through an inverter, gets multiplied by <strong>the</strong><br />
right motor gain G2, <strong>and</strong> is added to <strong>the</strong> agent‟s overall velocity.<br />
Is it possible to determine <strong>the</strong> precise values for <strong>the</strong>se parameters, especially T1 <strong>and</strong> T2?<br />
Let us consider a trial from <strong>the</strong> experimental setup 1 in which <strong>the</strong> agents did not do very<br />
well in order to get an idea of <strong>the</strong> range of <strong>the</strong>se values. We can hypo<strong>the</strong>size, for<br />
167 | P a g e
example, that <strong>the</strong> low fitness found in Region 6 of Figure 9-2 can be explained in terms<br />
of a long contact time between <strong>the</strong> agents. To test this hypo<strong>the</strong>sis we chose a<br />
representative trial starting from (557, 32). In this case <strong>the</strong> agents spontaneously<br />
disengage <strong>the</strong>ir interaction (after 2000 time steps) after having found each o<strong>the</strong>r in <strong>the</strong><br />
very beginning of <strong>the</strong> trial. They <strong>the</strong>n continue to explore <strong>the</strong> environment, finally reestablish<br />
perceptual crossing (after 10000 time steps), <strong>and</strong> continue to interact<br />
appropriately until <strong>the</strong> end of <strong>the</strong> trial.<br />
This initial coordination failure highlights <strong>the</strong> importance of co-regulation for<br />
perceptual crossing to be established. It is not enough for <strong>the</strong> agents to simply encounter<br />
each o<strong>the</strong>r, but <strong>the</strong>y also have to encounter each o<strong>the</strong>r with <strong>the</strong> right kind of velocity. It<br />
is also interesting to note that <strong>the</strong> decisive contact was <strong>the</strong> 4 th encounter in this case<br />
ra<strong>the</strong>r than <strong>the</strong> usual 3 rd . Might this have something to do with <strong>the</strong> cause of <strong>the</strong><br />
spontaneous failure? Initially this idea appears to be rejected by a quick test of<br />
decreasing <strong>the</strong> size of both agents to 3 units of space (instead of <strong>the</strong> normal 4) just<br />
before <strong>the</strong> 4 th encounter, which results in a performance with an almost maximum score.<br />
It is a matter of stimulation length after all, but <strong>the</strong> number of contacts appears to be less<br />
important. Moreover, this modified contact lasts 55 steps, which is short even for <strong>the</strong><br />
duration of normal contact with ano<strong>the</strong>r agent. Might this be an indication for <strong>the</strong> value<br />
of T1? However, <strong>the</strong>re is a serious problem for our hypo<strong>the</strong>sis: <strong>the</strong> original final<br />
encounter between <strong>the</strong> two agents only lasted 64 steps as well. This value is still within<br />
<strong>the</strong> range of contact durations with <strong>the</strong> o<strong>the</strong>r agent that we determined in <strong>the</strong> previous<br />
series of tests (between T1 <strong>and</strong> T2). So why do <strong>the</strong> agents disengage?<br />
At this point we have to reject <strong>the</strong> possibility that we can explain <strong>the</strong> agent‟s sensitivity<br />
to social contingency in terms of an objective cut-off point of contact duration, <strong>and</strong> turn<br />
to a more dynamical explanation. As a first step in this direction, it is important to get<br />
an idea of <strong>the</strong> general shape of <strong>the</strong> dynamical l<strong>and</strong>scape of <strong>the</strong> CTRNN shown in Figure<br />
9-1. When <strong>the</strong> CTRNN is decoupled from <strong>the</strong> 1-D environment it is characterized by<br />
two fixed point attractors, depending on whe<strong>the</strong>r <strong>the</strong> input parameter I is set to 0 or 1<br />
(see Figure 9-9). It turns out that <strong>the</strong> velocity of <strong>the</strong> agents is strongly coupled to <strong>the</strong><br />
value of this parameter. The velocity of <strong>the</strong> agent at attractor 0 , when input I = 0, is 0.86<br />
units of space per unit of time, whereas for <strong>the</strong> attractor 1 , when I = 1, <strong>the</strong> velocity is -<br />
168 | P a g e
0.96. This is indeed <strong>the</strong> basis for a tight sensory-motor coupling: <strong>the</strong> value of <strong>the</strong> input<br />
parameter is largely determined by <strong>the</strong> movement of <strong>the</strong> agent, <strong>and</strong> <strong>the</strong> movement of <strong>the</strong><br />
agent is largely determined by <strong>the</strong> input parameter (for a similar result on <strong>the</strong> basis of a<br />
related task, see <strong>Froese</strong> & Di Paolo 2008a).<br />
(a)<br />
(b)<br />
Figure 9-9. The attractor l<strong>and</strong>scape for <strong>the</strong> CTRNN shown in Figure 9-1. For 50 times <strong>the</strong> node<br />
activations were initialized to r<strong>and</strong>om activation values drawn from <strong>the</strong> trial shown in Figure 9-3, <strong>and</strong> <strong>the</strong><br />
network was allowed to settle for 8000 time steps. The input was ei<strong>the</strong>r (a) set to 0, which revealed a<br />
fixed point attractor at (0.04, 0.91, 0.02), or (b) set to 1, which revealed a fixed point attractor at (0.98,<br />
0.02, 0.88). The attractors are represented by a „*‟. Input-driven switching between <strong>the</strong> two attractors<br />
results in a hysteresis of motor outputs.<br />
It is also noteworthy that <strong>the</strong> switching behavior between <strong>the</strong> two attractors is<br />
characterized by a form of hysteresis (see Figure 9-9 <strong>and</strong> Figure 9-10). We will explain<br />
this behavior by means of binary approximation of <strong>the</strong> sigmoided outputs (left motor<br />
node, right motor node, node 3). The only way for <strong>the</strong> system to settle at attractor 0 (0, 1,<br />
0), for instance, is to approach it from (0, 0, 0). Thus, if input I = 0 <strong>and</strong> <strong>the</strong> system<br />
happens to be at (1, 1, 0), <strong>the</strong>n it first needs to pass through (0, 1, 0) <strong>and</strong> (0, 0, 0) before<br />
finally reaching (0, 1, 0). Similarly, for <strong>the</strong> system to settle at attractor 1 , if input I = 1<br />
<strong>and</strong> <strong>the</strong> system currently happens to be at attractor 0 (0, 1, 0), <strong>the</strong>n it first switches to (1,<br />
1, 0) before finally shifting toward attractor 1 at (1, 0, 1).<br />
It is also important to note that <strong>the</strong>re are significant differences in <strong>the</strong> trajectory speeds<br />
of <strong>the</strong> different regions of activation space, something which cannot be seen in Figure<br />
9-9 (but cf. Figure 9-10). In <strong>the</strong> case of attractor 0 <strong>the</strong> trajectories going from (1, 1, 0) to<br />
169 | P a g e
(0, 1, 0) are initially relatively slow. But when <strong>the</strong> output of <strong>the</strong> right motor node drops<br />
into <strong>the</strong> region of somewhat below 0.5, <strong>the</strong> system almost instantaneously switches off<br />
<strong>the</strong> output of <strong>the</strong> left motor node. Then <strong>the</strong> trajectories continue to slowly make <strong>the</strong>ir<br />
way from somewhere near (0, 0, 0) to <strong>the</strong> attractor 0 at (0, 1, 0). In <strong>the</strong> case of attractor 1 ,<br />
if <strong>the</strong> system happens to be near attractor 0 at (0, 1, 0), <strong>the</strong>n <strong>the</strong> system almost<br />
immediately switches on <strong>the</strong> left motor by going into state (1, 1, 0), <strong>and</strong> only <strong>the</strong>n<br />
slowly begins to turn off <strong>the</strong> right motor as it approaches <strong>the</strong> attractor 1 at (1, 0, 1). This<br />
effectively means that <strong>the</strong> time taken to switch between output velocities depends not<br />
only on <strong>the</strong> current state of <strong>the</strong> input parameter, but is also determined by <strong>the</strong> current<br />
state of <strong>the</strong> system. In o<strong>the</strong>r words, <strong>the</strong> behavior of <strong>the</strong> agents is not purely reactive to<br />
<strong>the</strong> input parameter but, due to <strong>the</strong> hysteresis of <strong>the</strong> motor outputs with different<br />
trajectory speeds, crucially depends on <strong>the</strong> agent‟s history of interactions. The influence<br />
of internal state is completely missed out in <strong>the</strong> hypo<strong>the</strong>tical circuit diagram shown in<br />
Figure 9-8.<br />
Moreover, this historical dimension in terms of <strong>the</strong> influence of previous interactions<br />
brings us to <strong>the</strong> next step of our dynamical analysis, since up to now we have only<br />
considered <strong>the</strong> CTRNN in isolation. How does <strong>the</strong> system behave when it is coupled to<br />
<strong>the</strong> 1-D environment <strong>and</strong> interacts with o<strong>the</strong>r objects including <strong>the</strong> o<strong>the</strong>r agent? The<br />
change in state of <strong>the</strong> system during <strong>the</strong> first 8000 time steps of <strong>the</strong> trial starting from<br />
position (100, 500), as was shown in Figure 9-3, is of interest here. How does <strong>the</strong><br />
dynamical analysis help us to better underst<strong>and</strong> what is going on?<br />
Let us consider <strong>the</strong> hysteresis relationship between <strong>the</strong> left <strong>and</strong> right motors during this<br />
trial, as shown in Figure 9-10. Before engaging in a new interaction in <strong>the</strong> environment,<br />
<strong>the</strong> system is usually near attractor 0 (top-left) <strong>and</strong> thus with <strong>the</strong> output of <strong>the</strong> right motor<br />
on full power. When <strong>the</strong> input parameter I changes from 0 to 1 <strong>the</strong> system jumps to<br />
Region 1, where it starts to slowly pass down through Region 2. Here both motors are<br />
competing somewhat <strong>and</strong> thus slow down <strong>the</strong> agent on <strong>the</strong> way back toward <strong>the</strong> source<br />
of stimulation. This typically results in ano<strong>the</strong>r contact that is more extended, <strong>and</strong> which<br />
pushes <strong>the</strong> system far down into Region 3. From this region in state space <strong>the</strong> activation<br />
of <strong>the</strong> left motor node decays rapidly toward 0, <strong>the</strong>reby returning <strong>the</strong> agent to its initial<br />
rightwards motion, <strong>and</strong> slowly moving its state upwards to attractor 0 .<br />
170 | P a g e
Region 1<br />
Region 2<br />
Region 3<br />
Figure 9-10. Change of state between <strong>the</strong> left <strong>and</strong> right motors during <strong>the</strong> trial that was shown in Figure<br />
9-3. Attractor 0 <strong>and</strong> attractor 1 are marked by „*‟ in <strong>the</strong> top-left <strong>and</strong> bottom-right, respectively. Arrows<br />
indicate direction <strong>and</strong> relative velocity of trajectories (i.e. long arrows = fast trajectories, short arrows =<br />
slow trajectories). For an explanation of Regions 1-3 in <strong>the</strong> state space, see <strong>the</strong> description in <strong>the</strong> text.<br />
While this description of <strong>the</strong> hysteresis between <strong>the</strong> two motor nodes explains <strong>the</strong><br />
observed oscillatory behavior of <strong>the</strong> agent when it passes objects in its environment, it<br />
does not yet explain how <strong>the</strong> agent distinguishes between <strong>the</strong> static object <strong>and</strong> <strong>the</strong> o<strong>the</strong>r<br />
agent. We have already determined that <strong>the</strong> third contact is crucial for this decision<br />
process. How can we explain this dynamically?<br />
The third contact happens after <strong>the</strong> agent has made initial contact, turned around for<br />
ano<strong>the</strong>r contact, <strong>and</strong> is now on its way to return to its original rightward velocity. In<br />
o<strong>the</strong>r words, <strong>the</strong> system is still tracing its way back up toward attractor 0 when it gets<br />
perturbed again, <strong>and</strong> this pushes it directly across into Region 2 (ra<strong>the</strong>r than Region 1,<br />
which happens after it is perturbed for <strong>the</strong> first time). The duration of stimulation during<br />
this encounter determines how far down in <strong>the</strong> state space from Region 2 to Region 3<br />
<strong>the</strong> system will end up. The more stimulation, <strong>the</strong> fur<strong>the</strong>r toward Region 3, <strong>and</strong> <strong>the</strong> more<br />
quickly <strong>the</strong> activation of <strong>the</strong> left motor node will decay, rapidly shifting <strong>the</strong> state back<br />
leftwards. In this case <strong>the</strong> system will quickly resume its rightward velocity. In o<strong>the</strong>r<br />
words, a lengthy stimulation results in a significantly quicker return to rightward<br />
velocity, which prevents ano<strong>the</strong>r contact with <strong>the</strong> object to occur, <strong>and</strong> <strong>the</strong> agent<br />
<strong>the</strong>refore moves away. O<strong>the</strong>rwise, if <strong>the</strong> third simulation is short enough, for example<br />
due to <strong>the</strong> responsiveness of <strong>the</strong> o<strong>the</strong>r agent, <strong>the</strong> system will spend some time slowly<br />
moving down Region 2, before eventually reaching Region 3 <strong>and</strong> <strong>the</strong>n switching<br />
171 | P a g e
quickly as before. This difference in time needed for <strong>the</strong> left motor to become<br />
deactivated, which is provided by <strong>the</strong> responsive counter-movement of <strong>the</strong> o<strong>the</strong>r agent,<br />
is essentially at <strong>the</strong> basis of <strong>the</strong> agent‟s ability to distinguish between its static object<br />
<strong>and</strong> that o<strong>the</strong>r agent.<br />
In summary, contrary to <strong>the</strong> methodologically individualistic perspective that we briefly<br />
entertained at <strong>the</strong> start of this dynamical analysis, we found that <strong>the</strong> discriminatory<br />
ability exhibited by <strong>the</strong> agents only emerges during interaction. The processes that drive<br />
<strong>the</strong> necessary internal state changes via appropriate input-switching are external to <strong>the</strong><br />
agent, <strong>and</strong> in this case <strong>the</strong>y are partly constituted by <strong>the</strong> responsive behavior of <strong>the</strong> o<strong>the</strong>r<br />
agent. An agent in an empty 1-D environment would be forever doomed to linear<br />
movement, lacking <strong>the</strong> ability to internally switch between <strong>the</strong> two attractor l<strong>and</strong>scapes.<br />
Moreover, <strong>the</strong> hypo<strong>the</strong>tical circuit of <strong>the</strong> agent‟s internal operations turned out to be<br />
wholly inadequate. A more detailed dynamical analysis failed to find internal threshold<br />
mechanisms, but instead revealed <strong>the</strong> important role of different temporal scales<br />
distributed across <strong>the</strong> state-space, as well as <strong>the</strong> externalization of processes necessary<br />
for <strong>the</strong> agent‟s discriminatory ability. These external processes did not perturb <strong>the</strong><br />
system along trajectories on a fixed state-space, but caused <strong>the</strong> entire state-space itself<br />
to switch between different attractor l<strong>and</strong>scapes. The behavior of <strong>the</strong> agents was thus<br />
largely determined by dynamical transients ra<strong>the</strong>r than fixed attractors.<br />
9.4 Discussion<br />
The modeling results presented in this chapter have provided support for <strong>the</strong> idea that<br />
<strong>the</strong> organization of an inter-individual interaction process can enable <strong>and</strong> constrain<br />
individual behavior in ways that are beneficial for problem-solving. With experimental<br />
setup 1 we replicated <strong>the</strong> findings of previous studies (Auvray, et al. 2009; Di Paolo, et<br />
al. 2008), namely that <strong>the</strong> interaction process can enable <strong>the</strong> agents to complete a task<br />
that appears impossible from an individual‟s perspective. The emergence of ongoing<br />
perceptual crossing between two interacting agents gives rise to a form of multi-agent<br />
interaction that depends on <strong>the</strong> ongoing interaction process, <strong>and</strong> at <strong>the</strong> same time also<br />
makes it more likely for that kind of mutual interaction to persist. This reciprocal<br />
dependency between individual agent behavior <strong>and</strong> overall interaction dynamics in this<br />
172 | P a g e
modeling experiment is a paradigmatic example of <strong>the</strong> constitutive autonomy of <strong>the</strong><br />
interaction process (De Jaegher & <strong>Froese</strong> 2009).<br />
The fact that <strong>the</strong> results of Auvray <strong>and</strong> colleagues can be replicated in a simulation<br />
model involving relatively simple interacting dynamical systems indicates that such<br />
autonomous interaction processes (<strong>and</strong> <strong>the</strong>ir enabling/constraining effects) might be<br />
much more pervasive than at first assumed. Thus, while De Jaegher <strong>and</strong> Di Paolo (2007)<br />
illustrate <strong>the</strong>ir enactive approach to social cognition with examples drawn from human<br />
interactions, <strong>the</strong>se models make a plausible case that more basic forms of <strong>life</strong> can give<br />
rise to autonomous interaction processes, too. Indeed, it turns out that it is not even<br />
necessary for an agent involved in such an interaction to intend to interact with ano<strong>the</strong>r<br />
agent 33 . Interactions in a multi-agent system are sufficient to effectively organize<br />
individual behavior into joint actions that exceed <strong>the</strong> capacities of each agent alone,<br />
even without <strong>the</strong> agents realizing that this is <strong>the</strong> case. As such, this model provides<br />
concrete support for <strong>the</strong> claim that <strong>the</strong> interaction process has <strong>the</strong> capacity to exp<strong>and</strong> an<br />
agent‟s behavioral domain, even in <strong>the</strong> most minimal cases of multi-agent interaction.<br />
We have shown how <strong>the</strong> interaction process is constitutive of <strong>the</strong> agents‟ successful<br />
strategy, an essential contribution that remains hidden when focusing on <strong>the</strong> behavior of<br />
an individual alone. If this constitutive impact is relevant for even such minimal forms<br />
of interaction, we can begin to wonder: How much of abstract reasoning, <strong>the</strong> traditional<br />
hallmark of human cognition, is based on our being embodied <strong>and</strong> embedded in <strong>the</strong><br />
complex multi-agent systems that define our socio-cultural context? This model can<br />
thus provide us with a first sense of how sociality can play an important explanatory<br />
role in defending <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis.<br />
In experimental setup 2, we disrupted <strong>the</strong> sensory-motor loops of <strong>the</strong> modeled agents in<br />
such a way that <strong>the</strong>y could no longer properly regulate <strong>the</strong>ir individual behaviors.<br />
However, engaging in reciprocal perceptual crossing remained a possibility within this<br />
modified setup. The results show that <strong>the</strong> agents managed to complete <strong>the</strong> task in spite<br />
33 Moreover, in <strong>the</strong> case of <strong>the</strong>se model agents, <strong>the</strong>y do not even fulfill <strong>the</strong> requirement for constitutive<br />
autonomy (since <strong>the</strong>ir identity is externally defined), <strong>and</strong> thus <strong>the</strong>ir interaction does not strictly satisfy <strong>the</strong><br />
definition of multi-agent interaction developed in Chapter 4. On <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>, could this be a model for<br />
<strong>the</strong> emergence of an autonomous entity out of non-autonomous elements (cf. <strong>Froese</strong> & Di Paolo 2008b)?<br />
173 | P a g e
of this sensory-motor disruption. It turns out that <strong>the</strong>ir remaining interactional capacity<br />
was sufficient for <strong>the</strong> emergence of <strong>the</strong> relevant interaction dynamics, which <strong>the</strong>n<br />
effectively organized <strong>the</strong> impaired individual behaviors appropriately. This is an<br />
indication that, given <strong>the</strong> right circumstances, an interaction process can enable <strong>and</strong><br />
organize <strong>the</strong> capacities of <strong>the</strong> interactors in such a way that individual impairment is<br />
overcome in a collective manner, even without <strong>the</strong> need for external control. Here we<br />
might have <strong>the</strong> beginning of an explanation of why human subjects with impaired<br />
sensory-motor capacities can perform normally under social conditions (cf. Chapter 6,<br />
pp. 104-111).<br />
In experimental setup 3 we modified <strong>the</strong> task of <strong>the</strong> agents in such a way that <strong>the</strong>ir<br />
individual behaviors <strong>and</strong> <strong>the</strong> overall interaction dynamics are in conflict, namely by<br />
requiring <strong>the</strong> agents to interact with <strong>the</strong> o<strong>the</strong>r‟s shadow (an inherently unstable situation<br />
in this setup). The results for this experiment give fur<strong>the</strong>r support to De Jaegher <strong>and</strong> Di<br />
Paolo‟s (2007) claim that, under some circumstances, an interaction process can<br />
constrain <strong>the</strong> behaviors of <strong>the</strong> interacting agents such that it continues to persist even<br />
despite <strong>the</strong> efforts of <strong>the</strong> individual interactors. In <strong>the</strong> simulation model, <strong>the</strong> agents<br />
„struggle‟ to stay close to <strong>the</strong> o<strong>the</strong>r‟s shadow object, <strong>and</strong> never<strong>the</strong>less continually fall<br />
back into <strong>the</strong> more stable interaction pattern of mutual perceptual crossing. This<br />
experiment indicates that we should be careful when assessing responsibility for an<br />
individual‟s behavior. The outcome of our actions can be unconsciously constrained in<br />
undesirable directions by certain processes existing in our social context, thus leading to<br />
conflict despite our intentions to behave o<strong>the</strong>rwise.<br />
Lastly, a lesson to be learnt from <strong>the</strong> dynamical analysis presented in Section 9.3 is that<br />
we should be wary of positing hypo<strong>the</strong>tical cybernetic circuits (box diagrams) that could<br />
explain observed behavior. As Hurley writes in relation to her own „shared circuits<br />
model‟ (SCM) of cognition: “While SCM is described cybernetically, dynamic systems<br />
<strong>the</strong>ory could represent interactions of its implementing neural processes <strong>and</strong> embodied<br />
activity over time as evolution of a phase space, <strong>and</strong> investigate its attractor structure”<br />
(Hurley 2008, p. 20). In this simple example we determined that <strong>the</strong> system was<br />
sensitive to <strong>the</strong> duration of a stimulus, posited a comparator mechanism on <strong>the</strong><br />
operational level, <strong>and</strong> finally had to concede that no such mechanism, as a reified unit of<br />
174 | P a g e
operation, exists within <strong>the</strong> actual system. The observed sensitivity is an emergent<br />
outcome of <strong>the</strong> shape of <strong>the</strong> state space of <strong>the</strong> system <strong>and</strong> <strong>the</strong> temporality of <strong>the</strong><br />
environment with which it is coupled. It is <strong>the</strong> interaction process between <strong>the</strong> agents<br />
which constrains <strong>the</strong>ir behavioral response so as to sustain this interaction. The duration<br />
of stimulation is not an independent feature of <strong>the</strong> agent‟s environment, but something<br />
which is co-determined through <strong>the</strong> velocities of both agents <strong>and</strong> <strong>the</strong>ir manner of<br />
interacting. Thus, at best such models might help us to make <strong>the</strong> system‟s operations<br />
more intelligible, but <strong>the</strong>re is <strong>the</strong> danger that we are merely providing a re-description of<br />
<strong>the</strong> behavior which emerges from <strong>the</strong> system‟s interactions with its environment (a<br />
danger we have already alluded to in terms of <strong>the</strong> ER models presented in Chapter 7, pp.<br />
114-120). It would be a category mistake to reify components of such a re-description<br />
as real entities existing at <strong>the</strong> sub-personal level.<br />
9.5 Summary<br />
We replicated a recent simulation model of a minimalist experiment in perceptual<br />
crossing <strong>and</strong> confirmed <strong>the</strong> results with significantly simpler artificial agents. A series<br />
of psycho-physical tests of <strong>the</strong>ir behavior informed a hypo<strong>the</strong>tical circuit model of <strong>the</strong>ir<br />
internal operation. However, a detailed study of <strong>the</strong> actual internal dynamics reveals this<br />
circuit model to be unfounded, <strong>the</strong>reby offering a tale of caution for those hypo<strong>the</strong>sizing<br />
about sub-personal processes in terms of behavioral observations (e.g. additional neural<br />
mechanisms to account for Ian‟s gesturing, cf. Chapter 6, pp. 104-111). In particular, it<br />
has been shown that <strong>the</strong> appropriate behavior of <strong>the</strong> agents largely emerges out of <strong>the</strong><br />
interaction process itself ra<strong>the</strong>r than being an individual achievement alone.<br />
We also extended <strong>the</strong> original simulation model in two novel directions in order to<br />
fur<strong>the</strong>r test <strong>the</strong> extent to which perceptual crossing between agents can self-organize in<br />
a robust manner. These modeling results suggest new hypo<strong>the</strong>ses that can become <strong>the</strong><br />
basis for fur<strong>the</strong>r psychological experiments. This chapter <strong>the</strong>reby has contributed to <strong>the</strong><br />
ongoing efforts to establish a mutually informative dialogue between psychology <strong>and</strong><br />
evolutionary robotics (cf. Rohde 2008), especially in order to investigate <strong>the</strong> dynamics<br />
of social interaction.<br />
175 | P a g e
The simulation experiments presented in this paper lend support to <strong>the</strong> enactive<br />
approach to social cognition. The results show that, under some circumstances, an<br />
interaction process can take on a self-organizing identity of its own, <strong>and</strong> that such a<br />
process can effectively enable <strong>and</strong> constrain <strong>the</strong> behavioral repertoire of <strong>the</strong> individuals<br />
involved in <strong>the</strong> interaction. More precisely, <strong>the</strong> experiments indicate (i) that some<br />
interaction processes can beneficially extend <strong>the</strong> behavioral domain of individuals in<br />
novel directions without requiring any form of external supervision, <strong>and</strong> conversely (ii)<br />
that some interaction processes can organize behavioral repertoires in ways that are in<br />
conflict with an individual‟s aims. More research is needed in order to better underst<strong>and</strong><br />
<strong>the</strong> circumstances which lead to one or <strong>the</strong> o<strong>the</strong>r situation. Future work might eventually<br />
help us to better structure our social environment such that beneficial situations are<br />
more likely to emerge spontaneously.<br />
176 | P a g e
10 Investigating social interaction<br />
The original experimental setup by Auvray, et al. (2009) presented participants with a<br />
task that is largely epistemic in nature, i.e. click when you think that you have located<br />
<strong>the</strong> o<strong>the</strong>r participant. Under <strong>the</strong> circumstances this kind of task invites an individual<br />
strategy based on detached reflection <strong>and</strong> action. The surprising result is that <strong>the</strong><br />
organization of <strong>the</strong> setup made it possible for <strong>the</strong> participants to solve <strong>the</strong> task with<br />
which <strong>the</strong>y were individually faced in a collective manner, though <strong>the</strong>y were <strong>the</strong>mselves<br />
unaware of <strong>the</strong> collective nature of <strong>the</strong>ir success. The autonomy of <strong>the</strong> interaction<br />
process itself ensured that <strong>the</strong>ir individual strategies were effectively complementary in<br />
<strong>the</strong>ir impact. The modeling experiments presented in <strong>the</strong> previous chapter presented<br />
fur<strong>the</strong>r evidence for <strong>the</strong> robustness of this effect.<br />
But are <strong>the</strong>se modeling experiments of <strong>the</strong> previous chapter also investigations into<br />
social interaction as we have defined it in Chapter 4 (cf. pp. 64-71)? It appears that what<br />
we have been investigating so far is <strong>the</strong> kind of interaction we have defined as a multiagent<br />
interaction (cf. Chapter 4, pp. 58-64). The existence of an autonomous interaction<br />
process between two or more agents is not enough. What is missing is a co-regulated<br />
act. In o<strong>the</strong>r words, a social act requires that a participant regulates <strong>the</strong>ir sensory-motor<br />
coupling such that <strong>the</strong> behavior is completed by <strong>the</strong> regulation of at least one o<strong>the</strong>r<br />
agent. Only with this particular kind of co-regulation of interaction does an agent‟s<br />
behavior qualify as properly social, ra<strong>the</strong>r than being essentially a solitary behavior that<br />
sometimes just happens to involve ano<strong>the</strong>r individual. Accordingly, <strong>the</strong> experimental<br />
setups we have investigated in <strong>the</strong> previous chapter have given us insights into <strong>the</strong><br />
dynamics of <strong>the</strong> interaction process <strong>and</strong> its power to organize <strong>the</strong> behavior of individual<br />
agents even without <strong>the</strong> presence of social interaction. This is an achievement in its own<br />
right, as well as an important step toward showing that a consideration of sociality<br />
enables us to address <strong>the</strong> „cognitive gap‟ of <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis from <strong>the</strong><br />
bottom-up.<br />
In this chapter we will continue along this path by using an evolutionary robotics (ER)<br />
approach to fine-tune <strong>the</strong> original experimental design so as to determine <strong>the</strong> minimal<br />
conditions under which it becomes possible to study social interaction within a<br />
177 | P a g e
perceptual crossing setup. First, we will present two modeling experiments that<br />
motivate changes to <strong>the</strong> original setup by fur<strong>the</strong>r marginalizing <strong>the</strong> role of individualbased<br />
strategies, <strong>and</strong> <strong>the</strong>n we report on a modeling experiment which motivates a<br />
change to <strong>the</strong> task given to <strong>the</strong> participants in order to encourage social interaction. The<br />
chapter concludes with a novel hypo<strong>the</strong>sis about <strong>the</strong> role of social interaction in<br />
detecting social contingency that is open to empirical verification.<br />
10.1 Experimental setup 4: Infinitely small objects<br />
It has been demonstrated in Chapter 9 that <strong>the</strong> simulated agents make use of <strong>the</strong><br />
duration of contact with objects in order to discriminate interactions with a static object<br />
(always same length of stimulation for same velocity) <strong>and</strong> <strong>the</strong> o<strong>the</strong>r agent (potentially<br />
shorter or longer stimulation, depending on whe<strong>the</strong>r <strong>the</strong> o<strong>the</strong>r passes by in in-phase or<br />
anti-phase movement). This is a reliable basis of distinction because <strong>the</strong> o<strong>the</strong>r agent is<br />
always moving. It is unlikely, however, that this is <strong>the</strong> main strategy employed by<br />
humans in <strong>the</strong> original psychological study.<br />
To be sure, during <strong>the</strong> training phase <strong>the</strong> participants were asked to interact with a fourpixel<br />
wide object in three conditions. The target object was ei<strong>the</strong>r (i) static, (ii) moving<br />
at a constant speed of 15 units/second, or (iii) moving at a constant speed of 30<br />
units/second, <strong>and</strong> each of <strong>the</strong>se one min. training phases was announced as such. It<br />
could thus be possible that <strong>the</strong> participants learned <strong>the</strong> correlation between contact<br />
duration <strong>and</strong> whe<strong>the</strong>r an object is static or moving. In practice, however, <strong>the</strong> difference<br />
in duration is small enough such that it is unlikely to be <strong>the</strong> main strategy of <strong>the</strong><br />
participants, though <strong>the</strong>re is some evidence that shorter contact duration made a<br />
difference, leading to 31.3% of clicking response (cf. event E6 [1, p. 40]). Still, we<br />
hypo<strong>the</strong>size that <strong>the</strong> successful behavior of <strong>the</strong> participants could be based on different<br />
types of interactions afforded by <strong>the</strong> static object <strong>and</strong> <strong>the</strong> o<strong>the</strong>r active participant, ra<strong>the</strong>r<br />
than <strong>the</strong>ir differing durations of isolated contacts.<br />
We know from <strong>the</strong> experiments in Chapter 9 that even when a difference in duration of<br />
contact was detectable, <strong>the</strong> evolved strategy still depended constitutively on some coregulatory<br />
activity of <strong>the</strong> agents, even though <strong>the</strong>y were greatly aided by an external<br />
178 | P a g e
factor: body <strong>and</strong> object size. The question we want to address is: can we use ER to<br />
investigate <strong>the</strong> kinds of strategies that are available when such a duration-based strategy<br />
is excluded from <strong>the</strong> experimental design? One way to approach this is to make all<br />
objects (i.e. agents, shadow objects, <strong>and</strong> static objects) within <strong>the</strong> virtual environment<br />
infinitely small. This can be done by simply checking whe<strong>the</strong>r <strong>the</strong> sign of <strong>the</strong> difference<br />
of <strong>the</strong> locations (of <strong>the</strong> agent <strong>and</strong> some target object) has changed compared to <strong>the</strong><br />
previous time step. If <strong>the</strong> sign has changed, <strong>the</strong>n we activate <strong>the</strong> agent‟s receptor field.<br />
Since in this case all objects afford an equal duration of contact (i.e. 1 time step) no<br />
matter <strong>the</strong> velocity, it is no longer possible for <strong>the</strong> agents to trivially rely on <strong>the</strong> fact that<br />
o<strong>the</strong>r moving objects entail a shorter (or longer) duration of contact. Can we use ER to<br />
generate a strategy that enables <strong>the</strong> agents to successfully locate each o<strong>the</strong>r even under<br />
this more ambiguous situation?<br />
We successfully evolved a 6-node CTRNN to cope with this modified experimental<br />
setup. To make <strong>the</strong> solutions more evolvable it was necessary to include a large<br />
magnitude of input gains (range [-1000, 1000]) so as to compensate for <strong>the</strong> minimal<br />
period of stimulation, <strong>and</strong> reducing <strong>the</strong> amount of sensory delay to 5 units of time was<br />
also helpful. As in <strong>the</strong> previous modeling experiments, <strong>the</strong> solutions were evaluated in<br />
terms of how close <strong>the</strong> two agents were to each o<strong>the</strong>r on average during a trial. To test<br />
<strong>the</strong> robustness of this solution we performed a comprehensive set of test trials. An<br />
overview of this solution‟s fitness is depicted in Figure 10-1. The average score is not<br />
significantly different from that of experimental setup 1 (cf. Chapter 9, pp. 148-157).<br />
179 | P a g e
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
0.9<br />
0.85<br />
0.8<br />
0.75<br />
0.7<br />
0.65<br />
100 200 300 400 500 600<br />
Figure 10-1. Graphical representation of fitness scores achieved at each possible combination of starting<br />
positions for agent „up‟ (x-axis) <strong>and</strong> agent „down‟ (y-axis). Note that <strong>the</strong> axes wrap around due to <strong>the</strong> 1-D<br />
circular shape of <strong>the</strong> virtual environment. Fitness scores range from 0.62 to 0.94 with an average of 0.84.<br />
How does <strong>the</strong> evolved solution manage to consistently solve <strong>the</strong> task under <strong>the</strong>se<br />
modified conditions? It is helpful to describe <strong>the</strong> strategy in terms of a representative<br />
trial (shown in Figure 10-2). Unfortunately, it turns out that <strong>the</strong> strategy of <strong>the</strong> agents is<br />
based on <strong>the</strong> close proximity of <strong>the</strong> two shadow objects. All <strong>the</strong> agents have to do is<br />
distinguish between one stimulation <strong>and</strong> two consecutive stimulations. This is a robust<br />
individual-based strategy to locate <strong>the</strong> o<strong>the</strong>r since: (i) passing <strong>the</strong> static object only<br />
causes one activation of <strong>the</strong> receptor, <strong>and</strong> (ii) passing <strong>the</strong> o<strong>the</strong>r agent with its attached<br />
shadow results in two activations. In o<strong>the</strong>r words, <strong>the</strong> evolutionary process has found<br />
ano<strong>the</strong>r solution that essentially relies on a factor that is external to <strong>the</strong> interaction<br />
process, namely <strong>the</strong> spatial relationship between <strong>the</strong> agents <strong>and</strong> <strong>the</strong>ir shadows.<br />
180 | P a g e
Figure 10-2. Illustration of <strong>the</strong> behavior of <strong>the</strong> agents during a representative trial starting from point<br />
(246, 436) with a score of 0.79. They first encounter <strong>the</strong>ir respective static objects, <strong>the</strong>n continue<br />
searching, <strong>and</strong> finally locate each o<strong>the</strong>r <strong>and</strong> establish perceptual crossing until <strong>the</strong> end of <strong>the</strong> run (16000<br />
time steps). Top: <strong>the</strong> position of <strong>the</strong> agents <strong>and</strong> objects over time. Middle: <strong>the</strong> status of <strong>the</strong> receptor field<br />
<strong>and</strong> <strong>the</strong> velocity of agent „up‟ over time. Bottom: <strong>the</strong> status of <strong>the</strong> receptor field <strong>and</strong> <strong>the</strong> velocity of agent<br />
„down‟ over time. Note that a change of receptor field status reaches <strong>the</strong> agent‟s controller only after a<br />
delay of 5 units of time (50 time steps).<br />
Ironically, this behavioral strategy is robust because <strong>the</strong> shadow, which was meant to<br />
introduce an essential ambiguity into <strong>the</strong> experiment, has been appropriated to<br />
disambiguate <strong>the</strong> target from <strong>the</strong> static object. Is this a strategy that would be used by<br />
<strong>the</strong> human participants of <strong>the</strong> original study? Participants were indeed told about <strong>the</strong><br />
experimental setup, including <strong>the</strong> three types of objects that <strong>the</strong>y could encounter, but<br />
“<strong>the</strong> precise relation of <strong>the</strong> mobile lure yoked to <strong>the</strong> avatar was not explained” (Auvray,<br />
181 | P a g e
et al. 2009, p. 38). Never<strong>the</strong>less, a large percentage of responses was preceded by a<br />
double stimulation (event E2, 32.3%, ibid. p. 40), indicating that <strong>the</strong> shadow-link might<br />
have played a role in <strong>the</strong> positive empirical results.<br />
10.2 Experimental setup 5: Maximally distant shadows<br />
What kind of strategies would be available if participants cannot take advantage of <strong>the</strong><br />
external agent-shadow relationship? Of course, we do not want to completely sever <strong>the</strong><br />
link between <strong>the</strong> movements of <strong>the</strong> agents <strong>and</strong> <strong>the</strong>ir shadow objects, since this is an<br />
essential aspect of <strong>the</strong> experimental setup (socially contingent vs. active but noncontingent<br />
interactions). Instead, we make <strong>the</strong> link between <strong>the</strong>m maximally distant<br />
(150 units) 34 . The rest of <strong>the</strong> setup remains <strong>the</strong> same as in <strong>the</strong> previous experiment. In<br />
this manner we want to test whe<strong>the</strong>r <strong>the</strong> evolutionary algorithm can come up with<br />
strategies that depend on <strong>the</strong> co-regulated activity of <strong>the</strong> agents alone. We evolved 6-<br />
node CTRNNs for several thous<strong>and</strong> generations <strong>and</strong> <strong>the</strong>n chose <strong>the</strong> fittest solutions to<br />
run test trials. The outcome of a typical trial is shown in Figure 10-3.<br />
34 Strictly speaking, in a 600 unit-wide circular world, being apart 300 units would be maximally distant.<br />
However, in this case <strong>the</strong>re is ano<strong>the</strong>r stable perceptual crossing situation, in which <strong>the</strong> agents can interact<br />
at a distance by stimulating each o<strong>the</strong>r with <strong>the</strong>ir shadow objects.<br />
182 | P a g e
Figure 10-3. Illustration of <strong>the</strong> behavior of <strong>the</strong> agents during a representative trial starting from point<br />
(314, 411) with a score of 0.56. See caption of Figure 10-2 <strong>and</strong> text for a more detailed description of <strong>the</strong><br />
graphs. Note <strong>the</strong> same patterns of receptor stimulation for agent „up‟ (middle graph) both when it interacts<br />
with its static object as well as when it first interacts with agent „down‟ during <strong>the</strong> middle of <strong>the</strong> trial.<br />
First, <strong>the</strong> agents explore <strong>the</strong>ir static objects for some time. Then <strong>the</strong>y proceed to explore<br />
<strong>the</strong> rest of <strong>the</strong> space, encounter each o<strong>the</strong>r <strong>and</strong> engage in some initial perceptual<br />
crossing. This mutual interaction breaks down after a while, <strong>and</strong> <strong>the</strong>y continue exploring<br />
until <strong>the</strong>y re-establish perceptual crossing at ano<strong>the</strong>r location. Such break-downs occur<br />
more often than in <strong>the</strong> original setup, because agents are more likely to miss each o<strong>the</strong>r<br />
with infinitely small object sizes. Indeed, <strong>the</strong> possibility of interaction break-down<br />
183 | P a g e
could be a first indication that <strong>the</strong> agents have to be much more active <strong>and</strong> responsive in<br />
<strong>the</strong>ir interaction in order to disambiguate <strong>the</strong> situation. They cannot make use of<br />
persistent <strong>and</strong> reliable external factors to assess <strong>the</strong> viability of <strong>the</strong>ir behavior, <strong>and</strong> thus<br />
<strong>the</strong>y are more open to commit errors <strong>and</strong> make mistaken responses. Due to this<br />
variability <strong>the</strong> behavior of <strong>the</strong> agents during <strong>the</strong> trial run thus looks much more <strong>life</strong>like<br />
than that of previous solutions. This modeling experiment leads us to <strong>the</strong> prediction that<br />
<strong>the</strong> performance of human participants under <strong>the</strong>se modified conditions would not be<br />
significantly different than from <strong>the</strong> original setup. In o<strong>the</strong>r words, while object size <strong>and</strong><br />
shadow link width might aid some individual-based strategies, <strong>the</strong>se factors are nei<strong>the</strong>r<br />
necessary nor essential for <strong>the</strong> collective success.<br />
However, <strong>the</strong>re still remains a problem in terms of this model. When <strong>the</strong> agents meet<br />
without receiving different stimulation beforeh<strong>and</strong>, <strong>the</strong>y engage on <strong>the</strong> basis of identical<br />
controllers (same structure <strong>and</strong> same internal state) such that <strong>the</strong>y will mirror <strong>the</strong>ir<br />
behavior perfectly. This produces <strong>the</strong> same sensory-motor correlation as if <strong>the</strong>y were<br />
oscillating around <strong>the</strong>ir static object. And since agents are more likely to encounter each<br />
o<strong>the</strong>r, evolution produces solutions which treat <strong>the</strong> occurrence of this sensory-motor<br />
pattern as always being due to <strong>the</strong> o<strong>the</strong>r ra<strong>the</strong>r than to <strong>the</strong> static object (a good choice,<br />
given <strong>the</strong> circumstances). However, occasionally this will result in both agents getting<br />
stuck oscillating around <strong>the</strong>ir static objects for <strong>the</strong> whole of a trial, giving rise to what<br />
looks like truly pathological behavior (for similar problems on a related task, cf. Rohde<br />
& Di Paolo, 2008).<br />
10.3 Experimental setup 6: Coordinated behavior<br />
How can we use ER to generate solutions that are better at distinguishing <strong>the</strong> o<strong>the</strong>r<br />
agent from <strong>the</strong> static object? As a first step, we remove <strong>the</strong> possibility of functionally<br />
identical CTRNNs encountering each o<strong>the</strong>r by simply activating <strong>the</strong> receptor field of a<br />
r<strong>and</strong>omly chosen agent at <strong>the</strong> start of <strong>the</strong> trial, thus ensuring a minimal difference in<br />
individual histories. Moreover, in Chapter 8 we showed that sensitivity to social<br />
contingency can emerge from <strong>the</strong> interaction process if agents are required to coordinate<br />
<strong>the</strong>ir behaviors in a way that forces <strong>the</strong>m to break <strong>the</strong> symmetry of <strong>the</strong>ir interactions.<br />
We <strong>the</strong>refore introduce some additional requirements into <strong>the</strong> fitness function. First, we<br />
184 | P a g e
also explicitly reward <strong>the</strong> agents for crossing each o<strong>the</strong>r, ra<strong>the</strong>r than just remaining in<br />
spatial proximity. Second, if <strong>the</strong>y engage in perceptual crossing (defined as at least two<br />
consecutive crosses, less than 10 units of time apart), we increase <strong>the</strong> reward<br />
proportionally to <strong>the</strong> distance traveled toge<strong>the</strong>r (<strong>the</strong>y are not rewarded for traveling<br />
alone). The evaluation function combines <strong>the</strong>se factors as follows:<br />
// If no perceptual crossing (PC), <strong>the</strong>n just use mean distance<br />
if (NumOfPercCrossings == 0)<br />
trialFitness = meanDistance;<br />
// If <strong>the</strong>re has been some PC, <strong>the</strong>n increase reward incrementally<br />
else if (NumOfPercCrossings < 20)<br />
trialFitness = meanDistance + NumOfPercCrossings;<br />
// If more PC, <strong>the</strong>n increase reward according to displacement<br />
else<br />
trialFitness = meanDistance + 20 + DistanceTraveled;<br />
We found that capping <strong>the</strong> influence of number of perceptual crossings at 20, <strong>and</strong> only<br />
<strong>the</strong>n taking <strong>the</strong> distance traveled toge<strong>the</strong>r into account increased <strong>the</strong> evolvability of <strong>the</strong><br />
solutions. Note that since <strong>the</strong> agents are structural clones, <strong>the</strong> traveling toge<strong>the</strong>r is a nontrivial<br />
activity since it requires <strong>the</strong>m to break <strong>the</strong> symmetry of <strong>the</strong>ir interactions. Note<br />
also that because it is impossible to coordinate this maneuver with a static object, we<br />
have fur<strong>the</strong>r emphasized <strong>the</strong> possibility of distinguishing <strong>the</strong> o<strong>the</strong>r in terms of its<br />
responsiveness (i.e. response to attempts at breaking behavioral symmetry).<br />
We evolved 8-node CTRNNs with this modified fitness function. The agents are able to<br />
coordinate <strong>the</strong>ir behavior so as to travel toge<strong>the</strong>r while interacting (cf. Figure 10-4).<br />
While engaging in perceptual crossing, <strong>the</strong> agents eventually start to drift toge<strong>the</strong>r<br />
horizontally. In o<strong>the</strong>r words, even though <strong>the</strong> agents are structurally identical, have<br />
minimally different histories (internal states), <strong>and</strong> are not affected by noise during <strong>the</strong><br />
trial, <strong>the</strong>y are never<strong>the</strong>less able to regulate <strong>the</strong> interaction such that <strong>the</strong> symmetry of<br />
<strong>the</strong>ir individual behaviors is broken. In fact, when one of <strong>the</strong> agents encounters its static<br />
object during this coordination process, <strong>the</strong> agents are able to re-negotiate <strong>the</strong> direction<br />
of drift <strong>and</strong> return <strong>the</strong> o<strong>the</strong>r way, much like what was found in <strong>the</strong> pioneering work by<br />
Quinn, et al. (2003). It is important to emphasize that <strong>the</strong> ability of <strong>the</strong> agents to<br />
negotiate <strong>the</strong> direction of travel in ei<strong>the</strong>r direction amounts to an interactively mediated<br />
185 | P a g e
expansion of <strong>the</strong>ir individual behavioral domains. As solitary agents <strong>the</strong>y can only<br />
traverse <strong>the</strong> environment in one direction.<br />
Figure 10-4. A representative trial run starting from (90, 390) <strong>and</strong> scoring 122 points. Initially, one agent<br />
gets stuck on its static object, <strong>and</strong> <strong>the</strong> o<strong>the</strong>r on <strong>the</strong> o<strong>the</strong>r‟s shadow. Then <strong>the</strong> shadow interaction breaks<br />
down, <strong>the</strong> agents meet <strong>and</strong> start moving toge<strong>the</strong>r until <strong>the</strong> end of <strong>the</strong> trial. Note that <strong>the</strong>y jointly bounce<br />
back from <strong>the</strong> static object located at position 448.<br />
Never<strong>the</strong>less, it is still <strong>the</strong> case that this strategy is not as robust as <strong>the</strong> solutions which<br />
we have excluded by modifying <strong>the</strong> experimental design. For instance, if both agents<br />
happen to encounter <strong>the</strong>ir static objects at <strong>the</strong> start of <strong>the</strong> trial, it is possible that <strong>the</strong>y<br />
simply continue to oscillate around <strong>the</strong>m until <strong>the</strong> trial is terminated (typically obtaining<br />
a fitness score between 0.25 <strong>and</strong> 0). A related problem occurs if <strong>the</strong> two agents meet<br />
while having <strong>the</strong> same internal state (e.g. due to same history of interactions), perform<br />
exactly <strong>the</strong> same behavior, <strong>and</strong> <strong>the</strong>n move apart. In both cases <strong>the</strong> problem is that <strong>the</strong>re<br />
is no principled way for <strong>the</strong> agents to tell apart an interaction with a static object <strong>and</strong><br />
interaction with ano<strong>the</strong>r agent with identical internal state. Both give rise to <strong>the</strong> same<br />
basic pattern of stimulation. Surprisingly, this problem occurs even if <strong>the</strong> possibility of<br />
identity of internal state has been removed in principle. The initial binary difference in<br />
stimulus, which we introduced to give <strong>the</strong> agents at least a minimally different history<br />
of interactions, is ineffective because it was not exploited by any long-term activity of<br />
<strong>the</strong> CTRNN controllers. The fast time constants of <strong>the</strong> evolved CTRNN entail that this<br />
difference is not carried for long as a difference in internal state.<br />
At first sight <strong>the</strong> failure to break away from <strong>the</strong> static object appears to be evidence that<br />
<strong>the</strong> agents are not sensitive to <strong>the</strong> social contingency of <strong>the</strong>ir interaction; o<strong>the</strong>rwise <strong>the</strong>y<br />
would presumably notice <strong>the</strong> lack of responsiveness of <strong>the</strong> object <strong>and</strong> eventually move<br />
away. But this way of looking at <strong>the</strong> problem essentially dem<strong>and</strong>s an individual<br />
186 | P a g e
esponse alone, i.e. detecting a lack of social contingency when <strong>the</strong>re is none to be<br />
detected. In contrast, perhaps <strong>the</strong> existence of this pathological behavior is an indication<br />
of <strong>the</strong> truly social nature of <strong>the</strong> evolved solution? Indeed, if only one agent becomes<br />
trapped it will eventually be freed by <strong>the</strong> o<strong>the</strong>r agent, which entrains it in an interaction<br />
process such that <strong>the</strong>y move away toge<strong>the</strong>r. In o<strong>the</strong>r words, this solution depends on <strong>the</strong><br />
responsiveness of <strong>the</strong> o<strong>the</strong>r to such an extent that if <strong>the</strong> presumed „o<strong>the</strong>r‟ is not<br />
responsive to <strong>the</strong> interaction <strong>the</strong> strategy fails 35 .<br />
This individual failure can also be related to an important insight we can learn from <strong>the</strong><br />
experimental design process, namely just how difficult it is to evolve a behavioral<br />
strategy that is primarily (or exclusively) based on mutually contingent interaction. One<br />
important aspect of this difficulty is surely that detecting an object‟s responsiveness as<br />
such is a much more dem<strong>and</strong>ing task than detecting environmental cues (e.g. difference<br />
in stimulus duration, difference in number of contacts, difference in noise, etc.). This is<br />
because <strong>the</strong> latter phenomena can become manifest in conditions that are largely<br />
independent of an agent‟s behavior (e.g. as long as an agent moves, passing <strong>the</strong> o<strong>the</strong>r<br />
„agent-shadow‟ will cause two stimulations, while passing <strong>the</strong> static object will cause<br />
one stimulation). Moreover, basing a behavioral strategy on <strong>the</strong> o<strong>the</strong>r introduces an<br />
inherent risk to <strong>the</strong> situation. The o<strong>the</strong>r‟s behavior can be influenced by your own<br />
actions, but it evades your direct control in principle. This is especially problematic<br />
when <strong>the</strong> presumed „o<strong>the</strong>r‟ does not react to you in a suitable manner, but your<br />
individual ability depends on <strong>the</strong> o<strong>the</strong>r‟s behavior. This is <strong>the</strong> case for <strong>the</strong> „pathological‟<br />
behavior of <strong>the</strong> simulated agents.<br />
The increasing complexity of <strong>the</strong> task is also indicated by <strong>the</strong> practical need that we<br />
have had to increase <strong>the</strong> number of nodes in <strong>the</strong> evolving CTRNN controller to support<br />
35 The fact that <strong>the</strong> agents in this modeling experiment are unable to distinguish between <strong>the</strong> static object<br />
<strong>and</strong> <strong>the</strong> o<strong>the</strong>r individually, but can do so when <strong>the</strong> o<strong>the</strong>r is present, deserves fur<strong>the</strong>r study, especially in<br />
relation to empirical findings in <strong>the</strong> cognitive sciences. For example, studies of rehabilitation after brain<br />
damage have shown that patients often find (i) sensory-motor tasks impossible to achieve individually in<br />
an abstract context, (ii) have difficulty with <strong>the</strong>m in a pragmatic context, <strong>and</strong> (iii) can function almost<br />
normally in socially situated circumstances (cf. Gallagher & Marcel 1999).<br />
187 | P a g e
etter evolvability (three nodes for <strong>the</strong> experiments in Chapter 8, four for Chapter 9, six<br />
for Section 10.1 <strong>and</strong> 10.2, eight for this section). In order to determine whe<strong>the</strong>r it is<br />
possible to syn<strong>the</strong>size a solution that performs robustly even when confronted by <strong>the</strong><br />
static object in isolation, we <strong>the</strong>refore repeated <strong>the</strong> evolutionary process but now with<br />
10-node CTRNNs. Presumably, giving <strong>the</strong> individual agents more complex controllers<br />
will make it more likely that <strong>the</strong>y are able to resolve <strong>the</strong> situation appropriately in terms<br />
of individual-based strategies as well. For this setup we also removed <strong>the</strong> r<strong>and</strong>om initial<br />
stimulus, since this attempt to influence <strong>the</strong> internal state of <strong>the</strong> agents made no<br />
difference to <strong>the</strong> previous strategy. We comprehensively tested <strong>the</strong> best solution after a<br />
few thous<strong>and</strong> generations of optimization. The results are shown in Figure 10-5.<br />
Figure 10-5. Graphical representation of fitness scores at each possible combination of starting positions<br />
for agent „up‟ (x-axis) <strong>and</strong> agent „down‟ (y-axis). Note that <strong>the</strong> axes wrap around due to <strong>the</strong> 1-D circular<br />
shape of <strong>the</strong> environment. Left: Trials of 1600 units of time. Fitness scores range from 3.53 to 316 with an<br />
average of 128. Right: Trials of 3200 units of time. Fitness scores range from 121 to 316 with an average<br />
of 129.<br />
Two things are immediately evident from <strong>the</strong>se test results. First, <strong>the</strong> evolved strategy is<br />
very robust in coping with different starting positions, <strong>and</strong> second, even <strong>the</strong> worst trials<br />
are not complete failures. Moreover, <strong>the</strong> regions of low fitness visible in <strong>the</strong> fitness map<br />
shown on <strong>the</strong> left of Figure 10-5 are due to unfavorable starting conditions, which<br />
require more time to resolve successfully. A comprehensive test with trials that are<br />
twice as long does not show any problematic regions (cf. <strong>the</strong> fitness map on <strong>the</strong> right of<br />
Figure 10-5). This demonstrates that <strong>the</strong> agents are able to disentangle <strong>the</strong>mselves from<br />
<strong>the</strong> static object even without <strong>the</strong> aid of <strong>the</strong> o<strong>the</strong>r agent (o<strong>the</strong>rwise <strong>the</strong>re would be<br />
188 | P a g e
egions of low fitness for those initial conditions where both agents first encounter <strong>the</strong>ir<br />
respective static objects). A representative trial of this situation is shown in Figure 10-6.<br />
Figure 10-6. A representative trial run starting from (238, 289) <strong>and</strong> scoring 122. Initially, <strong>the</strong> agents get<br />
stuck on <strong>the</strong>ir static objects. Then this interaction breaks down without outside interference, <strong>the</strong> agents<br />
eventually meet <strong>and</strong> start moving away toge<strong>the</strong>r until <strong>the</strong> end of <strong>the</strong> trial. Note that agent „down‟ gets<br />
perturbed by <strong>the</strong> o<strong>the</strong>r‟s shadow on <strong>the</strong> way to its static object <strong>and</strong> <strong>the</strong>n undergoes an additional iteration<br />
of interactions.<br />
We have thus managed to evolve a strategy that can cope with <strong>the</strong> high ambiguity of<br />
this modified experimental setup successfully. The agents never get completely trapped<br />
by <strong>the</strong>ir static objects or <strong>the</strong> shadow object of <strong>the</strong>ir partner. When an agent meets its<br />
static object it engages in bursts of interaction that slowly decrease in frequency until<br />
<strong>the</strong> agent eventually breaks free <strong>and</strong> continues on its way. To be sure, if <strong>the</strong> agents meet<br />
189 | P a g e
each o<strong>the</strong>r with identical internal states <strong>the</strong>n <strong>the</strong>y will disengage from <strong>the</strong> interaction in<br />
<strong>the</strong> same manner, because this situation is in principle identical to interacting with a<br />
static object. But if <strong>the</strong>ir internal states are at variance due to differing histories of<br />
interaction, <strong>the</strong>n <strong>the</strong>y are capable of breaking <strong>the</strong> symmetry of <strong>the</strong>ir behavior <strong>and</strong><br />
engage in perceptual crossing while jointly traveling around <strong>the</strong> environment. Indeed,<br />
<strong>the</strong>ir internal states are different most of <strong>the</strong> time, especially because <strong>the</strong> occasional <strong>and</strong><br />
uncorrelated perturbations due to <strong>the</strong> o<strong>the</strong>r agent‟s shadow object function as a source<br />
of noise. In Figure 10-6, for example, we can see that this kind of perturbation causes<br />
agent „down‟ to undergo an additional burst of interactions with its static object. Thus, if<br />
<strong>the</strong> two agents meet with different states we know that <strong>the</strong>ir behavior will not be<br />
identical <strong>and</strong> that <strong>the</strong>y will create a qualitatively different pattern of interaction. Toward<br />
<strong>the</strong> end of <strong>the</strong> trial shown in Figure 10-6 we can see this: <strong>the</strong> mutual interaction does not<br />
exhibit <strong>the</strong> same slow decrease in frequency of bursts of interactions, but is more<br />
irregular.<br />
Since <strong>the</strong>se agents can successfully solve <strong>the</strong> task, we can hypo<strong>the</strong>size that <strong>the</strong> outcome<br />
of <strong>the</strong> original psychological study will not be significantly altered when making objects<br />
infinitely small <strong>and</strong> displacing <strong>the</strong> shadow object by 150 units. The evolutionary<br />
robotics methodology has thus allowed us to fine-tune <strong>the</strong> essential elements of <strong>the</strong><br />
experimental design. Indeed, we can venture a fur<strong>the</strong>r hypo<strong>the</strong>sis that part of <strong>the</strong> reason<br />
why <strong>the</strong> original study found less response to <strong>the</strong> static object, when compared to <strong>the</strong><br />
two mobile objects, was that entrainment with this object was often broken by <strong>the</strong><br />
actions of <strong>the</strong> o<strong>the</strong>r participant (ei<strong>the</strong>r because of its shadow or its receptor).<br />
10.4 Summary<br />
The strategy of <strong>the</strong> agents in <strong>the</strong> final experimental setup is composed of both individual<br />
<strong>and</strong> interactive factors: <strong>the</strong> agents are individually capable of avoiding static objects<br />
(<strong>and</strong> shadow objects), <strong>and</strong> <strong>the</strong>y can also perturb each o<strong>the</strong>r through <strong>the</strong>ir interaction so<br />
as to sustain <strong>the</strong>ir mutual interaction without depending on factors that are external to<br />
that interaction. But have we finally managed to syn<strong>the</strong>size a model of social interaction<br />
ra<strong>the</strong>r than just multi-agent interaction as proposed in <strong>the</strong> introduction to this chapter?<br />
190 | P a g e
We know that we are dealing with models of social interaction in <strong>the</strong> case of those<br />
experiments where <strong>the</strong> agents are required to travel toge<strong>the</strong>r, because such coordinated<br />
movement is a type of activity that cannot be achieved by <strong>the</strong> regulation of any<br />
individual alone. In particular, <strong>the</strong> agents must break <strong>the</strong> symmetry of <strong>the</strong>ir behaviors;<br />
coordinate a direction of movement, <strong>and</strong> <strong>the</strong>n move toge<strong>the</strong>r in that direction while<br />
continuing to interact. Moreover, this coordination is flexible in that <strong>the</strong> direction of<br />
movement can be re-negotiated if necessary, for example when faced by a static object.<br />
It is also worth emphasizing that, because we eliminated external clues about <strong>the</strong><br />
location of <strong>the</strong> o<strong>the</strong>r agent (all objects give rise to <strong>the</strong> same minimal, solitary stimulus<br />
upon contact), <strong>the</strong> agents have to disambiguate <strong>the</strong> experimental situation by means of<br />
<strong>the</strong> properties of <strong>the</strong> interaction process itself. A successful strategy requires, in<br />
anthropomorphic terms, that an agent proposes a particular direction of travel, <strong>and</strong> that<br />
this offer is accepted by <strong>the</strong> o<strong>the</strong>r agent. O<strong>the</strong>rwise <strong>the</strong>re is an opening for fur<strong>the</strong>r<br />
negotiation, or <strong>the</strong> interaction simply fails (as in <strong>the</strong> case of interacting with <strong>the</strong> static<br />
object). Here we thus have all <strong>the</strong> necessary ingredients to speak of a model of social<br />
interaction.<br />
On <strong>the</strong> basis of <strong>the</strong>se modeling results it is possible to hypo<strong>the</strong>size that if <strong>the</strong><br />
psychological experiment is repeated with this modified setup (e.g. infinitely small<br />
objects <strong>and</strong> maximally distant shadow objects) <strong>and</strong> modified task (e.g. primarily<br />
pragmatic ra<strong>the</strong>r than epistemic), we will find, in contrast to <strong>the</strong> original study, that<br />
<strong>the</strong>re is a statistically significantly higher probability of clicking in response to<br />
encounters with <strong>the</strong> o<strong>the</strong>r participant. In fact, it is easy to imagine that if participants<br />
only clicked when <strong>the</strong>y have managed to establish sideways perceptual crossing 36 , <strong>the</strong>n<br />
<strong>the</strong> clicking responses could easily be 100% correct (since such co-regulated behavior is<br />
simply impossible with <strong>the</strong> shadow or <strong>the</strong> static object). This would be an example<br />
where <strong>the</strong> presence of social interaction significantly improves <strong>the</strong> individual abilities<br />
when compared to multi-agent interaction alone.<br />
36 That such coordinated behavior is in fact possible for human participants, at least under original<br />
experimental conditions, has already been demonstrated by some exploratory studies. The subjects were<br />
asked to engage in sideways movement toge<strong>the</strong>r without being told any direction in advance (Di Paolo,<br />
personal communication).<br />
191 | P a g e
Finally, in addition to this increase in behavioral performance, we can expect to find<br />
some qualitative differences in <strong>the</strong> participants‟ experience as well. Indeed, whereas in<br />
<strong>the</strong> original psychological experiment <strong>the</strong> socially contingent <strong>and</strong> non-contingent<br />
situations entailed no difference in meaning for <strong>the</strong> participants (i.e. <strong>the</strong>re was no<br />
statistically significant difference in clicking response), we can hypo<strong>the</strong>size that<br />
successful completion of this modified task is associated with a specifically social<br />
phenomenology. As such, we might have found minimal experimental conditions for<br />
participatory sense-making, which would make <strong>the</strong> modified psychological study an<br />
especially appropriate target for phenomenological investigation, perhaps by means of<br />
explicitation interviews (e.g. Petitmengin 2006). This would be a novel opportunity to<br />
determine <strong>the</strong> structural <strong>and</strong> qualitative differences between object- <strong>and</strong> o<strong>the</strong>rperception<br />
under minimalist <strong>and</strong> controllable conditions. We will return to <strong>the</strong><br />
phenomenology of intersubjectivity in Chapter 12.<br />
10.5 Discussion<br />
The simulation models that were presented in Chapters 8 to 10 toge<strong>the</strong>r form a series of<br />
thought experiments that were inspired by <strong>the</strong> enactive approach to cognition, especially<br />
by <strong>the</strong> <strong>the</strong>oretical framework that was developed in Chapter 4. However, in order count<br />
as strong support for <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis proposed by <strong>the</strong> enactive paradigm,<br />
a final problem must be addressed. There is still a lingering worry that <strong>the</strong> results of<br />
<strong>the</strong>se modeling experiments can simply be appropriated by a more general, embodied<br />
<strong>and</strong> dynamical approach to cognitive science (cf. Section 2.2). In o<strong>the</strong>r words, if <strong>the</strong><br />
model agents are not autonomous in <strong>the</strong> enactive sense of <strong>the</strong> term, <strong>the</strong>n what makes<br />
<strong>the</strong>se results supportive of that paradigm ra<strong>the</strong>r than of a broadly conceived embodiedembedded<br />
cognitive science? To be sure, <strong>the</strong> models can be insightful for a variety of<br />
different approaches. But do <strong>the</strong>y also provide insights that are specifically „enactive‟<br />
such that <strong>the</strong>y could not also have simply been accounted for by, for instance, a generic<br />
dynamical approach to social cognition?<br />
In response to <strong>the</strong>se worries it is important first to emphasize <strong>the</strong> widespread impact<br />
which <strong>the</strong> enactive paradigm has had ever since <strong>the</strong> publication of The Embodied Mind<br />
by Varela et al. in 1991. In o<strong>the</strong>r words, many of <strong>the</strong> core <strong>the</strong>mes of contemporary<br />
192 | P a g e
embodied-embedded cognitive science, including <strong>the</strong> role of embodiment, situatedness,<br />
dynamics, emergence <strong>and</strong> active perception, have been strongly influenced by <strong>the</strong> ideas<br />
of that book. Moreover, certainly <strong>the</strong> evolutionary robotics methodology that we have<br />
used to syn<strong>the</strong>size <strong>the</strong> models is also a popular tool for o<strong>the</strong>r embodied-embedded<br />
approaches. But <strong>the</strong>y are rarely aware of <strong>the</strong> fact that <strong>the</strong> methodology‟s dynamical<br />
perspective on behavior <strong>and</strong> cognition follows directly from an autopoietic perspective<br />
on <strong>life</strong> when two key abstractions are made (Beer 1995a): (i) we focus our investigation<br />
on an agent‟s behavioral dynamics alone, <strong>and</strong> (ii) we abstract <strong>the</strong> set of destructive<br />
perturbations that this agent can undergo as a viability constraint. Since <strong>the</strong>se two<br />
abstractions basically make or break <strong>the</strong> relevance of <strong>the</strong> dynamical approach for <strong>the</strong><br />
enactive paradigm, it is worth spelling out Beer‟s argument in more detail.<br />
Beer (2004) begins with <strong>the</strong> observation that a natural agent‟s normal behavior takes<br />
place within a highly structured subset of its total domain of interaction. This makes it<br />
possible to capture <strong>the</strong> behavioral dynamics while ignoring o<strong>the</strong>r structural details<br />
which may not be directly relevant. Moreover, since it is only meaningful to study an<br />
agent‟s behavior while it is living, we can largely take an agent‟s ongoing metabolic<br />
autonomy for granted in our models. This abstraction fits nicely with Bar<strong>and</strong>iaran <strong>and</strong><br />
Moreno‟s (2006) claim that <strong>the</strong> hierarchical decoupling of <strong>the</strong> nervous system entails<br />
that <strong>the</strong> dynamic organization of cognition is metabolically underdetermined. Finally,<br />
<strong>the</strong> possibility of undergoing a lethal interaction is represented as a viability constraint<br />
on <strong>the</strong> agent‟s behavior such that if any actions are ever taken that carry <strong>the</strong> agent into<br />
this terminal state, no fur<strong>the</strong>r behavior is possible.<br />
It follows from <strong>the</strong>se considerations that research with artificial embodied-embedded<br />
systems has <strong>the</strong> potential to develop a mutually informing relationship with some of <strong>the</strong><br />
<strong>the</strong>oretical foundations of enactive cognitive science. However, <strong>the</strong> insights generated<br />
by this research are equally informative for o<strong>the</strong>r approaches in cognitive science, even<br />
those whose interest in embodied, embedded <strong>and</strong> dynamical phenomena is merely in<br />
terms of an extension to functionalism (e.g. Wheeler 2005; Clark 1997). This ongoing<br />
appropriation should come as a warning to <strong>the</strong> enactive paradigm, especially because<br />
<strong>the</strong>re is a growing consensus that functionalism is fundamentally incompatible with <strong>the</strong><br />
enactive notion of autonomy (cf. Di Paolo 2009; Thompson & Stapleton 2009). Indeed,<br />
193 | P a g e
due to its functional level of abstraction much of embodied-embedded research cannot<br />
aid our underst<strong>and</strong>ing of how natural cognition arises through <strong>the</strong> precarious, selfconstituted<br />
activity of biological systems (cf. Section 3.4). As such, <strong>the</strong> embodiedembedded<br />
approach is unable to systematically address <strong>the</strong> kind of criticisms which<br />
have recently been leveled against it by <strong>Dr</strong>eyfus <strong>and</strong> o<strong>the</strong>rs (cf. Section 2.1). Have <strong>the</strong><br />
models presented in this <strong>the</strong>sis managed to avoid this dilemma?<br />
Arguably, this is at least partly <strong>the</strong> case. There is one clear sense in which this modeling<br />
work is specifically situated within <strong>the</strong> enactive paradigm alone, namely <strong>the</strong> constant<br />
focus on constitutive autonomy. To be sure, <strong>the</strong> simulated agents <strong>the</strong>mselves were not<br />
autonomous in this sense, but this was <strong>the</strong> result of a pragmatic choice to focus on <strong>the</strong><br />
dynamics of <strong>the</strong> interaction process instead. Given <strong>the</strong> current state of <strong>the</strong> art, <strong>the</strong> design<br />
of a system, which when run gives rise to autonomous entities which happen to interact<br />
with <strong>the</strong>ir environment in such a manner so as to engage in mutual interactions that give<br />
rise to an autonomous interaction process, would have first required us to address <strong>the</strong><br />
unresolved problem of „second-order emergence‟ (<strong>Froese</strong> & Ziemke 2009). While this<br />
is a worthwhile goal in itself, it would have unnecessarily distracted from <strong>the</strong> actual<br />
target of this investigation, namely <strong>the</strong> constitutive role of <strong>the</strong> autonomous interaction<br />
process for bootstrapping individual behavior. This view on <strong>the</strong> interaction process in<br />
itself, as an autonomous system, is thoroughly enactive. It has opened up <strong>the</strong> possibility<br />
of a systematic research program that can bridge <strong>the</strong> „cognitive gap‟ of <strong>the</strong> <strong>life</strong>-<strong>mind</strong><br />
<strong>continuity</strong> <strong>the</strong>sis (<strong>Froese</strong> & Di Paolo 2009; De Jagher & <strong>Froese</strong> 2009): <strong>the</strong> enactive<br />
paradigm can potentially integrate <strong>the</strong> distance between <strong>the</strong> <strong>life</strong> of a single cell <strong>and</strong> <strong>the</strong><br />
<strong>mind</strong> of a human being into one complex meshwork of autonomous systems.<br />
In addition, by means of <strong>the</strong> enactive detour through a focus on <strong>the</strong> autonomy of <strong>the</strong><br />
interaction process, <strong>the</strong>se models might actually represent an important step toward<br />
using evolutionary robotics not as a method to optimize predefined „agents‟, but ra<strong>the</strong>r<br />
as a way to generate <strong>the</strong> conditions for <strong>the</strong> self-organization of such systems (cf. <strong>Froese</strong><br />
& Di Paolo 2008b). As such, <strong>the</strong> insights generated by <strong>the</strong>se models will not be easily<br />
appropriated by some form of dynamical functionalism. In fact, it is not even clear how<br />
<strong>the</strong> organization of <strong>the</strong> self-sustaining dynamics of <strong>the</strong> interaction process can be<br />
captured in quantitative terms, or if that is going to be possible at all (cf. Stewart 2000).<br />
194 | P a g e
It remains to be seen to what extent we can go beyond psycho-physical tests <strong>and</strong><br />
qualitative descriptions when it comes to <strong>the</strong> study of an autonomous organization.<br />
Finally, <strong>the</strong>re is ano<strong>the</strong>r crucial difference between <strong>the</strong> enactive paradigm <strong>and</strong> generic<br />
embodied-embedded cognitive science apart from <strong>the</strong> concept of autonomy, <strong>and</strong> that is<br />
<strong>the</strong> role played by our very own experience. In fact, lived experience has arguably been<br />
<strong>the</strong> central <strong>the</strong>me of its research program, though many proponents of „enaction‟ have<br />
preferred to focus on <strong>the</strong> „embodied-embedded‟ insights of The Embodied Mind, ra<strong>the</strong>r<br />
than on developing <strong>the</strong> implications of its provocative subtitle Cognitive Science <strong>and</strong><br />
Human Experience. None<strong>the</strong>less, <strong>the</strong> inception of <strong>the</strong> enactive paradigm started on <strong>the</strong><br />
foundation of human experience, <strong>and</strong> its recent incorporation of <strong>the</strong> autopoietic tradition<br />
was similarly motivated by existential concerns (Weber & Varela 2002): How can we<br />
account for <strong>the</strong> fact that we care, that we are concerned beings with a point of view that<br />
enables a world to show up in a meaningful manner? We must have recourse to our own<br />
experience in order to verify whe<strong>the</strong>r we in fact are such beings <strong>and</strong>, consequently,<br />
whe<strong>the</strong>r <strong>the</strong>se questions require an answer or not. It is from this experiential basis that<br />
<strong>the</strong> <strong>the</strong>oretical framework of <strong>the</strong> enactive paradigm ultimately derives, <strong>and</strong> it is to this<br />
basis that our investigations must ultimately return.<br />
At this point we can submit a more general criticism to embodied-embedded cognitive<br />
science. While <strong>the</strong> field has been happy to appeal to <strong>the</strong> existential phenomenology of<br />
Heidegger, Merleau-Ponty, <strong>Dr</strong>eyfus <strong>and</strong> o<strong>the</strong>rs in order to support <strong>the</strong> use of an<br />
embodied, embedded <strong>and</strong> dynamical framework, this effort has largely remained a<br />
scholarly exercise ra<strong>the</strong>r than an experiential inquiry. To be sure, so far this approach<br />
has been a productive one, but signs of stagnation are already appearing. We ourselves<br />
need to start diving into <strong>the</strong>se forgotten realms of experience if we want to continue to<br />
push cognitive science in new directions. More yet: we need to make sure that <strong>the</strong> kind<br />
of science we derive from our insights lets us return to lived experience with newfound<br />
underst<strong>and</strong>ing. This is a tall order indeed, with profound implications, <strong>and</strong> <strong>the</strong> methods<br />
employed in this <strong>the</strong>sis have hardly done justice to <strong>the</strong> task. But at least <strong>the</strong> models have<br />
in <strong>the</strong> end led us to propose a testable hypo<strong>the</strong>sis about <strong>the</strong> dynamical conditions that<br />
must hold in order for an experience to be lived as qualitatively social. Of course,<br />
fur<strong>the</strong>r work remains to be done before such a test is possible with sufficient scientific<br />
195 | P a g e
igor, especially because <strong>the</strong>re is still a need to devise adequate means of specifying <strong>the</strong><br />
structure <strong>and</strong> content of experience. While a more detailed response to this challenge is<br />
beyond <strong>the</strong> scope of this <strong>the</strong>sis, in <strong>the</strong> following chapters we will conduct some initial<br />
explorations in <strong>the</strong> phenomenology of social <strong>life</strong>.<br />
196 | P a g e
11 Beyond methodological physicalism<br />
In <strong>the</strong> previous chapters we have used evolutionary robotics <strong>and</strong> a dynamical systems<br />
approach to gain a better underst<strong>and</strong>ing of <strong>the</strong> processes by which individual <strong>and</strong> social<br />
domains of phenomena can be constitutively interrelated. The modeling experiments<br />
have demonstrated that it is possible to question <strong>the</strong> widespread assumption of<br />
methodological individualism, <strong>and</strong> to do so without <strong>the</strong>reby descending into some<br />
mysterious notion of social <strong>life</strong>. On <strong>the</strong> contrary, <strong>the</strong> systemic approach has enabled us<br />
to capture something of <strong>the</strong> specificity of social phenomena, namely <strong>the</strong> organization of<br />
<strong>the</strong>ir relational structure, <strong>and</strong> <strong>the</strong>reby given novel credibility to <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong><br />
<strong>the</strong>sis. We have used <strong>the</strong> mutually supportive concepts of <strong>the</strong> enactive approach, i.e. <strong>the</strong><br />
notions of autonomy, sense-making, embodiment, <strong>and</strong> emergence, to outline <strong>the</strong><br />
beginnings of a research program into <strong>the</strong> origins of cognition on <strong>the</strong> basis of <strong>the</strong>se<br />
foundational biological principles.<br />
It appears that we have accomplished what we set out to achieve, <strong>and</strong> so it might be<br />
expected that <strong>the</strong> <strong>the</strong>sis ends here. But if we stopped our investigation here we would<br />
neglect an essential component of enactive cognitive science, namely <strong>the</strong> investigation<br />
of <strong>the</strong> subjective dimension of our bodily existence. The systematic integration of <strong>the</strong><br />
living body (systems biology) <strong>and</strong> <strong>the</strong> lived body (phenomenological philosophy)<br />
around a normatively laden conception of <strong>life</strong> is one of <strong>the</strong> great achievements of <strong>the</strong><br />
enactive approach. However, apart from indicating some structural constraints, so far<br />
we have said nothing about what it is like to be in a social situation. And it is precisely<br />
by addressing this first-person aspect of subjectivity that we move beyond <strong>the</strong> purely<br />
systemic (e.g. Luhmann 1984; Maturana & Varela 1987) <strong>and</strong> dialectical materialist (e.g.<br />
Vygotsky 1934) approaches to <strong>the</strong> social. Both of <strong>the</strong>se have much in common with <strong>the</strong><br />
enactive approach, though <strong>the</strong>y are limited by an impoverished conceptualization of <strong>the</strong><br />
target phenomenon.<br />
Of course, <strong>the</strong> task of explicating <strong>the</strong> characteristic nature of experiential phenomena in<br />
a manner that is scientifically respectable is beset by many difficulties, <strong>and</strong> an important<br />
part of future research will be to develop sound first- <strong>and</strong> second-person methodologies<br />
(e.g. Depraz, et al. 2003; Petitmengin 2006). It is beyond <strong>the</strong> scope of this <strong>the</strong>sis to enter<br />
197 | P a g e
into this complex debate (cf. Varela & Shear 1999), except to point out that even in this<br />
endeavor minimalist technological interfaces could feature as a crucial tool (cf. <strong>Froese</strong><br />
& Spiers 2007; Auvray, et al. 2007). Instead, we will have recourse to insights from <strong>the</strong><br />
phenomenological tradition, an important philosophical movement that was founded by<br />
Husserl in <strong>the</strong> beginning of <strong>the</strong> 20 th century <strong>and</strong> was continued by Heidegger, Merleau-<br />
Ponty, Scheler, Sartre <strong>and</strong> o<strong>the</strong>rs. Phenomenology has recently been brought into closer<br />
connection with cognitive science (cf. Gallagher & Zahavi 2008; Zahavi 2005; Roy, et<br />
al. 1999; Gallagher 1997), <strong>and</strong> has been an integral part to <strong>the</strong> enactive approach from<br />
<strong>the</strong> start (e.g. Varela, et al. 1991; Varela 1996b). Here we are going to appeal to central<br />
insights that have more or less withstood <strong>the</strong> test of time. Never<strong>the</strong>less, it is essential<br />
that this is not just a scholarly exercise, so <strong>the</strong> validity of <strong>the</strong>se insights will be verified<br />
as best as possible by illustrating <strong>the</strong>m with concrete examples from everyday<br />
experience.<br />
So what is it like to live through a social phenomenon? This question, if directed to our<br />
everyday selves, throws us right into <strong>the</strong> middle of a complex mixture of experience,<br />
preconception, <strong>and</strong> interpretation which is hard to disentangle. As already alluded to in<br />
Chapter 2, just like our assumptions determine <strong>the</strong> way we make sense of <strong>the</strong>ories <strong>and</strong><br />
empirical data, so <strong>the</strong>y shape <strong>the</strong> sense-making of our experience. We have already<br />
addressed <strong>the</strong> problem of methodological individualism, but <strong>the</strong>re is ano<strong>the</strong>r widespread<br />
assumption that is preventing a proper appreciation of <strong>the</strong> social. To put it boldly: most<br />
research in social cognition is trapped within an idealized world of physical forces. This<br />
is because much of mainstream science, with <strong>the</strong> exception of some progressive physics<br />
(cf. Bitbol 2002), is governed by a degenerate form of Cartesian metaphysics. To be<br />
sure, it differs somewhat from <strong>the</strong> substance dualism of <strong>the</strong> 17 th century, which divided<br />
all phenomena into <strong>the</strong> physical (res extensa) <strong>and</strong> <strong>the</strong> mental (res cogitans), by<br />
collapsing both domains into <strong>the</strong> physical 37 . However, as a remainder of this original<br />
abstraction, materialism is no different from idealism: Both give ontological primacy to<br />
<strong>the</strong> abstract organization of knowledge (ideas) over <strong>the</strong> concrete manifestation of <strong>the</strong><br />
37 The shift from a dualistic to a degenerate monist metaphysics has not been limited to <strong>the</strong> scientific<br />
community alone, as attested to by <strong>the</strong> common reference to <strong>the</strong> brain when actually talking about mental<br />
phenomena (e.g. “that is too much for my brain to compute”).<br />
198 | P a g e
phenomena (being) which <strong>the</strong>y are meant to explain. In <strong>the</strong> case of materialism <strong>the</strong> most<br />
basic form of this knowledge is expressed in terms of <strong>the</strong> laws of physics, which is why<br />
we will call this assumption methodological physicalism. In its most extreme form this<br />
doctrine is expressed as what has been called „scientism‟, i.e. <strong>the</strong> belief that nothing<br />
really exists but what is scientifically demonstrated. It will be <strong>the</strong> aim of <strong>the</strong> following<br />
two chapters to question <strong>the</strong> validity of methodological physicalism.<br />
How is this assumption expressed in cognitive science? For example, it is accepted as<br />
an unquestioned fact that our primary <strong>and</strong> most basic access to <strong>the</strong> world is objectcentered<br />
<strong>and</strong> defined by purely physical terms, such that “all mammals live in basically<br />
<strong>the</strong> same sensory-motor world of permanent objects arrayed in a representational space”<br />
(<strong>Tom</strong>asello 1999, p. 16). In <strong>the</strong> case of humans this basic access eventually gets<br />
complemented during development by a folk psychology that lets us „see through <strong>the</strong><br />
surface‟ of bodily movements such as arm extensions, finger curlings, etc., to <strong>the</strong> hidden<br />
intentions of o<strong>the</strong>rs (Meltzoff 1995). But this is only an achievement of higher cognitive<br />
functions. Thus, for example, when an adult talks to an infant too young to comprehend<br />
a joint attentional scene, for <strong>the</strong> infant “<strong>the</strong> adult is just making noises” (<strong>Tom</strong>asello<br />
1999, p. 100), i.e. <strong>the</strong> physical sounds produced by wet tissue rapidly slapping toge<strong>the</strong>r<br />
in <strong>the</strong> throat cavity. Methodological physicalism is thus at <strong>the</strong> root of <strong>the</strong> famous<br />
„problem of o<strong>the</strong>r <strong>mind</strong>s‟, i.e. <strong>the</strong> question of how it is possible to „see through <strong>the</strong><br />
surface‟, <strong>and</strong> guides <strong>the</strong> interpretation of most empirical results.<br />
In contrast to <strong>the</strong> metaphysically biased starting point of mainstream cognitive science<br />
<strong>the</strong> enactive approach is based on phenomenological considerations, i.e. a return to <strong>the</strong><br />
primacy of experience itself. Its central starting point is <strong>the</strong> claim that our primary mode<br />
of underst<strong>and</strong>ing <strong>the</strong> world is in terms of sense-making. If we pay close attention to <strong>the</strong><br />
manner in which we live through concrete situations, for example, we can say that <strong>the</strong><br />
brewing storm feels menacing, a cozy pub looks inviting, a wind chime can sound<br />
playful, etc. And <strong>the</strong>se expressive qualities are not mere poetic qualifiers added to our<br />
perceptual experience after we are presented with a raw physical object. More precisely,<br />
this expressiveness comes before <strong>the</strong> object, it is <strong>the</strong> object‟s condition of possibility: an<br />
uneasy feeling about <strong>the</strong> wea<strong>the</strong>r focuses our attention on <strong>the</strong> darkening horizon, <strong>the</strong> bits<br />
of joyful conversation coming from a soft light down an alleyway reveal a pub on closer<br />
199 | P a g e
inspection, soft melodies drifting in from <strong>the</strong> balcony turn out to be dangling pieces of<br />
wood gently making contact in a late summer breeze, <strong>and</strong> so forth. In this manner we<br />
primarily perceive aspects of <strong>the</strong> world in <strong>the</strong>ir immediate sense for us, as menacing,<br />
inviting, playful, etc., even before attentive reflection reveals „a storm‟, „a pub‟, „a wind<br />
chime‟, etc., as detached from our situation <strong>and</strong> devoid of any significance.<br />
By taking <strong>the</strong> phenomenology of our situatedness as our starting point, ra<strong>the</strong>r than <strong>the</strong><br />
materialist half of Cartesian dualist metaphysics, we transform <strong>the</strong> „problem of o<strong>the</strong>r<br />
<strong>mind</strong>s‟ into something more manageable. Instead of explaining why children start<br />
perceiving o<strong>the</strong>rs as acting for reasons on <strong>the</strong> basis of seeing mere physical objects <strong>and</strong><br />
movements (an absolute gap between meaningless physical change <strong>and</strong> meaningful<br />
behavior), we need to account for how a general perception of sense can develop into a<br />
perception of goals <strong>and</strong> intentions (a merely relative gap between different kinds of<br />
meaningful behavior). As a first step in this direction, we introduce <strong>the</strong> phenomenology<br />
of intersubjectivity (Chapter 12). On <strong>the</strong> basis of this change from a metaphysical to a<br />
phenomenological starting point it is possible to organize <strong>the</strong> empirical data of social<br />
psychology, developmental studies, <strong>and</strong> primatology in a novel manner, <strong>the</strong>reby giving<br />
us a fresh perspective on <strong>the</strong> issue of cumulative cultural development (Chapter 13).<br />
Finally, <strong>the</strong> <strong>the</strong>sis finishes by offering some reflections on what has been accomplished<br />
<strong>and</strong> pointing to potential implications that are still in need of fur<strong>the</strong>r investigation.<br />
200 | P a g e
12 Phenomenological considerations<br />
In this chapter we will draw on some central insights from Husserlian phenomenology,<br />
especially in relation to <strong>the</strong> problem of intersubjectivity which has taken center stage for<br />
most of <strong>the</strong> tradition. While <strong>the</strong>re is a growing recognition of <strong>the</strong> importance of a<br />
phenomenologically informed approach to intersubjectivity for <strong>the</strong> cognitive sciences<br />
(e.g. Gallagher & Zahavi 2008; Zahavi 2005; 2001), <strong>and</strong> <strong>the</strong> enactive approach in<br />
particular (e.g. Thompson 2007; 2001), <strong>the</strong>se are only <strong>the</strong> tentative beginnings of a<br />
mutually informative research program. For an insightful discussion of <strong>the</strong> relevant<br />
phenomenological literature <strong>the</strong> reader is referred to Zahavi‟s (1996) excellent treatment<br />
of <strong>the</strong> phenomenology of intersubjectivity, to which this chapter owes much. In <strong>the</strong><br />
context of this <strong>the</strong>sis, we will limit our efforts to a consideration of how <strong>the</strong><br />
phenomenology of intersubjectivity can contribute to a better underst<strong>and</strong>ing of <strong>the</strong> <strong>life</strong><strong>mind</strong><br />
<strong>continuity</strong> <strong>the</strong>sis. Of particular interest will be to determine how <strong>the</strong> presence of<br />
o<strong>the</strong>rs impacts on <strong>the</strong> structures of agency <strong>and</strong> sense-making (perception).<br />
This chapter will unfold in three parts: In Section 12.1 we will review aspects of <strong>the</strong><br />
phenomenology of intersubjectivity in relation to perception, which is how Husserl first<br />
got drawn into a serious appreciation of <strong>the</strong> constitutive role of intersubjectivity, <strong>and</strong><br />
which has also recently been used to motivate a similar turn in <strong>the</strong> cognitive sciences<br />
(e.g. Gallagher 2008a; Zahavi 2005, pp. 166-167). In Section 12.2 we will clarify <strong>the</strong><br />
way in which o<strong>the</strong>rs are given to us in our experience, giving special attention to how<br />
this encounter affects our perspective on <strong>the</strong> world. These phenomenological reflections<br />
indicate how <strong>the</strong> objectivist epistemic perspective that is characteristic of detached<br />
human sense-making (e.g. <strong>the</strong> scientific attitude) is constitutively dependent on our<br />
relationship with o<strong>the</strong>rs. In Section 12.3 <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis is reformulated<br />
to include this phenomenological support for <strong>the</strong> constitutive role of sociality.<br />
12.1 The phenomenology of perception<br />
One of <strong>the</strong> key insights of <strong>the</strong> phenomenology of intersubjectivity is that <strong>the</strong> existence<br />
of o<strong>the</strong>rs plays a constitutive role for our perception. In order to illustrate this insight, let<br />
us begin with a concrete example of object perception: how is <strong>the</strong> wall that is located<br />
201 | P a g e
ehind <strong>the</strong> desk given to me in my experience? How is its presence for me as an object<br />
constituted? The general proposal of <strong>the</strong> enactive approach to perception is that <strong>the</strong><br />
world (our experiential world) is enacted, that is, it claims that perception consists in<br />
perceptually guided action. The perceiver does not have access to some metaphysically<br />
supposed pre-given, perceiver independent world of objects; an autonomous agent is<br />
always embedded within a particular sensory-motor loop that is shaped by <strong>the</strong> overall<br />
dynamics of <strong>the</strong> organism-environment systemic whole. Accordingly, as discussed at<br />
length in Chapter 3, such an agent can only perceive its surroundings by appropriately<br />
regulating its sensory-motor interactions:<br />
Thus <strong>the</strong> overall concern of an enactive approach to perception is not to<br />
determine how some perceiver-independent world is to be recovered; it is,<br />
ra<strong>the</strong>r, to determine <strong>the</strong> common principles or lawful linkages between sensory<br />
<strong>and</strong> motor systems that explain how action can be perceptually guided in a<br />
perceiver-dependent world. (Varela, et al. 1991, p. 173)<br />
On this view, perceived objects appear as <strong>the</strong> invariants that happen to emerge from <strong>the</strong><br />
closed loop of an agent‟s ongoing embodied activity <strong>and</strong> <strong>the</strong> resulting stimulations (an<br />
idea inherited from <strong>the</strong> cybernetic tradition, e.g. von Foerster 1976). It is important to<br />
emphasize that we are specifically talking about sensory-motor invariants, which differ<br />
from <strong>the</strong> ecological invariants of Gibsonian psychology in that only <strong>the</strong> former give a<br />
constitutive role to motor activity (cf. Mossio & Taraborelli 2008). In brief, as <strong>the</strong><br />
philosopher Alva Noë has recently put it, “<strong>the</strong> invariant structure of reality unfolds in<br />
<strong>the</strong> active exploration of appearances” (Noë 2004, p. 85). Since it is my behavior which<br />
enables me to establish such regularities, it follows that my perceptual capabilities are<br />
enabled <strong>and</strong> constrained by my behavioral capabilities:<br />
How things look to me is constrained by my sensorimotor knowledge. It is my<br />
possession of basic sensorimotor skills (which include <strong>the</strong> abilities to move <strong>and</strong><br />
point <strong>and</strong> <strong>the</strong> dispositions to respond by turning <strong>and</strong> ducking, <strong>and</strong> <strong>the</strong> like) that<br />
enables my experience to acquire visual content at all. (Noë 2004, p. 90; cf.<br />
O‟Regan & Noë 2001).<br />
202 | P a g e
Thus, according to <strong>the</strong> enactive approach, my perception of <strong>the</strong> wall behind my desk is<br />
constituted by my possession of basic sensory-motor skills, such as perhaps scanning it<br />
with my eyes or moving my head <strong>and</strong> body in a certain manner 38 . Perception is, after<br />
all, conceived as perceptually guided action. This sensory-motor account of perception<br />
has generated a lot of excitement in cognitive science, but also a number of criticisms<br />
(e.g. Prinz 2006; Clark 2006; Velmans 2007). These need to be taken seriously <strong>and</strong><br />
carefully addressed in order to move <strong>the</strong> <strong>the</strong>ory of sense-making <strong>and</strong> enactive<br />
perception forward (e.g. Thompson 2005; Di Paolo 2009). Here we will focus on one<br />
problematic aspect for postulating a sensory-motor basis for perception, namely <strong>the</strong><br />
perceptual presence of absent phenomena, or what is sometimes called <strong>the</strong> problem of<br />
“perceptual presence” (Noë 2004, p. 59). This problem is of particular interest in <strong>the</strong><br />
current context because a related worry in phenomenology has been resolved by appeal<br />
to <strong>the</strong> constitutive role of open intersubjectivity.<br />
Let us return to <strong>the</strong> example of <strong>the</strong> wall behind <strong>the</strong> desk. How come it is given to me in<br />
my perceptual experience as a 3D object that has ano<strong>the</strong>r side which is currently out of<br />
view? How come it is given to me as separating my current location from whatever is<br />
outside <strong>the</strong> room, ra<strong>the</strong>r than just as a flat 2D appearance? Noë suggests that our<br />
experience of absent profiles should be understood as a type of „virtual‟ presence:<br />
“They are present to perception as accessible” (Noë 2004, p. 63; emphasis added). In<br />
o<strong>the</strong>r words, on this account I do not experience <strong>the</strong> wall as a flat facade that separates<br />
me from some meaningless void because of my embodied sensory-motor skills, which<br />
would let me view its o<strong>the</strong>r side from <strong>the</strong> outside, if I left <strong>the</strong> room to inspect it.<br />
But is this appeal to „virtual‟ presence <strong>and</strong> sensory-motor accessibility an adequate<br />
description of our perceptual experience? According to <strong>the</strong> later work of Husserl <strong>and</strong> <strong>the</strong><br />
subsequent phenomenological tradition, as well as recent work in cognitive science, this<br />
is not <strong>the</strong> case (cf. Gallagher 2008a; Zahavi 1996, pp. 43-51; Gallagher & Zahavi 2008,<br />
38 It is important to emphasize again that <strong>the</strong>re are important differences between <strong>the</strong> enactive paradigm,<br />
which has been developed by Varela <strong>and</strong> colleagues <strong>and</strong> is <strong>the</strong> focus of this <strong>the</strong>sis, <strong>and</strong> <strong>the</strong> „enactive‟<br />
sensory-motor approach to perception that has been proposed by Noë (cf. Torrance 2005; Thompson<br />
2005). However, <strong>the</strong>y are sufficiently similar with respect to <strong>the</strong> role of embodied action for perception<br />
with respect to <strong>the</strong> current issue that <strong>the</strong>se differences need not concern us at this point.<br />
203 | P a g e
p. 100-104; Thompson 2007, p. 384). The idea that we constitute <strong>the</strong> absent profiles of<br />
an object in terms of past or future embodied action necessitates an appeal to a temporal<br />
separation, an aspect that is itself not given in our perceptual experience. Even though I<br />
momentarily do not visually perceive <strong>the</strong> backside directly, I never<strong>the</strong>less experience<br />
<strong>the</strong> wall as having such a backside now while I am looking at it from inside my room. In<br />
o<strong>the</strong>r words, it is argued that since leaving <strong>the</strong> room to check <strong>the</strong> wall‟s backside would<br />
involve a temporal <strong>and</strong> spatial displacement from my current perceptual situation,<br />
nei<strong>the</strong>r <strong>the</strong> movement nor its potential accessibility can explain <strong>the</strong> fact that I currently<br />
perceive <strong>the</strong> wall as a whole object, no matter which side aspect is given to me, ra<strong>the</strong>r<br />
than as a temporarily distributed set of profiles. As Gallagher, following Husserl, points<br />
out: “When I perceive an object <strong>the</strong> present front is not a front with respect to a past or<br />
future back, but is determined through its reference to a present co-existing back. The<br />
object is perceived at any given moment as possessing a plurality of co-existing<br />
profiles” (Gallagher 2008a, p. 172). But since I can only be in one place at a time, how<br />
<strong>the</strong>n can we account for this perceptual presence of co-existing profiles?<br />
We can begin to resolve this problem for <strong>the</strong> enactive approach by noting that Varela<br />
<strong>and</strong> colleagues make use of a much broader notion of embodiment than that used in <strong>the</strong><br />
sensory-motor approach by Noë <strong>and</strong> O‟Regan. The term “embodied action” is indeed<br />
meant to highlight that cognition <strong>and</strong> perception depend on having a body with various<br />
kinds of skills. But <strong>the</strong> term is similarly meant to emphasize “that <strong>the</strong>se individual<br />
sensorimotor capacities are <strong>the</strong>mselves embedded in a more encompassing biological,<br />
psychological, <strong>and</strong> cultural context” (Varela, et al. 1991, pp. 172-173). To be sure,<br />
much of enactive cognitive science has focused on <strong>the</strong> biological <strong>and</strong> psychological<br />
context, but it is also informed by considerations of our social <strong>and</strong> cultural background,<br />
especially in terms of our role as practicing scientists (cf. Varela, et al. 1991, pp. 9-12;<br />
Thompson 2007; 2001). Moreover, even in <strong>the</strong> very beginning of <strong>the</strong> enactive approach<br />
<strong>the</strong> role of social context enters into <strong>the</strong> very definition of what it means to be an<br />
intelligent agent, such that “intelligence shifts from being <strong>the</strong> capacity to solve a<br />
problem to <strong>the</strong> capacity to enter into a shared world of significance” (Varela, et al., p<br />
207; emphasis added). How does this broader notion of embodiment, which includes<br />
our situatedness in a social <strong>and</strong> cultural context, help us to account for our perception of<br />
<strong>the</strong> wall as a full-fledged object?<br />
204 | P a g e
First, following Gallagher (2008a, p. 171), we can note that <strong>the</strong>re is good evidence from<br />
developmental psychology that we gain access to a meaningful world of objects through<br />
our interactions with o<strong>the</strong>rs. Not only are <strong>the</strong> most dominant <strong>and</strong> central experiences for<br />
a young infant its relations to o<strong>the</strong>rs, <strong>the</strong>se relations also shape its perception of <strong>the</strong><br />
world by engaging in joint attention. In o<strong>the</strong>r words, environmental objects, such as a<br />
wall, first take on meaning in <strong>the</strong> pragmatic contexts within which we see <strong>and</strong> imitate<br />
<strong>the</strong> actions of o<strong>the</strong>rs (cf. Merleau-Ponty 1960; Trevar<strong>the</strong>n & Hubley 1978; <strong>Tom</strong>asello<br />
1999). Second, even in our adult <strong>life</strong> we find that <strong>the</strong> wall is given as „a wall‟ within a<br />
common public totality of surroundings, i.e. in phenomenological terms <strong>the</strong> situation of<br />
our individual being is always already a form of „being-with o<strong>the</strong>rs‟ (cf. Heidegger<br />
1927; Zahavi 1996, pp. 124-127). Similarly, research in cognitive anthropology has<br />
demonstrated that our relations to <strong>the</strong> objects of our perception continue to be deeply<br />
interwoven with our relations to o<strong>the</strong>rs even in adult <strong>life</strong> (cf. Hutchins 1995). Thus, <strong>the</strong><br />
capacity for worldly engagement that is characteristic for adult humans is nei<strong>the</strong>r<br />
acquired nor performed in isolation. As Husserl famously puts it:<br />
Thus everything objective that st<strong>and</strong>s before me in experience <strong>and</strong> primarily in<br />
perception has an apperceptive horizon of possible experience, own <strong>and</strong> foreign.<br />
Ontologically speaking, every appearance that I have is from <strong>the</strong> very beginning<br />
a part of an open endless, but explicitly realized totality of possible appearances<br />
of <strong>the</strong> same, <strong>and</strong> <strong>the</strong> subjectivity belonging to this appearance is open<br />
intersubjectivity. (Hua XIV/289; see also Hua IX/294, XV/497; quoted by<br />
Zahavi 2005, p. 167)<br />
In this manner Husserl was led via a deepening phenomenological analysis of object<br />
perception from an individualistic <strong>the</strong>ory of <strong>the</strong> constitution of objects, which was<br />
somewhat reminiscent of Noë <strong>and</strong> O‟Regan‟s recent sensory-motor approach, to an<br />
appreciation of <strong>the</strong> constitutive role of o<strong>the</strong>r subjects. More precisely, following<br />
Zahavi‟s <strong>and</strong> Gallagher‟s interpretation, in order to account for <strong>the</strong> phenomenology of<br />
an object as something that transcends <strong>the</strong> current perceptual profile that we have of it,<br />
we need to posit <strong>the</strong> possibility of o<strong>the</strong>r subjects, which can potentially perceive <strong>the</strong><br />
o<strong>the</strong>r profiles at <strong>the</strong> same time. We could even go as far as to say that <strong>the</strong> transcendence<br />
205 | P a g e
of <strong>the</strong> world as such is derived from <strong>the</strong> radical transcendence of <strong>the</strong> o<strong>the</strong>r (e.g. Levinas<br />
1978). Or, as Sartre puts it: “it is not in <strong>the</strong> world that <strong>the</strong> O<strong>the</strong>r is first to be sought but<br />
at <strong>the</strong> side of consciousness as a consciousness in which <strong>and</strong> by which consciousness<br />
makes itself be what it is” (1943, p. 296). Here we have a phenomenological equivalent<br />
to <strong>the</strong> self-o<strong>the</strong>r co-determination of <strong>the</strong> enactive approach (Thompson 2001), though<br />
<strong>the</strong> latter would be more inclined to view <strong>the</strong> self-world-o<strong>the</strong>r relation in an essentially<br />
reciprocal manner since it is also <strong>the</strong> world which frames <strong>the</strong> factual possibility of<br />
encountering <strong>the</strong> o<strong>the</strong>r.<br />
For our present purposes we can leave <strong>the</strong> broader implications of intersubjectivity for<br />
<strong>the</strong> phenomenon of worldhood aside, <strong>and</strong> focus on <strong>the</strong> claim that it is <strong>the</strong> possibility of<br />
engaging in intersubjective interaction, an open intersubjectivity, which accounts for <strong>the</strong><br />
presence of <strong>the</strong> whole object. According to this phenomenological account, <strong>the</strong> reason<br />
that I experience <strong>the</strong> wall of <strong>the</strong> room in its full presence through its current partial<br />
profile (what is sometimes called „transcendence in immanence‟) is because I can<br />
engage in intersubjective interactions that shape <strong>the</strong> sense of my experience. I can, for<br />
example, hold a conversation with someone outside my window who informs me that<br />
<strong>the</strong> outer surface of <strong>the</strong> house could use some cleaning 39 . Of course, <strong>the</strong> claim is not that<br />
it is necessary that ano<strong>the</strong>r subject is factually present at <strong>the</strong> time for object-perception<br />
to be possible, which is why it specifically is an open intersubjectivity. Ra<strong>the</strong>r, it is<br />
claimed that objects are experienced as existing independently of us because we coconstitute<br />
<strong>the</strong> experiential domain in which <strong>the</strong> object is located as a public realm of<br />
shared meaning that includes <strong>the</strong> possibility of ano<strong>the</strong>r perspective.<br />
39 While it is beyond <strong>the</strong> scope of this chapter to address <strong>the</strong> role of language in <strong>the</strong> co-constitution of our<br />
objective world, it is clearly a crucial element: “In <strong>the</strong> experience of dialogue, <strong>the</strong>re is constituted<br />
between <strong>the</strong> o<strong>the</strong>r person <strong>and</strong> myself a common ground; […] we are collaborators for each o<strong>the</strong>r in<br />
consummate reciprocity. Our perspectives merge into each o<strong>the</strong>r, <strong>and</strong> we co-exist through a common<br />
world” (Merleau-Ponty 1945, p. 413). It is this kind of merging of perspectives which is necessary to<br />
account for <strong>the</strong> presence of an object as an object, namely by providing a syn<strong>the</strong>sis of co-existing<br />
perceptual profiles. More fascinating work remains to be done here, especially because language was a<br />
focal topic in <strong>the</strong> biology of cognition by Maturana (e.g. Maturana 1978; Maturana, et al. 1995).<br />
206 | P a g e
But this intersubjective resolution of <strong>the</strong> problem of „perceptual presence‟ leaves us<br />
with a dilemma with regard to <strong>the</strong> <strong>continuity</strong> aspect of <strong>the</strong> LMCT. For if we accept that<br />
<strong>the</strong> full presence of <strong>the</strong> wall as a 3D object is constituted in terms of <strong>the</strong> potentiality of<br />
my experience of o<strong>the</strong>rs as o<strong>the</strong>rs that perceive its currently hidden profiles, <strong>the</strong>n what<br />
does this entail for o<strong>the</strong>r forms of <strong>life</strong>? We seem to have two controversial options: (i)<br />
we claim that open intersubjectivity (in <strong>the</strong> sense of being able to take ano<strong>the</strong>r‟s<br />
perspective on <strong>the</strong> world) is present for o<strong>the</strong>r living beings as well, or (ii) we assert that<br />
<strong>the</strong> presence of a 3D world is a unique aspect of human phenomenology. While <strong>the</strong><br />
former claim remains highly contentious even with regard to our closest primate<br />
relatives (<strong>Tom</strong>asello, et al. 2003) <strong>and</strong> lacks scientific support for most o<strong>the</strong>r species<br />
(<strong>Tom</strong>asello 1999), <strong>the</strong> latter is in tension with <strong>the</strong> evident skill with which <strong>the</strong>se species<br />
negotiate <strong>the</strong>ir environments. Should we really conceive of <strong>the</strong> lived world of <strong>the</strong>se<br />
species as lacking <strong>the</strong> dimension of depth? At least for animals with evident depthadapted<br />
sense organs (e.g. vision, hearing, forms of touch, etc.), this is an unacceptable<br />
conclusion. We propose to resolve this dilemma in two steps.<br />
First, let us reconsider <strong>the</strong> constitution of perceptual presence in terms of sensory-motor<br />
invariants. In <strong>the</strong> case of humans, at least, psychological experiments using minimalist<br />
haptic interfaces have shown that <strong>the</strong> emergence of an experience of distal presence (in<br />
terms of an obstacle located in 3D space) is entailed by <strong>the</strong> skilful mastery of basic<br />
sensory-motor correlations <strong>and</strong> laws (Auvray, et al. 2005; 2007). However, as already<br />
indicated above, phenomenologists have argued that this kind of sensor-motor activity<br />
includes a temporal progression which prevents it from accounting for <strong>the</strong> perceptual<br />
presence of a complete object, because “<strong>the</strong> object is perceived at any given moment as<br />
possessing a plurality of co-existing profiles” (Gallagher 2008a, p. 172). How can <strong>the</strong><br />
sequence of sensory-motor events be integrated into a coherent whole? Is <strong>the</strong> appeal to<br />
open intersubjectivity necessary, even if it prevents us from attributing <strong>the</strong> experience of<br />
distal presence to solitary animals?<br />
We suggest a novel resolution of <strong>the</strong> tension by appealing to ano<strong>the</strong>r foundational <strong>the</strong>me<br />
of phenomenology <strong>and</strong> enactive cognitive science, namely <strong>the</strong> issue of temporality (cf.<br />
Varela 1999). Husserl himself has argued that every experiential moment is not simply<br />
an isolated point in time, but is ra<strong>the</strong>r temporally extended in terms of a tripartite<br />
207 | P a g e
etention-present-protention structure (cf. Hua X). Importantly, it is in this temporal<br />
horizon of backward retention <strong>and</strong> forward protention that perception incorporates its<br />
non-actualized possibilities. Accordingly, <strong>the</strong> existence of <strong>the</strong> temporal horizon of <strong>the</strong><br />
present moment allows us to explain <strong>the</strong> solitary constitution of an object as presently<br />
consisting of a plurality of o<strong>the</strong>r possible perspectives. These potential profiles can be<br />
contained in <strong>the</strong> present moment‟s retention <strong>and</strong> protention, even without <strong>the</strong> need for<br />
open intersubjectivity.<br />
While Husserl‟s account of inner time consciousness is derived from human experience,<br />
it is possible that his account of its tripartite structure defines a more general condition<br />
of lived experience as such. It could <strong>the</strong>refore potentially hold for o<strong>the</strong>r living beings as<br />
well. In o<strong>the</strong>r words, if we accept that depth perception can be grounded in tripartite<br />
temporality, <strong>the</strong>n we have managed to recover <strong>the</strong> possibility of such perception even<br />
for non-social animals, as long as <strong>the</strong>ir lived existence is characterized by this kind of<br />
temporality. To be sure, in <strong>the</strong> phenomenological literature <strong>the</strong> status of temporality for<br />
non-human animals remains ambiguous <strong>and</strong> controversial (e.g. Buchanan 2007; Hayes<br />
2007). From <strong>the</strong> perspective of <strong>the</strong> enactive paradigm, however, <strong>the</strong>re are reasons to be<br />
optimistic. For instance, <strong>the</strong>re has been work by Varela (1999) <strong>and</strong> van Gelder (1999)<br />
which links <strong>the</strong> phenomenology of time consciousness with a particular organization of<br />
neural dynamics. Accordingly, we can hypo<strong>the</strong>size that animals with nervous systems<br />
that are organized in a manner sufficiently similar to ours are also embedded in a<br />
tripartite temporality. More fundamentally, as Jonas (1996) has argued at length, <strong>the</strong>re<br />
are good reasons to claim that even metabolic forms of <strong>life</strong> have some of <strong>the</strong> existential<br />
credentials characteristic of human beings. The only time when a living being coincides<br />
with itself is when it has died. Life is essentially a temporal form of existence, as its<br />
essential form is only maintained by continuous material change. More precisely, <strong>the</strong><br />
satisfaction of metabolic needs is necessarily a precarious affair which creates a need to<br />
project toward future possibilities on <strong>the</strong> basis of past events. Whe<strong>the</strong>r this primordial<br />
temporality of <strong>life</strong> formally matches Husserl‟s description of human tripartite time<br />
consciousness would require fur<strong>the</strong>r work. In any case, <strong>the</strong>re is a strong possibility that<br />
<strong>the</strong> problem of perceptual presence faced by <strong>the</strong> sensory-motor approach can be<br />
resolved in terms of temporal integration without appeal to open intersubjectivity,<br />
perhaps even for non-human forms of <strong>life</strong>.<br />
208 | P a g e
Have we gone too far? It appears that our argument has undermined <strong>the</strong> phenomenology<br />
of intersubjectivity in preference for a solution that gives support to methodological<br />
individualism. To be sure, with <strong>the</strong> appeal to <strong>the</strong> temporal structure of experience we<br />
have indicated how it is possible to retain <strong>the</strong> insights of <strong>the</strong> sensory-motor approach to<br />
perceptual presence without requiring that o<strong>the</strong>r subjects must potentially be present as<br />
o<strong>the</strong>rs. But where does this leave <strong>the</strong> constitutive role of open intersubjectivity? This is<br />
where <strong>the</strong> second step comes into play. We claim that <strong>the</strong> aspect of perception which is<br />
intersubjectively co-constituted is a certain kind of detachment, an attitude of „taking<br />
as‟, which enables us to specifically perceive something as something. For example, I<br />
don‟t just see <strong>the</strong> wall from <strong>the</strong> perspective of my own current concerns (e.g. as trapping<br />
me inside <strong>the</strong> room) but also from <strong>the</strong> perspective of alternative concerns including<br />
those o<strong>the</strong>rs might have (e.g. as blocking <strong>the</strong> view, making parking difficult, etc). These<br />
alternative perspectives on <strong>the</strong> situation need not be actually <strong>the</strong>matized; <strong>the</strong>ir<br />
potentiality as legitimate concerns makes me see <strong>the</strong> wall as „a wall‟ that is independent<br />
from my current perspective of it, i.e. as an object in <strong>the</strong> strict sense of <strong>the</strong> word. Here<br />
we need to appeal to <strong>the</strong> possible presence of o<strong>the</strong>rs as o<strong>the</strong>rs in order to account for <strong>the</strong><br />
existence of <strong>the</strong>se potential co-existing perspectives of concern, as well as <strong>the</strong><br />
corresponding relativization or de-centering of our own current perceptual perspective.<br />
On this phenomenological view, it is open intersubjectivity which co-constitutes <strong>the</strong><br />
characteristically human de-centered presence, a presence for which <strong>the</strong> sensory-motor<br />
constitution of spatiality is a necessary but not sufficient condition. This appeal to <strong>the</strong><br />
constitutive role of o<strong>the</strong>rs as a means of going beyond <strong>the</strong> limitations of individual<br />
sensory-motor knowledge has occasionally been recognized in sensor-motor approaches<br />
to psychology (e.g. Piaget 1967), but is still lacking in cognitive science. We will return<br />
to this „de-centering‟ presence of <strong>the</strong> o<strong>the</strong>r in <strong>the</strong> next section, in <strong>the</strong> context of a more<br />
detailed phenomenological investigation of how we perceive o<strong>the</strong>rs.<br />
12.2 The phenomenology of intersubjectivity<br />
How precisely do we encounter o<strong>the</strong>rs as o<strong>the</strong>rs? In orthodox cognitive science this<br />
„problem of o<strong>the</strong>r <strong>mind</strong>s‟ is usually addressed according to <strong>the</strong> computationalist sense-<br />
209 | P a g e
model-plan-act schema: (i) we first perceive a set of physical facts, such as material<br />
bodies <strong>and</strong> movements; (ii) <strong>the</strong>se facts are <strong>the</strong> input for some kind of cognitive<br />
processing, such as inference or simulation, <strong>the</strong>reby producing a representation of <strong>the</strong><br />
o<strong>the</strong>r‟s mental states, (iii) that representation allows us to plan how to act in a socially<br />
appropriate manner, <strong>and</strong> (iv) we respond by executing that plan with motor comm<strong>and</strong>s<br />
as best as possible (for a critical analysis of inference <strong>and</strong> simulation based approaches,<br />
cf. Gallagher 2001).<br />
Though <strong>the</strong>re are some essential differences between <strong>the</strong> competing <strong>the</strong>ories of social<br />
cognition in mainstream cognitive science, <strong>the</strong>y are in agreement that <strong>the</strong> meaningful<br />
expressions of <strong>the</strong> o<strong>the</strong>r are essentially a secondary attribution to <strong>the</strong> primary perception<br />
of merely physical circumstances. This assumption is a direct outcome of what we have<br />
called methodological physicalism, <strong>and</strong> has already been criticized extensively by more<br />
phenomenologically oriented <strong>the</strong>orists. What is most peculiar about this assumption is<br />
that it does not match what is given in our experience: we directly perceive <strong>the</strong> o<strong>the</strong>r as<br />
ano<strong>the</strong>r subject in its own right without having to engage in inference or simulation (cf.<br />
Gallagher 2008b; Zahavi 2001). Moreover, if we accept that part of what it means to<br />
experience an objective world is that we encounter it as a shared existence for o<strong>the</strong>r<br />
subjects, i.e. <strong>the</strong> world is experienced as our common world (Hua I/123; cf. Zahavi<br />
1996, pp. 25-26), <strong>the</strong>n <strong>the</strong> traditional approach, based as it is on an objectivist premise,<br />
presupposes what it sets out to explain. The attempt to reduce <strong>the</strong> presence of o<strong>the</strong>rs to a<br />
combination of physical facts is doomed to failure because <strong>the</strong> condition of possibility<br />
for those facts includes <strong>the</strong> presence of o<strong>the</strong>rs. In brief, it is impossible to conceive of<br />
objectivity without positing intersubjectivity at <strong>the</strong> same time.<br />
However, <strong>the</strong> claim that o<strong>the</strong>rs are immediately given in our experience should not be<br />
misunderstood as asserting that we have direct access to ano<strong>the</strong>r‟s experience. On <strong>the</strong><br />
contrary, a crucial element that defines <strong>the</strong> o<strong>the</strong>r is its peculiar „o<strong>the</strong>rness‟, o<strong>the</strong>rwise<br />
<strong>the</strong> phenomenon of intersubjectivity would be logically inexplicable. In <strong>the</strong> words of<br />
Husserl: “if what belongs to <strong>the</strong> o<strong>the</strong>r‟s own essence were directly accessible, it would<br />
be merely a moment of my own essence, <strong>and</strong> ultimately he himself <strong>and</strong> I myself would<br />
be <strong>the</strong> same” (Hua I/139). To be sure, to say that <strong>the</strong> o<strong>the</strong>r is defined by his o<strong>the</strong>rness is<br />
not enough. After all, all real objects of our experience, including self <strong>and</strong> world, are<br />
210 | P a g e
characterized by a certain type of transcendence in relation to <strong>the</strong> constituting subject.<br />
Moreover, we can even find structures of o<strong>the</strong>rness (alterity) within <strong>the</strong> constituting<br />
subject itself (cf. Zahavi 1999). What is special about <strong>the</strong> transcendence of <strong>the</strong> o<strong>the</strong>r is<br />
that <strong>the</strong> o<strong>the</strong>r necessarily eludes our grasp in a unique manner.<br />
While it is true that objects also elude our grasp, i.e. <strong>the</strong>y are necessarily only given in<br />
profiles that never exhaust <strong>the</strong> constituted object as such, we can still posit <strong>the</strong>ir identity<br />
as an ideal limit point, which we pre-reflectively underst<strong>and</strong> through our mastery of<br />
sensory-motor engagement with <strong>the</strong>m 40 . Ano<strong>the</strong>r subject, in contrast, is always prone to<br />
change in such a way that it escapes any attempt at grasping its identity in <strong>the</strong> form of a<br />
simple object perception. As long as <strong>the</strong> o<strong>the</strong>r remains an autonomous subject in its own<br />
right, <strong>the</strong>re is always <strong>the</strong> possibility that <strong>the</strong> affordances of interaction change in<br />
surprising <strong>and</strong> unexpected ways. In relation to <strong>the</strong> transcendence of things we can say<br />
that “<strong>the</strong> real lends itself to unending exploration; it is inexhaustible” (Merleau-Ponty<br />
1945, p. 378). To be sure, this is also <strong>the</strong> case for our encounters with o<strong>the</strong>r subjects, but<br />
<strong>the</strong>re it runs into what we could call a meta- or second-order transcendence: o<strong>the</strong>rs do<br />
not only lend <strong>the</strong>mselves to unending exploration, <strong>the</strong>y also spontaneously lend<br />
<strong>the</strong>mselves to unending explorations of different styles of unending exploration.<br />
This insistence on <strong>the</strong> radical o<strong>the</strong>rness of <strong>the</strong> o<strong>the</strong>r might at first seem like a minor<br />
technical point, but it actually is at <strong>the</strong> heart of why a consideration of intersubjectivity<br />
is so important for any adequate account of human cognition. It is only in <strong>the</strong> case when<br />
<strong>the</strong> subject encounters <strong>the</strong> particular transcendence of <strong>the</strong> o<strong>the</strong>r that we can say that its<br />
immanent sphere of „ownness‟ is surpassed toward a shared world of objects (Hua<br />
XIV/442). In this way we have turned <strong>the</strong> traditional problem of o<strong>the</strong>r <strong>mind</strong>s on its<br />
head: “<strong>the</strong> o<strong>the</strong>rness of „someone else‟ becomes extended to <strong>the</strong> whole world, as its<br />
„Objectivity‟, giving it this sense in <strong>the</strong> first place” (Hua I/173). We thus find that <strong>the</strong><br />
40 “The ipseity is, of course, never reached: each aspect of <strong>the</strong> thing which falls to our perception is still<br />
only an invitation to perceive beyond it, still only a momentary halt in <strong>the</strong> perceptual process. […] What<br />
makes <strong>the</strong> „reality‟ of <strong>the</strong> thing is <strong>the</strong>refore precisely what snatches it from our grasp. The aseity of <strong>the</strong><br />
thing, its unchallengeable presence <strong>and</strong> <strong>the</strong> perpetual absence into which it withdraws, are two<br />
inseperable aspects of transcendence” (Merleau-Ponty 1945, p. 271). Might we here have <strong>the</strong> basis for a<br />
phenomenological explanation of Schrödinger‟s uncertainty principle?<br />
211 | P a g e
categories of transcendence, objectivity <strong>and</strong> reality are intersubjectively constituted, that<br />
is, <strong>the</strong>y can only be constituted by a subject who has experienced o<strong>the</strong>r subjects. More<br />
precisely, <strong>the</strong>se categories are co-constituted, as illustrated by <strong>the</strong> following example of<br />
<strong>the</strong>ir application to our underst<strong>and</strong>ing of ourselves:<br />
Thus for me <strong>the</strong> O<strong>the</strong>r is first <strong>the</strong> being for whom I am an object; that is, <strong>the</strong><br />
being through whom I gain my objectness. If I am to be able to conceive of even<br />
one of my properties in <strong>the</strong> objective mode, <strong>the</strong>n <strong>the</strong> O<strong>the</strong>r is already given. […]<br />
In experiencing <strong>the</strong> look, in experiencing myself as an unrevealed object-ness, I<br />
experience <strong>the</strong> inapprehensible subjectivity of <strong>the</strong> O<strong>the</strong>r directly <strong>and</strong> with my<br />
being. (Sartre 1943, p. 294)<br />
Accordingly, even our presence to ourselves as a temporal-spatial object in <strong>the</strong> world is<br />
a phenomenon that is mediated by <strong>the</strong> presence of <strong>the</strong> o<strong>the</strong>r (Sartre 1943, p. 290-291).<br />
Moreover, Husserl claims that <strong>the</strong> same holds for <strong>the</strong> categories of immanence,<br />
appearance, <strong>and</strong> inwardness. It is only when I experience myself as an object under <strong>the</strong><br />
gaze of ano<strong>the</strong>r subject, that I can distinguish my personal inwardness from its public<br />
external manifestation. Similarly, only when a subject experiences that <strong>the</strong> same object<br />
can be experienced by several subjects, <strong>and</strong> that it is given in various profiles, that <strong>the</strong><br />
subject is in a position to realize that <strong>the</strong>re is a distinction between <strong>the</strong> object itself <strong>and</strong><br />
its appearance, its being-for-me (cf. Zahavi 1996, pp. 38-39).<br />
What might lived experience be like for a subject who has not been able to perceive<br />
o<strong>the</strong>rs as o<strong>the</strong>rs? While this is extremely difficult to imagine from our socialized <strong>and</strong><br />
enculturated perspective, <strong>the</strong> transformative power of <strong>the</strong> o<strong>the</strong>r is still evident from<br />
within <strong>the</strong> perspective of our own adult existence, so we can entertain some tentative<br />
comparative reflections. Merleau-Ponty, for instance, observes that “no sooner has my<br />
gaze fallen upon a living body in <strong>the</strong> process of acting than <strong>the</strong> objects surrounding it<br />
immediately take on a fresh layer of significance: <strong>the</strong>y are no longer simply what I<br />
myself could make of <strong>the</strong>m, <strong>the</strong>y are what this o<strong>the</strong>r pattern of behavior is about to<br />
make of <strong>the</strong>m” (1945, p. 411-412). This is a good example of how our own sensemaking<br />
can be shaped by <strong>the</strong> presence of ano<strong>the</strong>r subject. Ano<strong>the</strong>r fitting example is<br />
212 | P a g e
Sartre‟s description of <strong>the</strong> experience of being in a public park, facing a lawn with a row<br />
of benches along its edge, when a man happens to walk by those benches:<br />
Thus suddenly an object has appeared which has stolen <strong>the</strong> world from me.<br />
Everything is in place; everything still exists for me; but everything is traversed<br />
by an invisible flight <strong>and</strong> fixed in <strong>the</strong> direction of <strong>the</strong> new object. The<br />
appearance of <strong>the</strong> O<strong>the</strong>r in <strong>the</strong> world corresponds <strong>the</strong>refore to a fixed sliding of<br />
<strong>the</strong> whole universe, to a decentralization of <strong>the</strong> world which undermines <strong>the</strong><br />
centralization which I am simultaneously effecting. (Sartre 1943, p. 279)<br />
It is in situations like <strong>the</strong>se, namely when we are prompted to make sense of <strong>the</strong> world<br />
in relation to <strong>the</strong> perspective of ano<strong>the</strong>r subject, that we also become aware of our own<br />
contributions to <strong>the</strong> structure of our experience, for example <strong>the</strong> centralization which we<br />
ourselves continually effect on our perceptual world. In this way we can reaffirm that<br />
<strong>the</strong> de-centered presence, which is characteristic of our existence in <strong>the</strong> world, is coconstituted<br />
by <strong>the</strong> presence of <strong>the</strong> o<strong>the</strong>r. It is likely that non-social forms of <strong>life</strong> exist in<br />
a centralized world that is not given as centralized (because <strong>the</strong> comparative experience<br />
of sharing <strong>the</strong> world with ano<strong>the</strong>r centralizing perspective is missing). We will return to<br />
<strong>the</strong>se comparative considerations in <strong>the</strong> next section.<br />
This completes <strong>the</strong> brief review of <strong>the</strong> phenomenology of intersubjectivity. Of special<br />
interest was how our experience of self <strong>and</strong> world is shaped <strong>and</strong> co-constituted by <strong>the</strong><br />
possible <strong>and</strong> actual presence of o<strong>the</strong>r subjects. We have seen that this guiding question<br />
touched upon <strong>the</strong> basic perceptual presence of objects, as well as on <strong>the</strong> conditions for<br />
notions of objectivity, appearance, inwardness, <strong>and</strong> transcendence. Our detached <strong>and</strong><br />
de-centered presence in <strong>the</strong> world, which is a condition of possibility for <strong>the</strong> scientific<br />
attitude, is an intersubjective achievement. On this basis we can now return to <strong>the</strong><br />
LMCT <strong>and</strong> provide some clarification of its key concepts.<br />
12.3 A phenomenologically informed <strong>continuity</strong> <strong>the</strong>sis<br />
The main motivation for this <strong>the</strong>sis was an analysis of <strong>the</strong> constitutive role of sociality<br />
for <strong>mind</strong> <strong>and</strong> cognition in order to support <strong>the</strong> LMCT as a unifying framework for<br />
213 | P a g e
cognitive science. We have pursued this goal by drawing on insights developed in two<br />
traditions, namely <strong>the</strong> enactive paradigm <strong>and</strong> Husserlian phenomenology. The aim of<br />
this section is to combine <strong>the</strong> insights of <strong>the</strong>se traditions in a mutually informative<br />
manner, <strong>and</strong> reformulate <strong>the</strong> LMCT accordingly.<br />
It should be clear that relating <strong>the</strong> enactive <strong>and</strong> <strong>the</strong> phenomenological traditions in a<br />
fruitful manner is both a challenge <strong>and</strong> an opportunity. We have argued that <strong>the</strong><br />
organizational (or behavioral) approach to <strong>the</strong> LMCT is not enough, <strong>and</strong> that we need to<br />
incorporate phenomenological considerations. However, <strong>the</strong> appeal to <strong>the</strong> first-person<br />
perspective might cause some resistance in cognitive science, especially in <strong>the</strong><br />
cognitivist mainstream, while introducing <strong>the</strong> enactive approach into current debates in<br />
phenomenology could also be met with some skepticism:<br />
Phenomenologists never conceive of intersubjectivity as an objectively existing<br />
structure in <strong>the</strong> world that can be described <strong>and</strong> analyzed from a third-person<br />
perspective. On <strong>the</strong> contrary, intersubjectivity as a relation between subjects<br />
must be analyzed, as such, from a first-person <strong>and</strong> a second-person perspective.<br />
(Zahavi 2005, p. 176)<br />
In order to resolve <strong>the</strong>se tensions it is helpful to re<strong>mind</strong> ourselves that we are dealing<br />
with a subjectivity that is embodied <strong>and</strong> embedded, both of which are characteristics<br />
that can be explored from <strong>the</strong> perspectives of science <strong>and</strong> phenomenology. And <strong>the</strong><br />
same applies to intersubjectivity as well: “we must consider <strong>the</strong> relation with o<strong>the</strong>rs not<br />
only as one of <strong>the</strong> contents of our experience but as an actual structure in its own right”<br />
(Merleau-Ponty 1960, p. 140). Indeed, <strong>the</strong> enactive approach to social cognition is well<br />
positioned to provide <strong>the</strong> phenomenological tradition with a dynamical account of <strong>the</strong><br />
interaction process, while <strong>the</strong> latter can sharpen <strong>the</strong> sensitivity of <strong>the</strong> former to <strong>the</strong><br />
phenomena that need explaining. Moreover, both traditions are joined in <strong>the</strong>ir focus on<br />
<strong>the</strong> primacy of embodied (ra<strong>the</strong>r than linguistic) interactions (cf. Zahavi 2005, p. 176).<br />
It is with respect to <strong>the</strong> constitutive impact of such embodied interactions that <strong>the</strong><br />
enactive approach can be of help to <strong>the</strong> phenomenological tradition, for example by<br />
relating <strong>the</strong> changing qualitative presence of <strong>the</strong> o<strong>the</strong>r to <strong>the</strong> particular dynamics of<br />
changes in ongoing bodily coordination:<br />
214 | P a g e
[The o<strong>the</strong>r‟s] autonomy dem<strong>and</strong>s frequent readjustments of my individual sensemaking.<br />
When interaction <strong>and</strong> individual intentions coordinate, we feel mutually<br />
skilful to navigate <strong>the</strong> interaction: we experience a kind of transparency of <strong>the</strong><br />
o<strong>the</strong>r-in-interaction. But when, for a variety of reasons, a breakdown occurs, <strong>and</strong><br />
until a new coordination is attained, we experience <strong>the</strong> o<strong>the</strong>r as opaque. (De<br />
Jaegher & Di Paolo 2007, p. 504)<br />
It is <strong>the</strong>refore likely that an integrative methodology that combines both dynamical <strong>and</strong><br />
phenomenological insights would be mutually beneficial. However, <strong>the</strong>re still remains a<br />
lingering conceptual tension. The phenomenological analysis of intersubjectivity has led<br />
us to argue that “intersubjectivity exists <strong>and</strong> develops in relation between world-related<br />
subjects, <strong>and</strong> <strong>the</strong> world is brought to articulation only in <strong>the</strong> relation between subjects”<br />
(Zahavi 2005, p. 177). However, this st<strong>and</strong>s in contrast to <strong>the</strong> basic enactive account of<br />
agency <strong>and</strong> sense-making, which posits <strong>the</strong> bringing forth of a world for <strong>the</strong> adaptive<br />
agent without any mention of <strong>the</strong> constitutive role of o<strong>the</strong>r agents.<br />
We will use this tension to our advantage by forcing us to become clearer about <strong>the</strong><br />
enactive approach to agency <strong>and</strong> sense-making. The lack of intersubjectivity is not fatal<br />
to <strong>the</strong> enactive account of <strong>the</strong>se basic notions, but we must be careful that we do not<br />
over-anthropomorphize <strong>the</strong>m. But how should we conceive of <strong>the</strong> experiential world of<br />
an agent who is incapable of interacting with o<strong>the</strong>rs as o<strong>the</strong>rs in <strong>the</strong>ir own right? To be<br />
sure, it is likely that it is characterized by some minimal transcendence, since sensorymotor<br />
regularities are partly dependent on environmental affordances, <strong>and</strong> thus<br />
inherently escape <strong>the</strong> agent‟s grasp to some extent. Moreover, we have argued that even<br />
basic sensory-motor skills will give rise to <strong>the</strong> perceptual presence of complete objects<br />
without <strong>the</strong> need to appeal to open intersubjectivity, namely due to <strong>the</strong> intrinsic tripartite<br />
temporal structure of lived experience. Never<strong>the</strong>less, without <strong>the</strong> relativizing <strong>and</strong> decentering<br />
presence of <strong>the</strong> o<strong>the</strong>r as o<strong>the</strong>r, this kind of alterity of <strong>the</strong> world is likely to<br />
remain undifferentiated from <strong>the</strong> alterity already encountered within <strong>the</strong> isolated subject<br />
itself (i.e. <strong>the</strong> opacity of <strong>the</strong> agent‟s relation to itself). Thus, whereas Jonas claims that<br />
“inwardness is coextensive with <strong>life</strong>” (1966, p. 58), a position which has been<br />
influential in <strong>the</strong> recent development of <strong>the</strong> notion of sense-making (cf. Weber & Varela<br />
215 | P a g e
2002; Di Paolo 2005; Thompson 2007), this is no longer a precise enough description of<br />
<strong>the</strong> phenomenon of <strong>life</strong>. As Husserl remarks in <strong>the</strong> case of an imagined solitary human:<br />
For <strong>the</strong> human being who has not undergone <strong>the</strong> experience of empathy, or from<br />
<strong>the</strong> st<strong>and</strong>point of <strong>the</strong> abstraction from any empathy, <strong>the</strong>re is no “inwardness” of<br />
an “externality”; such a human being would have all of <strong>the</strong> lived experiences –<br />
<strong>and</strong> all of <strong>the</strong> objectivities, of whatever sort – that are included under <strong>the</strong> title of<br />
inwardness, but <strong>the</strong> concept of inwardness would be lost. (Hua XIII/420; quoted<br />
by Zahavi 1996, p. 39)<br />
The solitary basic agent is thus best described as experiencing an „inwardness‟, but an<br />
inwardness which is not experienced as an inwardness. This agent would still be a<br />
center of needs <strong>and</strong> concerns embedded in a meaningful context related to its particular<br />
circumstances <strong>and</strong> viability constraints, as described by Jonas, but this differentiation as<br />
a center for a world is not a structure of its experience as such (cf. Heidegger 1929).<br />
What we have in this case is <strong>the</strong> „organism-Umwelt‟ dyad that is so well described in <strong>the</strong><br />
work of <strong>the</strong> biologists von Uexküll (1934). But in order for <strong>the</strong> distinction between<br />
„inner‟ <strong>and</strong> „outer‟ to become present in experience as such it is necessary that we are<br />
dealing with an intersubjectively constituted form of <strong>life</strong>, one which unfortunately does<br />
not get addressed by Jonas (1966) 41 .<br />
On this view, it is only with a certain process of socialization that <strong>the</strong>re is a possibility<br />
of making sense of one‟s existence as a sense-making existence. Thus, if we want to<br />
talk about <strong>the</strong> kind of sense-making activities that are involved in enacting <strong>the</strong> physical<br />
world which we experience from a detached human perspective, <strong>the</strong>n appealing to a<br />
simple „organism-Umwelt‟ dyad is not sufficient. Following <strong>the</strong> phenomenological<br />
tradition, we first have to elucidate <strong>the</strong> human condition in terms of a „self-world-o<strong>the</strong>r‟<br />
triad. This is because <strong>the</strong> subject-object dichotomy, which is at <strong>the</strong> heart of <strong>the</strong><br />
<strong>the</strong>oretical attitude that makes abstract knowledge possible, is only an idealized <strong>and</strong><br />
41 Indeed, considering that all known forms of <strong>life</strong> are closely interconnected in various kinds of<br />
networks, e.g. <strong>the</strong> relationships of predator <strong>and</strong> prey, mating partners <strong>and</strong> rivals, symbiosis, <strong>and</strong> <strong>the</strong> whole<br />
ecosystem context, <strong>the</strong> social nature of <strong>life</strong> is a striking omission in Jonas‟ bio-phenomenology.<br />
216 | P a g e
derivative perspective which requires intersubjectivity as its necessary foundation. The<br />
central question that needs to be addressed by <strong>the</strong> LMCT, <strong>the</strong>refore, is not how to get<br />
from <strong>the</strong> basic „organism-Umwelt‟ to <strong>the</strong> human „self-world‟ structure, but ra<strong>the</strong>r how to<br />
get from <strong>the</strong> former to a „self-world-o<strong>the</strong>r‟ structure. And <strong>the</strong>n, only on <strong>the</strong> basis of this<br />
transition, is it possible to determine <strong>the</strong> conditions for <strong>the</strong> emergence of <strong>the</strong> subjectobject<br />
dichotomy which has been mistakenly taken as <strong>the</strong> primary epistemic attitude by<br />
mainstream cognitive science. Nothing specific has been said about <strong>the</strong> conditions for<br />
this dichotomy in this <strong>the</strong>sis, but <strong>the</strong>re is no reason to believe that <strong>the</strong> LMCT cannot<br />
also accommodate this final transition. Indeed, it appears that Heidegger‟s (1927) claim<br />
that this transition is caused by break-downs in ongoing coping can be addressed in <strong>the</strong><br />
framework of enactive cognitive science, <strong>and</strong> is perhaps amenable to evolutionary<br />
robotics modeling (cf. Di Paolo & Iizuka 2008).<br />
217 | P a g e
13 Toward an enactive approach to culture<br />
Culture is a rich <strong>and</strong> important topic for <strong>the</strong> enactive approach that has only recently<br />
begun to be addressed (e.g. Thompson 2007; Steiner & Stewart 2009; Stewart, in press;<br />
<strong>Froese</strong> 2009). In Chapter 4 we ventured for a moment into <strong>the</strong> debate about culture by<br />
analyzing Steiner <strong>and</strong> Stewart‟s (2009) emphasis of <strong>the</strong> essential heteronomy of cultural<br />
values. In brief, <strong>the</strong> idea is that in order to undergo enculturation <strong>the</strong> living subject has<br />
to constrain its own behavioral autonomy in order to appropriate <strong>the</strong> pre-existing social<br />
practices that already form an established context of cultural normativity. Paradoxically,<br />
only through <strong>the</strong> incorporation of <strong>the</strong>se heteronomous constraints does <strong>the</strong> subject<br />
eventually gain an increase in autonomy that reaches beyond <strong>the</strong> acquired skills on <strong>the</strong><br />
socio-cultural stage <strong>and</strong> enables <strong>the</strong> expansion of individual behavioral abilities. This<br />
dialectic of autonomy/constraint, which Jonas (1966) founded on <strong>the</strong> emergence of <strong>life</strong><br />
<strong>and</strong> traced throughout <strong>the</strong> later transitions in evolution, thus finds its expression once<br />
more during enculturation. In effect, <strong>the</strong> process of becoming a part of a culture appears<br />
to be a more specific form of social learning whereby an especially large body of preestablished<br />
practices ends up being incorporated.<br />
However, simple emphasis of <strong>continuity</strong> between sociality <strong>and</strong> culture should not be <strong>the</strong><br />
final word on this manner. Indeed, this <strong>the</strong>sis would not be complete without at least a<br />
cursory discussion of how <strong>the</strong> enactive approach could also approach <strong>the</strong> perplexing<br />
phenomenon of human culture. In o<strong>the</strong>r words, while <strong>the</strong>se first steps toward an<br />
enactive approach to cultural cognition are promising in <strong>the</strong> sense that <strong>the</strong>y appear to<br />
follow <strong>the</strong> same bio-logic that we applied to our analysis of sociality, <strong>the</strong>re still remains<br />
one major worry: <strong>the</strong> specificity of human culture. Thus, even if we accept <strong>the</strong> relatively<br />
controversial claim that <strong>the</strong>re are legitimate ways for us to attribute culture to o<strong>the</strong>r<br />
species (cf. Byrne, et al. 2004), it never<strong>the</strong>less remains to be explained why <strong>the</strong>re is<br />
such a surprising diversity <strong>and</strong> prevalence of human culture. Can <strong>the</strong> enactive paradigm<br />
perhaps shed some light on this mystery?<br />
This chapter attempts to respond to this question in a twofold manner. First, it presents a<br />
preliminary critical analysis of <strong>the</strong> mainstream‟s answer to <strong>the</strong> challenge of explaining<br />
humanity‟s cumulative cultural development. The results of this analysis point to some<br />
218 | P a g e
significant problems that are derived from <strong>the</strong> mainstream‟s uncritical assumption of<br />
methodological individualism (cf. Chapter 5) <strong>and</strong> methodological physicalism (cf.<br />
Chapter 11). Second, throughout this analysis <strong>the</strong> enactive paradigm is presented as a<br />
favorable position from which to move <strong>the</strong> debate forward. More precisely, it can<br />
provide a fresh perspective on cumulative cultural development by diagnosing <strong>the</strong><br />
origins of <strong>the</strong> traditional framework‟s shortcomings, <strong>and</strong> also by offering some new<br />
paths for future research. Of course, at this stage much of this work remains speculative,<br />
<strong>and</strong> it is offered here merely as an opportunity for stimulating fur<strong>the</strong>r debate.<br />
13.1 The ‘ratchet effect’<br />
One popular way of explaining <strong>the</strong> mechanism of cumulative cultural development is in<br />
terms of <strong>the</strong> “ratchet effect” (<strong>Tom</strong>asello, et al. 1993): a combination of faithful imitation<br />
<strong>and</strong> creative innovation. The basic idea is that when a practice is first invented by an<br />
individual it can be quite primitive, but when it is replicated by o<strong>the</strong>rs <strong>the</strong>y might also<br />
make small improvements, which are <strong>the</strong>n in turn copied <strong>and</strong> potentially improved by<br />
o<strong>the</strong>rs, <strong>and</strong> so on over historical time. In this process <strong>the</strong> source of innovation can be as<br />
simple as trial <strong>and</strong> error, <strong>and</strong> imitation apparently only requires accurate copying of <strong>the</strong><br />
physical movements of <strong>the</strong> o<strong>the</strong>r. Note that <strong>the</strong>se two factors appear to be so basic that<br />
<strong>the</strong>y should also be within <strong>the</strong> behavioral capacity of most non-human social animals.<br />
However, cumulative cultural development is arguably a phenomenon that is unique to<br />
humans (<strong>Tom</strong>asello 2001). What explains this discrepancy?<br />
It might be thought that o<strong>the</strong>r animals lack <strong>the</strong> creative capacity for innovation, but this<br />
hypo<strong>the</strong>sis is not supported by <strong>the</strong> empirical evidence: “Perhaps surprisingly, for many<br />
animal species it is not <strong>the</strong> creative component, but ra<strong>the</strong>r <strong>the</strong> stabilizing ratchet<br />
component, that is <strong>the</strong> difficult feat” (<strong>Tom</strong>asello 1999, p. 5). Indeed, it turns out that<br />
many non-human animals, including our closest primate relatives, have great difficulty<br />
in copying <strong>the</strong> precise physical movements of o<strong>the</strong>rs. The traditional explanation of this<br />
incapacity is that imitative learning is “made possible by a very special form of social<br />
cognition, namely, <strong>the</strong> ability of individual organisms to underst<strong>and</strong> conspecifics as<br />
beings like <strong>the</strong>mselves who have intentional <strong>and</strong> mental lives like our own” (<strong>Tom</strong>asello<br />
1999, p. 5). And, so <strong>the</strong> rest of <strong>the</strong> argument goes, since (i) this intentional stance is a<br />
219 | P a g e
type of social underst<strong>and</strong>ing that non-human animals supposedly lack, <strong>and</strong> (ii) imitation<br />
is a necessary prerequisite for <strong>the</strong> emergence of <strong>the</strong> ratchet effect, we appear to have<br />
found a potential biological factor that can explain why human beings have given rise to<br />
cumulative cultural development <strong>and</strong> o<strong>the</strong>r species do not.<br />
As might be expected, from <strong>the</strong> point of view of <strong>the</strong> enactive paradigm this explanation<br />
is unsatisfactory in several respects. Even we assume that something like <strong>the</strong> ratchet<br />
effect is <strong>the</strong> driving mechanism behind cumulative cultural development, nei<strong>the</strong>r of its<br />
two essential components can simply be accepted at face value. However, since this<br />
<strong>the</strong>sis is primarily concerned with sociality, <strong>the</strong> following discussion will not include an<br />
analysis of what is required for <strong>the</strong> ability to innovate <strong>and</strong> only focus on <strong>the</strong> origins of<br />
<strong>the</strong> stabilizing component (i.e. imitative learning) 42 . In brief, <strong>the</strong> central problem of <strong>the</strong><br />
mainstream position is <strong>the</strong> unquestioned assumption of what we have diagnosed as<br />
„methodological physicalism‟. This assumption has created an explanatory blind spot<br />
because our ability to attend to o<strong>the</strong>rs in terms of abstract physical properties has been<br />
taken for granted. In o<strong>the</strong>r words, perception is understood as a form of information<br />
processing whereby mental representations of <strong>the</strong> physical world (including <strong>the</strong> bodies<br />
of o<strong>the</strong>rs), i.e. of <strong>the</strong> world as it is described by physics, are made available in <strong>the</strong> <strong>mind</strong><br />
for fur<strong>the</strong>r processing by o<strong>the</strong>r cognitive processes. It follows that <strong>the</strong> mainstream<br />
position‟s methodological physicalism leaves individual deficits in social underst<strong>and</strong>ing<br />
as <strong>the</strong> only valid explanation for a lack of imitative ability.<br />
In contrast to this position a critical analysis of some of <strong>the</strong> key empirical evidence from<br />
<strong>the</strong> perspective of <strong>the</strong> enactive approach lends support to three claims: (i) for most<br />
animals <strong>the</strong> default mode of perceiving o<strong>the</strong>rs is to perceive <strong>the</strong>m directly in terms of<br />
goals, intentions <strong>and</strong> general mental attitude, (ii) what makes humans special is <strong>the</strong>ir<br />
capacity to override this default mode of perception by attending to o<strong>the</strong>rs in terms of<br />
<strong>the</strong>ir abstract physical properties, <strong>and</strong> (iii) it is our capacity to override <strong>the</strong> default mode,<br />
42 There certainly remains a story to be told about how to ground creativity in <strong>the</strong> notion of autonomous<br />
agency. However, <strong>the</strong> intrinsic openness of an autonomous system to change its own organization in<br />
relation to its current structure <strong>and</strong> history of interactions already gives us a good starting point for this<br />
endeavor. Indeed, it has already been suggested that autonomy may be at <strong>the</strong> heart of creative play among<br />
animals, <strong>and</strong> thus partly responsible for <strong>the</strong>ir higher cognitive functions (Di Paolo, et al. in press).<br />
220 | P a g e
a<strong>the</strong>r than o<strong>the</strong>r animals‟ supposed lack of social underst<strong>and</strong>ing, which explains why<br />
we are better at imitative behavior, <strong>and</strong> <strong>the</strong>refore why specifically our species has given<br />
rise to cumulative cultural development.<br />
In <strong>the</strong> rest of this chapter we will spell out <strong>the</strong>se three claims in more detail. In<br />
particular, we will evaluate important experimental evidence from primatology as well<br />
as developmental <strong>and</strong> social psychology from <strong>the</strong> perspective provided by <strong>the</strong> enactive<br />
paradigm. This critical analysis will give empirical support to <strong>the</strong> claims <strong>and</strong> raise some<br />
questions for future research. The chapter concludes with brief reflections on some<br />
evidence in evolutionary anthropology.<br />
13.2 Primatology<br />
Just a decade ago it was widely believed that non-human primates did not underst<strong>and</strong><br />
<strong>the</strong> intentions of o<strong>the</strong>rs (cf. <strong>Tom</strong>asello 1999), but in recent years this consensus has been<br />
subjected to drastic revisions even though <strong>the</strong> form that a new hypo<strong>the</strong>sis should take is<br />
not entirely clear (<strong>Tom</strong>asello, et al. 2003). Here we will suggest that <strong>the</strong> empirical<br />
evidence of primatology supports our phenomenological starting point, <strong>and</strong> that o<strong>the</strong>r<br />
primates also experience o<strong>the</strong>rs primarily in terms of <strong>the</strong>ir goals, intentions <strong>and</strong> a<br />
general mental attitude. As a first step toward <strong>the</strong> establishment of this new hypo<strong>the</strong>sis<br />
we must question <strong>the</strong> validity of previous findings that appeared as evidence to <strong>the</strong><br />
contrary. Why has it taken researchers so long to realize that non-human primates have<br />
<strong>the</strong> capacity to underst<strong>and</strong> o<strong>the</strong>rs as o<strong>the</strong>r intentional beings?<br />
First of all, we must consider <strong>the</strong> psychological state of <strong>the</strong> lab animals, who often<br />
suffer from profound traumatic experiences. Accordingly, we should always be careful<br />
not to over-generalize any empirical findings in this area of research (cf. Racine, et al.<br />
2008). This is especially true of negative results, which might ra<strong>the</strong>r be <strong>the</strong> result of<br />
social withdrawnness <strong>and</strong> o<strong>the</strong>r individual deficits contingent on a ra<strong>the</strong>r unnatural<br />
developmental history, as well as <strong>the</strong> general arbitrariness of lab experiments from <strong>the</strong><br />
perspective of those being tested. For example, <strong>the</strong> reason why studies which required<br />
chimpanzees to follow a communicative sign to <strong>the</strong> location of food resulted in negative<br />
evidence could be that this kind of gesturing is something that <strong>the</strong>y would never do in<br />
221 | P a g e
<strong>the</strong> wild <strong>and</strong> <strong>the</strong>refore failed to underst<strong>and</strong> (<strong>Tom</strong>asello, et al. 2003). Indeed, it should<br />
thus come as no surprise that implementing experimental situations in more ecologically<br />
plausible ways has led to more positive results of social underst<strong>and</strong>ing (cf. Call &<br />
<strong>Tom</strong>asello 2008) 43 . Similarly, as would be expected, enculturated apes are better at<br />
coping with experimental settings, probably because <strong>the</strong>ir social underst<strong>and</strong>ing is more<br />
akin to that of <strong>the</strong> experimenters (e.g. Savage-Rumbaugh, et al. 2001). We should also<br />
not forget <strong>the</strong> subtle but prevalent power structures impinging on <strong>the</strong> animals that spend<br />
<strong>the</strong>ir <strong>life</strong> under lab conditions. If <strong>the</strong> social <strong>and</strong> cognitive identity of <strong>the</strong> animal has been<br />
significantly impaired (i.e. a kind of mental „death‟), <strong>the</strong>n we should expect to find little<br />
evidence for sociality. Thus, we follow De Jaegher <strong>and</strong> Di Paolo (2008, pp. 38-39) in<br />
insisting that <strong>the</strong> interactors must be autonomous agents in some broad sense so as to be<br />
even capable of engaging in social interactions. In sum, it cannot be emphasized enough<br />
that <strong>the</strong> practice of using a lab animal‟s irresponsiveness to social cues as <strong>the</strong> basis for<br />
denying social abilities to its species as a whole is simply unacceptable.<br />
Second, we can identify a strong tendency toward negative interpretative bias. This is at<br />
least implicitly acknowledged by <strong>Tom</strong>asello <strong>and</strong> colleagues who, as if almost doubting<br />
<strong>the</strong>ir own findings, assure <strong>the</strong> reader that <strong>the</strong>ir “studies show what <strong>the</strong>y seem to show,<br />
namely, that chimpanzees actually know something about <strong>the</strong> content of what o<strong>the</strong>rs see<br />
<strong>and</strong>, at least in some situations, how this governs <strong>the</strong>ir behavior” (<strong>Tom</strong>asello, et al.<br />
2003, p. 155; emphasis added). How does this apparent negative bias manifest itself in<br />
o<strong>the</strong>r experiments? For example, it has been argued in a well-known study by Povinelly<br />
<strong>and</strong> Eddy (1996) that because some apes were found gesturing (i.e. begging for food) to<br />
„blind‟ experimenters (i.e. who wearing buckets over <strong>the</strong>ir heads), <strong>the</strong>y cannot be said to<br />
perceive o<strong>the</strong>rs as intentional beings. However, it has also been shown that congenitally<br />
blind humans gesture normally, even when <strong>the</strong>y speak to ano<strong>the</strong>r blind listener alone<br />
(Iverson & Goldin-Meadow 1998). Strictly speaking, we must <strong>the</strong>refore ei<strong>the</strong>r insist that<br />
blind humans also have no underst<strong>and</strong>ing of o<strong>the</strong>rs as intentional beings, or we treat<br />
Povinelly <strong>and</strong> Eddy‟s empirical data as it is, namely deeply inconclusive. As it turns<br />
43 Human beings may also appear as socially incapable when confronted with unfamiliar situations, an<br />
experience many of us have had in <strong>the</strong> context of foreign cultures. For example, in <strong>the</strong> Philippines it is<br />
common practice to indicate <strong>the</strong> location of a joint attentional target by pointing with <strong>the</strong> lips or <strong>the</strong><br />
mouth, a gesture whose intended meaning might be lost on a foreigner, if it is noticed at all.<br />
222 | P a g e
out, this latter interpretation is even supported by <strong>the</strong> evidence presented by Povinelli<br />
<strong>and</strong> Eddy (1996). They had also found that chimpanzees did discriminate situations in<br />
which one human was facing <strong>the</strong>m <strong>and</strong> ano<strong>the</strong>r had her back turned.<br />
The unfortunate prevalence of developmental disorders, unnatural experimental settings,<br />
<strong>and</strong> a strong negative interpretative bias can help us to explain some of <strong>the</strong> negative<br />
findings of primatology, but <strong>the</strong>y do not cover all relevant cases. Consider for example<br />
an experimental study by <strong>Tom</strong>asello <strong>and</strong> colleagues (1997) in which a chimpanzee was<br />
removed from her group <strong>and</strong> taught arbitrary signals by means of which she obtained<br />
desired food from a human. When she was eventually returned to <strong>the</strong> group <strong>and</strong> began<br />
to use <strong>the</strong>se newly learned gestures to obtain treats from <strong>the</strong> experimenters while in full<br />
view of <strong>the</strong> o<strong>the</strong>r chimpanzees, <strong>the</strong>re was not one observed instance of o<strong>the</strong>rs attempting<br />
to imitate <strong>the</strong> effective gestures. This lack of imitation is especially surprising since <strong>the</strong><br />
rest of <strong>the</strong> chimpanzees were observing <strong>the</strong> gesturer in action <strong>and</strong> <strong>the</strong>y were <strong>the</strong>mselves<br />
highly motivated for <strong>the</strong> food. The conclusion proposed by <strong>Tom</strong>asello <strong>and</strong> colleagues<br />
was that chimpanzees do not underst<strong>and</strong> each o<strong>the</strong>r as intentional agents, since this is a<br />
necessary pre-requisite for imitation <strong>and</strong> <strong>the</strong> chimps failed to exhibit such behavior.<br />
But our phenomenologically clarified starting point offers a different explanation: what<br />
if <strong>the</strong> o<strong>the</strong>r chimpanzees did underst<strong>and</strong> <strong>the</strong> trained individual to be gesturing for food,<br />
but <strong>the</strong>y failed to attend to <strong>the</strong> physical manner in which this gesture was realized? In<br />
o<strong>the</strong>r words, <strong>the</strong>y perhaps perceived <strong>the</strong> expressiveness of <strong>the</strong> gesture as a goal-directed<br />
action aimed at obtaining some desired food, but <strong>the</strong>y could not abstractly attend to <strong>the</strong><br />
particular physical movements by which this gesture was embodied, <strong>and</strong> <strong>the</strong>refore failed<br />
to imitate <strong>the</strong> gesture correctly. Thus, if we bracket <strong>the</strong> assumption of methodological<br />
physicalism, it becomes conceivable that <strong>the</strong> problem that is faced by <strong>the</strong> chimpanzees<br />
is not a failure to underst<strong>and</strong> <strong>the</strong> intentions of <strong>the</strong> o<strong>the</strong>rs, but ra<strong>the</strong>r <strong>the</strong> inability to adopt<br />
a detached perspective which could help <strong>the</strong>m to reveal aspects of <strong>the</strong> o<strong>the</strong>r‟s presence<br />
in terms of its „mere‟ physical properties. In this way we have reversed <strong>the</strong> traditional<br />
epistemic hierarchy such that directly perceiving o<strong>the</strong>rs as expressive, intentional beings<br />
is <strong>the</strong> primary experience, <strong>and</strong> <strong>the</strong> capacity of perceiving o<strong>the</strong>rs as physical objects is a<br />
secondary, additional achievement.<br />
223 | P a g e
We are thus led to a reversal of <strong>the</strong> guiding questions of this field. Ra<strong>the</strong>r than being<br />
faced by <strong>the</strong> „problem of o<strong>the</strong>r <strong>mind</strong>s‟ we are now confronted with <strong>the</strong> „problem of<br />
o<strong>the</strong>r bodies‟: what are <strong>the</strong> evolutionary <strong>and</strong> developmental origins of our ability to<br />
attend to <strong>the</strong> presence of o<strong>the</strong>rs in terms of <strong>the</strong>ir abstract physical properties? In order to<br />
begin answering this question we can turn to <strong>the</strong> evidence of child development.<br />
13.3 Developmental <strong>and</strong> social psychology<br />
If we want to better underst<strong>and</strong> <strong>the</strong> origin of our ability to abstractedly attend to <strong>the</strong><br />
physical properties of o<strong>the</strong>r subjects, it is useful to consider work on neonate imitation<br />
because this latter ability might presuppose some capacity for <strong>the</strong> former. In Chapter 6<br />
we already discussed <strong>the</strong> case of neonate imitation as an interesting example of bodily<br />
coordination because it can take place even without knowledge of a visually formed<br />
body image (e.g. Meltzoff & Moore 1977). But did we not suggest that such imitation<br />
demonstrated <strong>the</strong> effective role of <strong>the</strong> interaction process for organizing an individual‟s<br />
behavior? Is <strong>the</strong> ability to adopt a detached perceptual attitude thus even a necessary<br />
postulate to explain <strong>the</strong>se results?<br />
A closer look at <strong>the</strong> experimental protocol reveals that <strong>the</strong> authors indeed invested a<br />
considerable amount of effort to make sure that elements of social interaction could not<br />
be an explanatory factor in <strong>the</strong> emergence of imitative behavior. For instance, <strong>the</strong> adults<br />
were instructed by <strong>the</strong> experimenters to perform gestures at equally long intervals,<br />
separated by a neutral face that was unresponsive to <strong>the</strong> infant‟s gestures. The idea was<br />
to show that neonates were capable of imitating gestures, as well as even improving<br />
<strong>the</strong>se gestures, without social feedback about how well <strong>the</strong>y were doing. Of course, if<br />
social feedback was really such a confounding factor, <strong>the</strong>n we can hypo<strong>the</strong>size that <strong>the</strong>y<br />
would perform even better within an appropriate social context. Moreover, it is very<br />
likely that <strong>the</strong> non-social experiments were conducted within a larger social context,<br />
though not much is said of this in <strong>the</strong> papers on neonate imitation. For example, <strong>the</strong><br />
baby probably has to be sufficiently animated <strong>and</strong> engaged by <strong>the</strong> experimenter so that<br />
it is interested to direct its attention appropriately <strong>and</strong> concentrates on <strong>the</strong> relevant<br />
events of what is going on around it.<br />
224 | P a g e
The problem is that if we cannot even appeal to such a minimal socially mediated<br />
context in order to explain neonate imitation, <strong>the</strong>n <strong>the</strong> results appear to be in conflict<br />
with those of <strong>the</strong> „double TV monitor‟ experiments by Murray <strong>and</strong> Trevar<strong>the</strong>n (1985),<br />
Nadel, et al. (1999) <strong>and</strong> o<strong>the</strong>rs (cf. Chapter 8). There it was demonstrated that if infants<br />
are faced with a „social‟ interaction that lacks social contingency, i.e. where <strong>the</strong> o<strong>the</strong>r<br />
participant is not responsive to <strong>the</strong>ir gestures, <strong>the</strong>y become distressed <strong>and</strong>/or removed.<br />
Though <strong>the</strong> double TV monitor experiments lack a temporal analysis of <strong>the</strong> progression<br />
of infant behavior during <strong>the</strong> replay condition, we can hypo<strong>the</strong>size that <strong>the</strong>re is a short<br />
period during which infants are still responsive, <strong>and</strong> that this period is sufficiently long<br />
to make Meltzoff <strong>and</strong> Moore‟s studies possible. Never<strong>the</strong>less, since Meltzoff <strong>and</strong> Moore<br />
are essentially interested in demonstrating <strong>the</strong> ability of neonates to imitate in situations<br />
that lack social contingency, it might be an appropriate challenge for future research to<br />
combine <strong>the</strong> two experimental paradigms so as to test <strong>the</strong> response of <strong>the</strong> neonates to a<br />
video playback of <strong>the</strong> experimenter‟s gestures. If <strong>the</strong> outcome in this control condition<br />
is significantly different <strong>the</strong>n we would have confirmed our suspicion that even in <strong>the</strong><br />
original study <strong>the</strong>re might have been some subtle social contingency at play that was<br />
affecting <strong>the</strong> behavior of <strong>the</strong> infants.<br />
Be this as it may, let us assume for <strong>the</strong> sake of argument that it is indeed possible to get<br />
young infants to imitate gestures of an unresponsive „partner‟, especially if <strong>the</strong>re was a<br />
preparatory period of social interaction beforeh<strong>and</strong> <strong>and</strong> <strong>the</strong> experiment itself does not<br />
last too long. In o<strong>the</strong>r words, we take it as given that human infants are well predisposed<br />
to displaying imitative behavior <strong>and</strong> that this capacity can be expressed to some extent<br />
even without <strong>the</strong> need for an ongoing interaction process. Moreover, we assume that <strong>the</strong><br />
internal enabling conditions for this imitative ability are much less developed or lacking<br />
in o<strong>the</strong>r primate species. Never<strong>the</strong>less, at this point it needs to be emphasized that, of<br />
course, <strong>the</strong>se assumptions are highly speculative <strong>and</strong> that more research remains to be<br />
done. Moreover, it is clear that simply shifting <strong>the</strong> burden of <strong>the</strong> explanation to some<br />
kind of innate bias that separates human infants from <strong>the</strong> young of o<strong>the</strong>r primate species<br />
does not accomplish much in terms of our underst<strong>and</strong>ing of <strong>the</strong> phenomenon. Let us<br />
<strong>the</strong>refore turn our attention to an investigation of <strong>the</strong> kind of situations in which <strong>the</strong><br />
imitative behavior of human infants is typically expressed.<br />
225 | P a g e
To begin with we can note that <strong>the</strong>re is empirical evidence which suggests that whereas<br />
chimpanzees are more likely to reproduce <strong>the</strong> target goal of a demonstration but not <strong>the</strong><br />
particular means of attaining that goal (a form of behavior sometimes called „emulation<br />
learning‟), human children typically imitate <strong>the</strong> demonstrator‟s actual physical actions<br />
(e.g. Call, et al. 2005). The traditional explanation of this behavioral difference, as we<br />
might already expect, is that non-human primates simply cannot underst<strong>and</strong> o<strong>the</strong>rs as<br />
being intentional subjects like <strong>the</strong>mselves. But this explanation is awkward given that<br />
(i) nothing is said about why <strong>the</strong> capacity to adopt an intentional stance should lead to<br />
imitation ra<strong>the</strong>r than emulation (i.e. it might be necessary but not sufficient for <strong>the</strong><br />
former option), <strong>and</strong> (ii) both chimpanzees <strong>and</strong> human children appear to underst<strong>and</strong><br />
equally well <strong>the</strong> goal of <strong>the</strong> demonstrator‟s unfolding behavior. Therefore, a much more<br />
parsimonious explanation is that both underst<strong>and</strong> <strong>the</strong> o<strong>the</strong>r in terms of goal-directed<br />
behavior, but that human infants appear to have <strong>the</strong> additional capacity to pay attention<br />
to <strong>the</strong> particular means of how this goal is achieved. In o<strong>the</strong>r words, it seems that human<br />
infants can more easily attend to o<strong>the</strong>rs in terms of <strong>the</strong>ir abstract physical properties <strong>and</strong><br />
<strong>the</strong>n make use of this additional information. Can we specify more precisely under what<br />
circumstances <strong>the</strong> infants make us of this ability?<br />
First of all, it is important to note that it has indeed been shown that infants are not<br />
simply imitators of <strong>mind</strong>less physical movements, but are actually able to underst<strong>and</strong><br />
<strong>the</strong> intentions of o<strong>the</strong>rs as well. In a well-known study by Meltzoff (1995), for example,<br />
it was demonstrated that 18-month-old children could infer <strong>the</strong> adult‟s intended act by<br />
watching failed attempts, since <strong>the</strong>y proceeded to complete <strong>the</strong> intended act <strong>the</strong>mselves.<br />
Moreover, compelling evidence suggests that cases in which infants directly imitate <strong>the</strong><br />
means of ano<strong>the</strong>r‟s goal-directed behavior can be explained by <strong>the</strong> fact that <strong>the</strong> o<strong>the</strong>r‟s<br />
action appeared to be arbitrary under <strong>the</strong> given circumstances (cf. Gergely, et al. 2002).<br />
In o<strong>the</strong>r words, if a performed action does not appear to be contingent on any factors<br />
that are relevant to <strong>the</strong> current situation, <strong>the</strong> performance of that particular action might<br />
never<strong>the</strong>less be a necessary means of achieving <strong>the</strong> goal, albeit a means which <strong>the</strong> infant<br />
does not (yet) underst<strong>and</strong>. Gergely <strong>and</strong> colleagues argue that in such ambiguous<br />
situations it is more rational for <strong>the</strong> infant to imitate <strong>the</strong> action, since it is better to err on<br />
<strong>the</strong> side of safety, but o<strong>the</strong>rwise it is more appropriate to simply select what appears to<br />
be one‟s best available means for achieving <strong>the</strong> goal (emulation). Indeed, it has been<br />
226 | P a g e
shown that even infants as young as 12 months underst<strong>and</strong> a demonstrator‟s action in<br />
terms of <strong>the</strong> meaning <strong>the</strong>se choices have to achieve <strong>the</strong> intended goal, <strong>and</strong> that <strong>the</strong>y use<br />
this underst<strong>and</strong>ing when deciding whe<strong>the</strong>r or not to imitate particular aspects of that<br />
action (Schwier, et al. 2006). We <strong>the</strong>refore suggest that human infants primarily make<br />
sense of o<strong>the</strong>rs as being agents like <strong>the</strong>mselves, <strong>and</strong> only resort to an abstract evaluation<br />
of <strong>the</strong> o<strong>the</strong>r‟s physical movements when <strong>the</strong>ir primary sense-making of <strong>the</strong> o<strong>the</strong>r agent<br />
remains inconclusive about that o<strong>the</strong>r‟s particular intentions.<br />
13.4 Evolutionary anthropology<br />
In <strong>the</strong> previous section we have presented some empirical evidence which nicely<br />
complements <strong>the</strong> phenomenologically informed critique of methodological physicalism<br />
that we developed in Chapters 11 <strong>and</strong> 12. It is interesting to recall that in <strong>the</strong> end of that<br />
phenomenological analysis we were led to suggest that <strong>the</strong> human ability for abstraction<br />
is perhaps based on a socially mediated form of object perception, i.e. object perception<br />
that has been decentered by open intersubjectivity. Interestingly, this appears to place us<br />
into a typical „chicken <strong>and</strong> egg‟ dilemma: (i) we have argued that detached object<br />
perception is an outcome of (adult) human intersubjectivity, <strong>and</strong> (ii) we have accepted<br />
<strong>the</strong> claim of <strong>Tom</strong>asello <strong>and</strong> colleagues that human culture requires imitation, which is a<br />
behavior that arguably depends on detached object perception itself. It <strong>the</strong>refore appears<br />
to be impossible to say which of <strong>the</strong> two came first!<br />
Fortunately, <strong>the</strong> enactive paradigm is well positioned in order to transform this apparent<br />
vicious circle into ano<strong>the</strong>r one of its „creative circles‟ (cf. Varela 1984). Indeed, <strong>the</strong><br />
modeling experiments that were presented in Chapters 7 to 10 have already shown <strong>the</strong><br />
possibility of an inter-individual interaction process that simultaneously enables <strong>and</strong> is<br />
enabled by <strong>the</strong> individual behavior. The relationship between imitation <strong>and</strong> sociality<br />
might <strong>the</strong>refore be ano<strong>the</strong>r example of a co-dependence that bootstraps itself into<br />
existence. We can now reformulate <strong>the</strong> question of cumulative cultural development in<br />
enactive terms: Could <strong>the</strong> crucial difference between chimpanzees <strong>and</strong> humans perhaps<br />
be that <strong>the</strong> latter are somehow better at providing <strong>the</strong> conditions of emergence for this<br />
autonomous process of enculturation? While it is beyond <strong>the</strong> scope of this <strong>the</strong>sis to<br />
provide an adequate response to this question, we can never<strong>the</strong>less submit it as a novel<br />
227 | P a g e
working hypo<strong>the</strong>sis for <strong>the</strong> enactive paradigm. At least <strong>the</strong> rest of this <strong>the</strong>sis has already<br />
begun to provide <strong>the</strong> <strong>the</strong>oretical, ma<strong>the</strong>matical <strong>and</strong> phenomenological framework which<br />
could be <strong>the</strong> basis for a more systematic investigation. Finally, we will simply conclude<br />
this chapter by highlighting some points of interest related to evolutionary anthropology<br />
which could be potential starting points of this future research.<br />
Let us begin by considering what is entailed by our newly developed underst<strong>and</strong>ing of<br />
imitative behavior. We have argued that <strong>the</strong> phenomenological st<strong>and</strong>point commits us to<br />
<strong>the</strong> view that perception of o<strong>the</strong>rs occurs primarily in terms of meaning <strong>and</strong> intentions,<br />
<strong>and</strong> that perception of o<strong>the</strong>rs in terms of <strong>the</strong>ir abstract physical properties is a secondary<br />
achievement. This perspective has allowed us to propose a novel interpretation of some<br />
crucial evidence in primatology <strong>and</strong> developmental studies, <strong>and</strong> led us to replace <strong>the</strong><br />
„problem of o<strong>the</strong>r <strong>mind</strong>s‟ with <strong>the</strong> „problem of o<strong>the</strong>r bodies‟. Note that this enactive<br />
reformulation of <strong>the</strong> central problem has important consequences for our underst<strong>and</strong>ing<br />
of hominid evolution. Thus, when investigating <strong>the</strong> historical beginnings of cumulative<br />
cultural development, we are no longer looking for <strong>the</strong> origin of <strong>the</strong> capacity for „<strong>mind</strong>reading‟,<br />
but ra<strong>the</strong>r for <strong>the</strong> emergence of our ability to specifically attend to <strong>the</strong> abstract<br />
physical properties <strong>and</strong> movements of o<strong>the</strong>rs‟ bodies in relation to <strong>the</strong> world.<br />
In fact, changing our perspective in this manner throws up new puzzles for evolutionary<br />
anthropology because <strong>the</strong> ability to perceive o<strong>the</strong>rs in terms of <strong>the</strong>ir intentions clearly<br />
has adaptive benefits. For example, it is more flexible in that it enables an underst<strong>and</strong>ing<br />
of o<strong>the</strong>rs not only in “previously observed or highly similar situations but also in novel<br />
situations” (Call & <strong>Tom</strong>asello 2008, p. 187). Conversely, it can even be maladaptive to<br />
copy <strong>the</strong> precise movements of someone else, especially if <strong>the</strong> purpose of <strong>the</strong> action is<br />
beyond one‟s underst<strong>and</strong>ing (i.e. it might be idiosyncratic <strong>and</strong> unnecessary or even plain<br />
wrong <strong>and</strong> dangerous). Never<strong>the</strong>less, we have seen that <strong>the</strong>re is strong evidence that this<br />
is precisely <strong>the</strong> typical behavior of young infants. We can <strong>the</strong>refore ask: what adaptive<br />
benefits could be associated with <strong>the</strong> ability to attend to o<strong>the</strong>rs in terms of <strong>the</strong> abstract<br />
movements of <strong>the</strong>ir physical bodies?<br />
One interpretation of <strong>the</strong> empirical evidence in infant studies is that imitation is <strong>the</strong><br />
rational choice of means when <strong>the</strong> reasons for <strong>the</strong> o<strong>the</strong>r‟s particular action are unclear in<br />
228 | P a g e
elation to <strong>the</strong> current circumstances <strong>and</strong> <strong>the</strong> target goal (cf. Gergely, et al. 2002). But<br />
we have already noted that imitation is not always beneficial. So <strong>the</strong> question is: what is<br />
<strong>the</strong> context that enables this to be a rational choice in most cases? To be sure, imitative<br />
behavior is advantageous in situations which often involve arbitrary actions of o<strong>the</strong>rs,<br />
<strong>and</strong> where it is important to attend to <strong>the</strong>se actions in detail. A paradigmatic example of<br />
such a situation is a complex symbolic context, in which <strong>the</strong> meaning of actions is fixed<br />
by historical convention alone. In o<strong>the</strong>r words, physical imitation highly out-competes<br />
emulation when trying to interact with o<strong>the</strong>rs in terms of language. It should <strong>the</strong>refore<br />
come as no surprise that Homo sapiens appears to be <strong>the</strong> only primate species which is<br />
capable of accomplished vocal imitation (Fitch 2000). However, this brings us back to<br />
<strong>the</strong> creative circularity of enculturation because we can now ask: is our ability to imitate<br />
an evolutionary outcome or an enabling condition of this linguistic context?<br />
We can get a better h<strong>and</strong>le on this question by considering <strong>the</strong> evidence of comparative<br />
psychology. For example, it has recently been demonstrated that at least enculturated<br />
chimpanzees are also capable of performing imitative behavior. Like <strong>the</strong> young infants<br />
in Gergely et al.‟s (2002) study, <strong>the</strong>y appear to have a similar underst<strong>and</strong>ing of <strong>the</strong><br />
rationality of o<strong>the</strong>rs‟ intentional actions in relation to <strong>the</strong> goal of <strong>the</strong> task, <strong>and</strong> <strong>the</strong>y can<br />
use this underst<strong>and</strong>ing to evaluate when it is better to imitate an action ra<strong>the</strong>r than to<br />
emulate it (Buttelmann, et al. 2007). This study is noteworthy in two important respects:<br />
(i) it provides additional evidence that chimpanzees can underst<strong>and</strong> o<strong>the</strong>rs as intentional<br />
beings like <strong>the</strong>mselves, which is what we already would expect, <strong>and</strong> (ii) it demonstrates<br />
that <strong>the</strong>y can also attend to <strong>the</strong> abstract physical movements. Most importantly, it turns<br />
out that <strong>the</strong> chimpanzees make use of this ability for abstraction only in situations when<br />
<strong>the</strong> o<strong>the</strong>r‟s reason for choosing a particular means of acting toward <strong>the</strong> intended goal<br />
has remained ambiguous. Thus, <strong>the</strong> fact that chimpanzees do not appear to imitate like<br />
human infants when observed in <strong>the</strong> wild, but are spontaneously able to do so when <strong>the</strong>y<br />
are appropriately embedded within a human cultural background, suggests that human<br />
imitation might have historically originated primarily as part of a developmental ra<strong>the</strong>r<br />
than an evolutionary process. Of course, it is possible that <strong>the</strong> prevalent existence of this<br />
special context put additional selective pressure on individuals to develop <strong>the</strong> ability to<br />
abstract <strong>and</strong> imitate as quickly as possible. In o<strong>the</strong>r words, <strong>the</strong> possibly innate ability of<br />
229 | P a g e
human neonates to imitate facial gestures might thus be a result of <strong>the</strong> Baldwin effect<br />
(Baldwin 1897) ra<strong>the</strong>r than a cause of cumulative cultural development.<br />
This enactive approach to culture lends itself to ano<strong>the</strong>r speculation. At a certain point<br />
in our history, when <strong>the</strong> ability to attend to <strong>the</strong> movement of o<strong>the</strong>rs abstractly <strong>and</strong> <strong>the</strong><br />
concomitant capacity for imitation were developed in sufficiently detailed manner, new<br />
complex practices could begin to be preserved in a precise manner. In o<strong>the</strong>r words, we<br />
suggest that it was through this refined combination of abstraction <strong>and</strong> imitation that a<br />
social medium could first be established. This medium has at least three important<br />
properties: (i) its manifestation is concrete because it is effectively embodied in <strong>the</strong><br />
behavioral practices of <strong>the</strong> individual participants; (ii) its structure is arbitrary because<br />
<strong>the</strong>se practices are essentially contingent on historically determined conventions; <strong>and</strong><br />
(iii) its existence is independent of <strong>the</strong> particular set of individuals which momentarily<br />
manifest it concretely through <strong>the</strong>ir arbitrarily structured behavior. All of <strong>the</strong>se factors<br />
potentially entail qualitative changes in <strong>the</strong> historical trajectories that are related to this<br />
medium: (i) its concrete manifestation can give rise to new forms of creative innovation<br />
that are based on <strong>the</strong> particular properties of <strong>the</strong> medium itself (Shanon 1998); (ii) its<br />
arbitrary structure enables it accommodate an open-ended complexity of forms that are<br />
underdetermined by immediate needs; <strong>and</strong> (iii) its relative independence enables <strong>the</strong>se<br />
forms to become almost self-sufficient.<br />
The combination of all of <strong>the</strong>se factors can potentially help us to explain <strong>the</strong> origin of<br />
cumulative cultural development. Indeed, toge<strong>the</strong>r <strong>the</strong>y appear to indicate <strong>the</strong> historic<br />
point when social processes can for <strong>the</strong> first time achieve conditions comparable to that<br />
of <strong>the</strong> biological autonomy of <strong>life</strong>, namely a kind of „needful freedom‟ (Jonas 1966) in<br />
relation to <strong>the</strong>ir constituent components. Accordingly, here we could have <strong>the</strong> birth of a<br />
cultural form of autonomy, which from <strong>the</strong> perspective of its component individuals<br />
would be encountered as a form of heteronomy (cf. Steiner & Stewart 2009). Since<br />
<strong>the</strong>se social processes are largely freed from <strong>the</strong> material <strong>and</strong> energetic necessities of<br />
realizing biological autonomy, <strong>the</strong>ir autonomous development could be less constrained<br />
than that of living systems. Indeed, this possibility is supported by empirical evidence in<br />
evolutionary anthropology which shows that human cumulative cultural evolution is a<br />
process that begins around 0.3 million years ago, at which point its development<br />
230 | P a g e
ecomes “increasingly autocatalytic” (Ambrose 2001, p. 1752). Of course, it remains to<br />
be seen to what extent <strong>the</strong> ra<strong>the</strong>r conservative concepts of <strong>the</strong> enactive paradigm, e.g.<br />
autonomy as maintenance of identity <strong>and</strong> adaptivity as compensation of perturbation,<br />
are adequate for dealing with such a process of developmental becoming. What is it that<br />
drives autonomous systems to be en-active in this radical way?<br />
231 | P a g e
14 Conclusion<br />
If enactive cognitive science wants its basic notions of autonomous agency <strong>and</strong> sensemaking<br />
to be <strong>the</strong> foundation for a general <strong>the</strong>ory of <strong>mind</strong> <strong>and</strong> cognition, <strong>the</strong>n it needs to<br />
appeal to a strong version of <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis. Only if <strong>the</strong> <strong>continuity</strong> <strong>the</strong>sis<br />
is in fact a valid working hypo<strong>the</strong>sis can this bottom-up approach to cognitive science<br />
systematically reject <strong>the</strong> criticism that it is dealing with interesting, but ultimately<br />
irrelevant, descriptions of biological phenomena. So far, however, proponents of <strong>the</strong><br />
enactive paradigm have been plagued by what we have called <strong>the</strong> cognitive gap, i.e. an<br />
inability to conceive of how <strong>the</strong> principles applicable to simple organisms can be used<br />
systematically to explain <strong>the</strong> highest reaches of human cognition. It has been argued that<br />
this impasse is largely due to <strong>the</strong> methodological individualism that is still prevalent in<br />
cognitive science, <strong>and</strong> that a proper consideration of <strong>the</strong> constitutive role of sociality for<br />
agency <strong>and</strong> sense-making is needed in order to make <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis a<br />
viable approach. We <strong>the</strong>refore have addressed <strong>the</strong> cognitive gap from a <strong>the</strong>oretical,<br />
experimental <strong>and</strong> phenomenological perspective.<br />
In terms of <strong>the</strong>ory, <strong>the</strong> enactive approach to social cognition is developed in a novel<br />
direction by highlighting <strong>the</strong> specific manner in which <strong>the</strong> dynamics of <strong>the</strong> interaction<br />
process opens up new behavioral domains. In particular, we develop novel definitions of<br />
multi-agent systems <strong>and</strong> social interaction, <strong>the</strong> latter of which emphasizes <strong>the</strong> essential<br />
co-regulation of social acts. This <strong>the</strong>oretical background provides <strong>the</strong> motivation for<br />
using an evolutionary robotics methodology to syn<strong>the</strong>size a set of novel minimalist<br />
simulation models. These are based on actual experiments in social psychology, so as to<br />
promote a mutually informative dialogue between <strong>the</strong> two disciplines. A detailed<br />
dynamical analysis of <strong>the</strong>se models supports <strong>the</strong> enactive approach; <strong>the</strong> behavior of <strong>the</strong><br />
agents in a multi-agent system is not an individual achievement alone but ra<strong>the</strong>r codetermined<br />
by <strong>the</strong>ir mutual interaction <strong>and</strong> organized effectively by this interaction<br />
process. It is demonstrated that when this interaction process is co-regulated as part of a<br />
social context (ra<strong>the</strong>r than just a multi-agent system), <strong>the</strong> extension to individual<br />
behavioral capacities is even fur<strong>the</strong>r increased.<br />
232 | P a g e
These results are complemented by a phenomenological investigation which has<br />
revealed ano<strong>the</strong>r common assumption in mainstream cognitive science that we have<br />
called methodological physicalism, i.e. <strong>the</strong> idea that <strong>the</strong> function of perception is to<br />
furnish <strong>the</strong> perceiver with information about <strong>the</strong> abstract physical properties of <strong>the</strong><br />
world. It is argued that this assumption prevents a proper appreciation of sociality<br />
because it mistakenly focuses scientific efforts on <strong>the</strong> „problem of o<strong>the</strong>r <strong>mind</strong>s‟: how<br />
social underst<strong>and</strong>ing is possible on <strong>the</strong> basis of physical facts that are devoid of any<br />
significance. A critical analysis of phenomenological observations, on <strong>the</strong> o<strong>the</strong>r h<strong>and</strong>,<br />
indicates that <strong>the</strong> detached perceptual attitude that is characteristic of adult human<br />
perception is essentially an intersubjective <strong>and</strong> socially mediated ability. Finally, <strong>the</strong><br />
systemic <strong>and</strong> phenomenological insights are combined to provide a novel perspective on<br />
<strong>the</strong> origins of cumulative cultural development. This perspective suggests a more<br />
coherent interpretation of <strong>the</strong> available empirical data. It is concluded that <strong>the</strong> <strong>life</strong>-<strong>mind</strong><br />
<strong>continuity</strong> <strong>the</strong>sis is a viable working hypo<strong>the</strong>sis even when accounting for specifically<br />
human abilities, <strong>and</strong> that an appreciation of <strong>the</strong> constitutive role of sociality for <strong>life</strong> <strong>and</strong><br />
<strong>mind</strong> confirms it to be a serious contender for a unified <strong>the</strong>ory of cognitive science.<br />
This <strong>the</strong>sis has demonstrated that <strong>the</strong> enactive paradigm can enable us to underst<strong>and</strong><br />
agency, sociality, <strong>and</strong> culture in a novel manner by drawing on <strong>the</strong>oretical, experimental<br />
<strong>and</strong> phenomenological methods. On this basis it has been possible to dissolve some<br />
outst<strong>and</strong>ing problems faced by more traditional approaches, as well as to formulate new<br />
hypo<strong>the</strong>ses that are open to validation by future empirical experiments. It has been<br />
acknowledged that <strong>the</strong> novel interpretations that have been suggested for evidence in<br />
social psychology, infant studies, primatology, <strong>and</strong> evolutionary anthropology still need<br />
to be worked out in more detail, but <strong>the</strong> outlines of a possible working hypo<strong>the</strong>sis have<br />
been indicated. Indeed, in many ways this <strong>the</strong>sis has only outlined <strong>the</strong> beginnings of<br />
what could become an important research program in its own right, by suggesting how<br />
<strong>the</strong> enactive approach is able to integrate scientific <strong>and</strong> phenomenological, artificial <strong>and</strong><br />
empirical, intellectual <strong>and</strong> practical traditions into one coherent framework.<br />
233 | P a g e
15 References<br />
Ambrose, S. H. (2001), “Paleolithic Technology <strong>and</strong> Human Evolution”, Science, 291, pp. 1748-1753<br />
Anderson, M. L. (2003), “Embodied Cognition: A field guide”, Artificial Intelligence, 149(1), pp. 91-130<br />
Auvray, M., Hanneton, S., Lenay, C. & O‟Regan, K. (2005), “There is something out <strong>the</strong>re: Distal<br />
attribution in sensory substitution, twenty years later”, Journal of Integrative Neuroscience, 4(4), pp.<br />
505-521<br />
Auvray, M., Lenay, C. & Stewart, J. (2009), “Perceptual interactions in a minimalist virtual<br />
environment”, New Ideas in Psychology, 27(1), pp. 32-47<br />
Auvray, M., Philipona, D., O‟Regan, J. K. & Spence, C. (2007), “The perception of space <strong>and</strong> form<br />
recognition in a simulated environment: The case of minimalist sensory-substitution devices”,<br />
Perception, 36, pp. 1736-1751<br />
Baldwin, J. M. (1897), “Organic Selection”, Nature, 55, p. 558<br />
Bar<strong>and</strong>iaran, X., Di Paolo, E. A. & Rohde, M. (2009), “Defining Agency: Individuality, Normativity,<br />
Asymmetry, <strong>and</strong> Spatio-temporality in Action”, Adaptive Behavior, 17(5), pp. 367-386<br />
Bar<strong>and</strong>iaran, X. & Moreno, A. (2006), “On what makes certain dynamical systems cognitive: A<br />
minimally cognitive organization program”, Adaptive Behavior, 14(2), pp. 171-185<br />
Bar<strong>and</strong>iaran, X. & Moreno, A. (2008), “Adaptivity: From Metabolism to Behavior”, Adaptive Behavior,<br />
16(5), pp. 325-344<br />
Beer, R. D. (1995a), “A dynamical systems perspective on agent-environment interaction”, Artificial<br />
Intelligence, 72(1-2), pp. 173-215<br />
Beer, R. D. (1995b), “On <strong>the</strong> dynamics of small continuous-time recurrent neural networks”, Adaptive<br />
Behavior, 3(4), pp. 471-511<br />
Beer, R. D. (1996), “Toward <strong>the</strong> evolution of dynamical neural networks for minimally cognitive<br />
behavior”, in: P. Maes, M. J. Mataric, J.-A. Arcady, J. Pollack & S. W. Wilson (eds.), From Animals<br />
to Animats 4: Proc. of <strong>the</strong> 4 th Int. Conf. on Simulation of Adaptive Behavior, Cambridge, MA: The<br />
MIT Press, pp. 421-429<br />
Beer, R. D. (1997), “The dynamics of adaptive behavior: A research program”, Robotics <strong>and</strong> Autonomous<br />
Systems, 20(2-4), pp. 257-289<br />
Beer, R. D. (2000), “Dynamical approaches to cognitive science”, Trends in Cognitive Sciences, 4(3), pp.<br />
91-99<br />
Beer, R. D. (2003), “The dynamics of active categorical perception in an evolved model agent”, Adaptive<br />
Behavior, 11(4), pp. 209-243<br />
Beer, R. D. (2004), “Autopoiesis <strong>and</strong> Cognition in <strong>the</strong> Game of Life”, Artificial Life, 10(3), pp. 309-326<br />
234 | P a g e
Bitbol, M. (2002), “Science as if situation mattered”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 1(2),<br />
pp. 181-224<br />
Bitbol, M. & Luisi, P. L. (2004), “Autopoiesis with or without cognition: defining <strong>life</strong> at its edge”,<br />
Journal of <strong>the</strong> Royal Society Interface, 1(1), pp. 99-107<br />
Boden, M. A. (1996) (ed.), The Philosophy of Artificial Life, New York, NY: Oxford University Press<br />
Boden, M. A. (2006a), Mind as Machine: A History of Cognitive Science, New York, NY: Oxford<br />
University Press<br />
Boden, M. A. (2006b), “Of Isl<strong>and</strong>s <strong>and</strong> Interactions”, Journal of Consciousness Studies, 13(5), pp. 53-63<br />
Bourgine, P., & Stewart, J. (2004), “Autopoiesis <strong>and</strong> Cognition”, Artificial Life, 10(3), pp. 327-345<br />
Bourgine, P. & Varela, F. J. (1992), “Introduction: Towards a Practice of Autonomous Systems”, in: F. J.<br />
Varela & P. Bourgine (eds.), Towards a Practice of Autonomous Systems: Proc. of <strong>the</strong> 1 st Euro. Conf.<br />
on Artificial Life, Cambridge, MA: The MIT Press, pp. 1 - 3<br />
Brooks, R. A. (1991a), “Intelligence without representation”, Artificial Intelligence, 47(1-3), pp. 139-160<br />
Brooks, R. A. (1991b), “New Approaches to Robotics”, Science, 253, pp. 1227-1232<br />
Brooks, R. A. (1997), “From earwigs to humans”, Robotics <strong>and</strong> Autonomous Systems, 20(2-4), pp. 291-<br />
304<br />
Brooks, R. A. (2001), “The relationship between matter <strong>and</strong> <strong>life</strong>”, Nature, 409(6818), pp. 409-411<br />
Buchanan, B. (2007), “The Time of <strong>the</strong> Animal”, PhaenEx, 2(2), pp. 61-80<br />
Buckley, C. L., Fine, P., Bullock, S. & Di Paolo, E. A. (2008), “Monostable controllers for adaptive<br />
behavior”, in: M. Asada, J. C. T. Hallam, J.-A. Meyer & J. Tani (eds.), From Animals to Animats 10:<br />
Proc. of <strong>the</strong> 10 th Int. Conf. on Simulation of Adaptive Behavior, Berlin, Germany: Springer-Verlag,<br />
pp. 103-112<br />
Buttelmann, D., Carpenter, M., Call, J. & <strong>Tom</strong>asello, M. (2007), “Encultured chimpanzees imitate<br />
rationally”, Developmental Science, 10(4), pp. F31-F38<br />
Buzsáki, G. (2006), Rhythms of <strong>the</strong> Brain, New York, NY: Oxford University Press<br />
Byrne, R. W., Barnard, P. J., Davidson, I., Janik, V. M., McGrew, W. C., Miklósi, Á. & Wiessner, P.<br />
(2004), “Underst<strong>and</strong>ing culture across species”, Trends in Cognitive Sciences, 8(8), pp. 341-346<br />
Call, J., Carpenter, M. & <strong>Tom</strong>asello, M. (2005), “Copying results <strong>and</strong> copying actions in <strong>the</strong> process of<br />
social learning: chimpanzees (Pan troglodytes) <strong>and</strong> human children (Homo sapiens)”, Animal<br />
Cognition, 8, pp. 151-163<br />
Call, J. & <strong>Tom</strong>asello, M. (2008), “Does <strong>the</strong> chimpanzee have a <strong>the</strong>ory of <strong>mind</strong>? 30 years later”, Trends in<br />
Cognitive Sciences, 12(5), pp. 187-192<br />
235 | P a g e
Carru<strong>the</strong>rs, P. (1996), "Simulation <strong>and</strong> self-knowledge: A defense of <strong>the</strong>ory-<strong>the</strong>ory", in: P. Carru<strong>the</strong>rs &<br />
P. K. Smith (eds.), Theories of Theories of Mind, Cambridge, UK: Cambridge University Press, pp.<br />
22-38<br />
Catmur, C., Walsh, V. & Heyes, C. M. (2007), “Sensorimotor learning configures <strong>the</strong> human mirror<br />
system”, Current Biology, 17, pp. 1527-1531<br />
Chalmers, D. J. (1996), The Conscious Mind: In Search of a Fundamental Theory, New York, NY:<br />
Oxford University Press<br />
Chiel, H. J. & Beer, R. D. (1997), “The brain has a body: Adaptive behaviour emerges from interactions<br />
of nervous system, body <strong>and</strong> environment”, Trends in Neurosciences, 20(12), pp. 553-557<br />
Clark, A. (1997), Being There: Putting brain, body, <strong>and</strong> world toge<strong>the</strong>r again, Cambridge, MA: The MIT<br />
Press<br />
Clark, A. (2001), Mindware: An Introduction to <strong>the</strong> Philosophy of Cognitive Science, Oxford, UK:<br />
Oxford University Press<br />
Clark, A. (2003), Natural-Born Cyborgs: Minds, Technologies, <strong>and</strong> <strong>the</strong> Future of Human Intelligence,<br />
Oxford, UK: Oxford University Press<br />
Clark, A. (2005), “Beyond <strong>the</strong> Flesh: Some Lessons from a Mole Cricket”, Artificial Life, 11(1-2), pp.<br />
233-244<br />
Clark, A. (2006), “That lonesome whistle: a puzzle for <strong>the</strong> sensorimotor model of perceptual experience”,<br />
Analysis, 66(289), pp. 22-25<br />
Clark, A. (2008), Supersizing <strong>the</strong> Mind: Embodiment, Action, <strong>and</strong> Cognitive Extension, New York, NY:<br />
Oxford University Press<br />
Cliff, D. (1991), “Computational Neuroethology: A Provisional Manifesto”, in: J.-A. Meyer & S. W.<br />
Wilson (eds.), From Animals to Animats: Proc. of <strong>the</strong> 1 st Int. Conf. on Simulation of Adaptive<br />
Behavior, Cambridge, MA: The MIT Press, pp. 29-39<br />
Cliff, D., Harvey, I. & Husb<strong>and</strong>s, P. (1993), “Explorations in Evolutionary Robotics”, Adaptive Behavior,<br />
2(1), pp. 73-110<br />
Cole, J. (1995), Pride <strong>and</strong> a Daily Marathon, Cambridge, MA: The MIT Press<br />
Cole, J. (2009), “Impaired embodiment <strong>and</strong> intersubjectivity”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive<br />
Sciences, 8(3), pp. 343-360<br />
Cole, J., Gallagher, S. & McNeill, D. (2002a), “Gesture following deafferentation: A phenomenologically<br />
informed experimental study”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 1(1), pp. 49-67<br />
Cole, J., Gallagher, S. & McNeill, D. (2002b), “Social cognition <strong>and</strong> primacy of movement revisited”,<br />
Trends in Cognitive Sciences, 6(4), pp. 155-156<br />
236 | P a g e
Colombetti, G. (in press), “Enaction, sense-making <strong>and</strong> emotion”, in: J. Stewart, O. Gapenne & E. A. Di<br />
Paolo (eds.), Enaction: Towards a New Paradigm for Cognitive Science, Cambridge, MA: The MIT<br />
Press<br />
Dennett, D. C. (1978), “Why not <strong>the</strong> whole iguana?”, Behavioral <strong>and</strong> Brain Sciences, 1, pp. 103-104<br />
Dennett, D. C. (1984), “Cognitive Wheels: The Frame Problem of AI”, in: C. Hookway (ed.), Minds,<br />
Machines, <strong>and</strong> Evolution: Philosophical Studies, Cambridge, UK: Cambridge University Press, pp.<br />
129-151<br />
Dennett, D. C. (1994), “Artificial Life as Philosophy”, Artificial Life, 1(3), pp. 291-292<br />
Depraz, N., Varela, F. J. & Vermersch, P. (2003), On Becoming Aware: A pragmatics of experiencing,<br />
The Ne<strong>the</strong>rl<strong>and</strong>s, Amsterdam: John Benjamins Publishing<br />
De Jaegher, H. (2006), Social Interaction Rhythm <strong>and</strong> Participatory Sense-Making: An Embodied,<br />
Interactional Approach to Social Underst<strong>and</strong>ing, with Some Implication for Autism, unpublished<br />
D.Phil. <strong>the</strong>sis, Brighton, UK: University of Sussex<br />
De Jaegher, H. (2009), “Social underst<strong>and</strong>ing though direct perception? Yes, by interacting”,<br />
Consciousness <strong>and</strong> Cognition, 18, pp. 535-542<br />
De Jaegher, H. & Di Paolo, E. A. (2007), “Participatory sense-making: An enactive approach to social<br />
cognition”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 6(4), pp. 485-507<br />
De Jaegher, H. & Di Paolo, E. A. (2008), "Making Sense in Participation: An Enactive Approach to<br />
Social Cognition", in: F. Morganti, A. Carassa & G. Riva (eds.), Enacting Intersubjectivity: A<br />
Cognitive <strong>and</strong> Social Perspective on <strong>the</strong> Study of Interactions, Amsterdam, Ne<strong>the</strong>rl<strong>and</strong>s: IOS Press,<br />
pp. 33-47<br />
De Jaegher, H. & <strong>Froese</strong>, T. (2009), “On <strong>the</strong> Role of Social Interaction in Individual Agency”, Adaptive<br />
Behavior, 17(5), pp. 444-460<br />
Di Paolo, E. A. (1999), On <strong>the</strong> Evolutionary <strong>and</strong> Behavioral Dynamics of Social Coordination: Models<br />
<strong>and</strong> Theoretical Aspects, unpublished D.Phil. <strong>the</strong>sis, Brighton, UK: University of Sussex<br />
Di Paolo, E. A. (2000), “Behavioral coordination, structural congruence <strong>and</strong> entrainment in a simulation<br />
of acoustically coupled agents”, Adaptive Behavior, 8(1), pp. 25-46<br />
Di Paolo, E. A. (2003), “Organismically-inspired robotics: homeostatic adaptation <strong>and</strong> teleology beyond<br />
<strong>the</strong> closed sensorimotor loop”, in: K. Murase & T. Asakura (eds.), Dynamical Systems Approach to<br />
Embodiment <strong>and</strong> <strong>Sociality</strong>, Adelaide, Australia: Advanced Knowledge International, pp. 19-42<br />
Di Paolo, E. A. (2005), “Autopoiesis, adaptivity, teleology, agency”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive<br />
Sciences, 4(4), pp. 429-452<br />
Di Paolo, E. A. (2008), “A Mind of Many”, Constructivist Foundations, 3(2), pp. 89-91<br />
Di Paolo, E. A. (2009), “Extended Life”, Topoi, 28(1), pp. 9-21<br />
237 | P a g e
Di Paolo, E. A. (in press), “Overcoming autopoiesis: An enactive detour on <strong>the</strong> way from <strong>life</strong> to society”,<br />
in: R. Magalhaes & R. Sanchez (eds.), Autopoiesis in Organizations <strong>and</strong> Information Systems,<br />
Elsevier<br />
Di Paolo, E. A. & Harvey, I. (2003), “Decisions <strong>and</strong> Noise: The Scope of Evolutionary Syn<strong>the</strong>sis <strong>and</strong><br />
Dynamical Analysis”, Adaptive Behavior, 11(4), pp. 284-288<br />
Di Paolo, E. A. & Iizuka, H. (2008), “How (not) to model autonomous behavior”, BioSystems, 91(2), pp.<br />
409-423<br />
Di Paolo, E. A., Noble, J. & Bullock, S. (2000), “Simulation Models as Opaque Thought Experiments”,<br />
in: M. A. Bedau, J. S. McCaskill, N. H. Packard & S. Rasmussen (eds.), Artificial Life VII: Proc. of<br />
<strong>the</strong> 7 th Int. Conf. on Artificial Life, Cambridge, MA: The MIT Press, pp. 497-506<br />
Di Paolo, E. A., Rohde, M. & De Jaegher, H. (in press), “Horizons for <strong>the</strong> Enactive Mind: Values, Social<br />
Interaction, <strong>and</strong> Play”, in: J. Stewart, O. Gapenne & E. A. Di Paolo (eds.), Enaction: Towards a New<br />
Paradigm for Cognitive Science, Cambridge, MA: The MIT Press<br />
Di Paolo, E. A., Rohde, M. & Iizuka, H. (2008), “Sensitivity to social contingency or stability of<br />
interaction? Modelling <strong>the</strong> dynamics of perceptual crossing”, New Ideas in Psychology, 26(2), pp.<br />
278-294<br />
<strong>Dr</strong>eyfus, H. L. (1972), What Computers Can’t Do: A Critique of Artificial Reason, New York, NY:<br />
Harper <strong>and</strong> Row<br />
<strong>Dr</strong>eyfus, H. L. (1981), “From Micro-Worlds to Knowledge Representation: AI at an Impasse”, in: J.<br />
Haugel<strong>and</strong> (ed.), Mind Design: Philosophy, Psychology, Artificial Intelligence, Cambridge, MA: The<br />
MIT Press, pp. 161-204<br />
<strong>Dr</strong>eyfus, H. L. (1991), Being-in-<strong>the</strong>-World: A Commentary on Heidegger’s Being <strong>and</strong> Time, Division 1,<br />
Cambridge, MA: The MIT Press<br />
<strong>Dr</strong>eyfus, H. L. (2007), “Why Heideggerian AI failed <strong>and</strong> how fixing it would require making it more<br />
Heideggerian”, Philosophical Psychology, 20(2), pp. 247-268<br />
<strong>Dr</strong>eyfus, H. L. & <strong>Dr</strong>eyfus, S. E. (1988), “Making a <strong>mind</strong> versus modelling <strong>the</strong> brain: artificial intelligence<br />
back at a branch-point”, Daedalus¸ 117(1), p. 15-44<br />
Dunbar, R. I. M. (1998), “The Social Brain Hypo<strong>the</strong>sis”, Evolutionary Anthropology, 6, pp. 178-190<br />
Dupuy, J.-P. (2009), On <strong>the</strong> Origins of Cognitive Science: The Mechanization of Mind, Cambridge, MA:<br />
The MIT Press<br />
Egbert, M. D. & Di Paolo, E. A. (2009), “Integrating Autopoiesis <strong>and</strong> Behavior: An Exploration in<br />
Computational Chemo-ethology”, Adaptive Behavior, 17(5), pp. 387-401<br />
Eldredge, N. & Gould, S. J. (1972), “Punctuated equilibria: An alternative to phyletic gradualism”, in: T.<br />
J. M. Schopf (ed.), Models in Paleobiology, San Francisco, CA: Freeman, Cooper & Co.<br />
238 | P a g e
Fitch, W. T. (2000), “The evolution of speech: a comparative review”, Trends in Cognitive Sciences, 4(7),<br />
pp. 258-267<br />
Fodor, J. A. (1975), The Language of Thought, Cambridge, MA: Harvard University Press<br />
Fogel, A. (1993), Developing through relationships: Origins of communication, self <strong>and</strong> culture, London,<br />
UK: Harvester Wheatsheaf<br />
<strong>Froese</strong>, T. (2007), “On <strong>the</strong> role of AI in <strong>the</strong> ongoing paradigm shift within <strong>the</strong> cognitive sciences”, in: M.<br />
Lungarella, F. Iida, J. Bongard, & R. Pfeifer (eds.), 50 Years of Artificial Intelligence: Essays<br />
Dedicated to <strong>the</strong> 50 th Anniversary of Artificial Intelligence, Berlin, Germany: Springer, pp. 63-75<br />
<strong>Froese</strong>, T. (2009), “Hume <strong>and</strong> <strong>the</strong> enactive approach to <strong>mind</strong>”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive<br />
Sciences, 8(1), pp. 95-133<br />
<strong>Froese</strong>, T. & Di Paolo, E. A. (2008a), “Stability of coordination requires mutuality of interaction in a<br />
model of embodied agents”, in: M. Asada, J. C. T. Hallam, J.-A. Meyer & J. Tani (eds.), From<br />
Animals to Animats 10: Proc. of <strong>the</strong> 10 th Int. Conf. on Simulation of Adaptive Behavior, Berlin,<br />
Germany: Springer-Verlag, pp. 52-61<br />
<strong>Froese</strong>, T. & Di Paolo, E. A. (2008b), “Can evolutionary robotics generate simulation models of<br />
autopoiesis?”, Cognitive Science Research Paper (CSRP), 598, Brighton, UK: University of Sussex<br />
<strong>Froese</strong>, T. & Di Paolo, E. A. (2009), “<strong>Sociality</strong> <strong>and</strong> <strong>the</strong> <strong>life</strong>-<strong>mind</strong> <strong>continuity</strong> <strong>the</strong>sis”, Phenomenology <strong>and</strong><br />
<strong>the</strong> Cognitive Sciences, 8(4), pp. 439-463<br />
<strong>Froese</strong>, T. & Di Paolo, E. A. (in press-a), “Modeling social interaction as perceptual crossing: An<br />
investigation into <strong>the</strong> dynamics of <strong>the</strong> interaction process”, Connection Science<br />
<strong>Froese</strong>, T. & Di Paolo, E. A. (in press-b), “Toward Minimally Social Behavior: Social Psychology Meets<br />
Evolutionary Robotics”, Advances in Artificial Life: Proc. of <strong>the</strong> 10 th Euro. Conf. on Artificial Life,<br />
Berlin, Germany: Springer-Verlag<br />
<strong>Froese</strong>, T. & Spier, E. (2008), “Convergence <strong>and</strong> Crossover: The Permutation Problem Revisited”,<br />
Cognitive Science Research Paper (CSRP), 596, Brighton, UK: University of Sussex<br />
<strong>Froese</strong>, T. & Spiers, A. (2007), “Toward a Phenomenological Pragmatics of Enactive Perception”, in:<br />
Enactive/07: Proc. of <strong>the</strong> 4 th Int. Conf. on Enactive Interfaces, Grenoble, France: Association<br />
ACROE, pp. 105-108<br />
<strong>Froese</strong>, T., Virgo, N. & Izquierdo, E. (2007), “Autonomy: a review <strong>and</strong> a reappraisal”, in: F. Almeida e<br />
Costa, L. M. Rocha, E. Costa, I. Harvey & A. Coutinho (eds.), Advances in Artificial Life: Proc. of<br />
<strong>the</strong> 9 th Euro. Conf. on Artificial Life, Berlin, Germany: Springer-Verlag, pp. 455-464<br />
<strong>Froese</strong>, T. & Ziemke, T. (2009), “Enactive Artificial Intelligence: Investigating <strong>the</strong> systemic organization<br />
of <strong>life</strong> <strong>and</strong> <strong>mind</strong>”, Artificial Intelligence, 173(3-4), pp. 366-500<br />
Funahashi, K. & Nakamura, Y. (1993), “Approximation of dynamical systems by continuous time<br />
recurrent neural networks”, Neural Networks, 6(6), pp. 801-806<br />
239 | P a g e
Gallagher, S. (1997), “Mutual enlightenment: Recent phenomenology in cognitive science”, Journal of<br />
Consciousness Studies, 4(3), pp. 195-214<br />
Gallagher, S. (2001), "The Practice of Mind: Theory, Simulation or Primary Interaction?", Journal of<br />
Consciousness Studies, 8(5-7), pp. 83-108<br />
Gallagher, S. (2005), How <strong>the</strong> Body Shapes <strong>the</strong> Mind, New York, NY: Oxford University Press<br />
Gallagher, S. (2007), “Simulation trouble”, Social Neuroscience, 2(3-4), pp. 353-365<br />
Gallagher, S. (2008a), “Intersubjectivity in perception”, Continental Philosophy Review, 41(2), pp. 163-<br />
178<br />
Gallagher, S. (2008b), “Direct perception in <strong>the</strong> intersubjective context”, Consciousness <strong>and</strong> Cognition,<br />
17(2), pp. 535-543<br />
Gallagher, S. (2008c), Brainstorming: View <strong>and</strong> Interviews on <strong>the</strong> Mind, Exeter, UK: Imprint Academic<br />
Gallagher, S. (2008d), "Inference or interaction: social cognition without precursors", Philosophical<br />
Explorations, 11(3), pp. 163-174<br />
Gallagher, S. & Cole, J. (1995), “Body schema <strong>and</strong> body image in a deafferented subject”, Journal of<br />
Mind <strong>and</strong> Behavior, 16, pp. 369-390<br />
Gallagher, S. & Marcel, A. J. (1999), “The self in contextualized action”, Journal of Consciousness<br />
Studies, 6(6), pp. 4-30<br />
Gallagher, S. & Meltzoff, A. N. (1996), “The earliest sense of self <strong>and</strong> o<strong>the</strong>rs: Merleau-Ponty <strong>and</strong> recent<br />
developmental studies”, Philosophical Psychology, 9, pp. 213-236<br />
Gallagher, S. & Zahavi, D. (2008), The Phenomenological Mind: An Introduction to Philosophy of Mind<br />
<strong>and</strong> Cognitive Science, London, UK: Routledge<br />
Gallese, V. & Goldman, A. (1998), "Mirror neurons <strong>and</strong> <strong>the</strong> simulation <strong>the</strong>ory of <strong>mind</strong> reading", Trends<br />
in Cognitive Science, 2(12), pp. 493-501<br />
Gergely, G., Bekkering, H. & Király, I. (2002), “Rational imitation in preverbal infants”, Nature, 415, p.<br />
755<br />
Gergely, G. & Watson, J. (1999), “Early social-emotional development: contingency perception <strong>and</strong> <strong>the</strong><br />
social biofeedback model”, in: P. Rochat (ed.), Early social cognition: Underst<strong>and</strong>ing o<strong>the</strong>rs in <strong>the</strong><br />
first months of <strong>life</strong>, Hillsdale, NJ: Lawrence Erlbaum, pp. 101-137<br />
Godfrey-Smith, P. (1996), “Spencer <strong>and</strong> Dewey on Life <strong>and</strong> Mind”, in: M. Boden (ed.), The Philosophy<br />
of Artificial Life, Oxford, UK: Oxford University Press, pp. 314-331<br />
Goldberg, D. E. (1989), Genetic Algorithms in Search, Optimization, <strong>and</strong> Machine Learning, Redwood<br />
City, CA: Addison-Wesley<br />
Hanna, R. & Thompson, E. (2003), “The Mind-Body-Body Problem”, Theoria et Historia Scientarum,<br />
7(1), pp. 24-44<br />
240 | P a g e
Harnad, S. (1990), “The symbol grounding problem”, Physica D, 42, pp. 335-346<br />
Harnish, R. M. (2002), Minds, Brains, Computers: An Historical Introduction to <strong>the</strong> Foundations of<br />
Cognitive Science, Malden, MA: Blackwell Publishers<br />
Harvey, I. (1996), “Untimed <strong>and</strong> misrepresented: connectionism <strong>and</strong> <strong>the</strong> computer metaphor”, Artificial<br />
Intelligence <strong>and</strong> Simulation of Behavior Quarterly, 96, pp. 20-27<br />
Harvey, I. (2000), “Robotics: Philosophy of Mind using a Screwdriver”, in: T. Gomi (ed.), Evolutionary<br />
Robotics. From Intelligent Robots to Artificial Life: Proc. of <strong>the</strong> 7 th Int. Symposium on Evolutionary<br />
Robotics, Ontario, Canada: AAI Books, pp. 207-230<br />
Harvey, I. (2001), “Artificial Evolution: A Continuing SAGA”, in: T. Gomi (ed.), Evolutionary Robotics.<br />
From Intelligent Robotics to Artificial Life: Proc. of <strong>the</strong> 8 th Int. Symposium on Evolutionary Robotics,<br />
Berlin, Germany: Springer-Verlag, pp. 94-109<br />
Harvey, I. (2004), “Homeostasis <strong>and</strong> Rein Control: From Daisyworld to Active Perception”, in: J.<br />
Pollack, M. Bedau, P. Husb<strong>and</strong>s, T. Ikegami, & R. A. Watson (eds.), Artificial Life IX: Proc. of <strong>the</strong><br />
9 th Int. Conf. on <strong>the</strong> Simulation <strong>and</strong> Syn<strong>the</strong>sis of Living Systems, Cambridge, MA: The MIT Press, pp.<br />
309-314<br />
Harvey, I., Di Paolo, E. A., Wood, R., Quinn, M. & Tuci, E. A. (2005), “Evolutionary Robotics: A new<br />
scientific tool for studying cognition”, Artificial Life, 11(1-2), pp. 79-98<br />
Haselager, W. F. G. (2005), “Robotics, philosophy <strong>and</strong> <strong>the</strong> problems of autonomy”, Pragmatics &<br />
Cognition, 13(3), pp. 515-532<br />
Haugel<strong>and</strong>, J. (1985), Artificial Intelligence: The Very Idea, Cambridge, MA: The MIT Press<br />
Haugel<strong>and</strong>, J. (1997), “What is <strong>mind</strong> design?”, in: J. Haugel<strong>and</strong> (ed.), Mind Design II: Philosophy,<br />
Psychology, Artificial Intelligence, Cambridge, MA: The MIT Press, pp. 1-28<br />
Hayes, J. (2007), “Heidegger‟s Fundamental Ontology <strong>and</strong> <strong>the</strong> Problem of Animal Life”, PhaenEx, 2(2),<br />
pp. 42-60<br />
Heath, J. (2009), “Methodological individualism”, in: E. N. Zalta (ed.), The Stanford Encyclopedia of<br />
Philosophy (Summer 2009 Ed.), http://plato.stanford.edu/archives/sum2009/entries/methodologicalindividualism/<br />
Heidegger, M. (1927), Sein und Zeit, trans. by: J. Macquarrie & E. Robinson, Being <strong>and</strong> Time, Oxford,<br />
UK: Blackwell Publishing Ltd., 1962<br />
Heidegger, M. (1929), Die Grundbegriffe der Metaphysik: Welt, Endlichkeit, Einsamkeit, trans. by: W.<br />
McNeill & N. Walker, The Fundamental Concepts of Metaphysics: World, Finitude, Solitude,<br />
Bloomington, IN: Indiana University Press, 1995<br />
Holl<strong>and</strong>, J. H. (1975), Adaptation in Natural <strong>and</strong> Artificial Systems, Ann Arbor, MI: University of<br />
Michigan Press<br />
241 | P a g e
Hurley, S. (2008), “The shared circuits model (SCM): How control, mirroring, <strong>and</strong> simulation can enable<br />
imitation, deliberation, <strong>and</strong> <strong>mind</strong>reading”, Behavioral <strong>and</strong> Brain Sciences, 31(1), pp. 1-22<br />
Husb<strong>and</strong>s, P., Holl<strong>and</strong>, O. & Wheeler, M. (eds.) (2008), The Mechanical Mind in History, Cambridge,<br />
MA: The MIT Press<br />
Husb<strong>and</strong>s, P., Smith, T., Jakobi, N. & O‟Shea, M. (1998), “Better Living through Chemistry: Evolving<br />
GasNets for Robot Control”, Connection Science, 10(3-4), pp. 185-210<br />
Husserliana (Hua)<br />
Husserliana I: Husserl, E. (1929), Cartesianische Meditationen und Pariser Vorträge, Den Haag,<br />
Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1950; trans. by: D. Cairns, Cartesian Meditations: An Introduction to<br />
Phenomenology, The Hague, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1960<br />
Husserliana VI: Husserl, E. (1936), Die Krisis der europäischen Wissenschaften und die transzendentale<br />
Phänomenologie. Eine Einleitung in die phänomenologische Philosophie, Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s:<br />
Martinus Nijhoff, 1976; trans. by: D. Carr, The Crisis of European Sciences <strong>and</strong> Transcendental<br />
Phenomenology. An Introduction to Phenomenology, Evanston, IL: Northwestern Uni. Press, 1970<br />
Husserliana IX: Husserl, E. (1925), Phänomenologische Psychologie. Vorlesungen Sommersemester<br />
1925, Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1962; trans. by: J. Scanlon, Phenomenological<br />
Psychology: Lectures, Summer Semester, 1925, The Hague, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1977<br />
Husserliana X: Husserl, E. (1893-1917), Zur Phänomenologie des inneren Zeitbewusstseins (1893-1917),<br />
Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1966; trans. by: J. B. Brough, On <strong>the</strong> Phenomenology of<br />
<strong>the</strong> Consciousness of Internal Time (1893-1917), Dordrecht, Ne<strong>the</strong>rl<strong>and</strong>s: Kluwer Academic, 1991<br />
Husserliana XIII: Husserl, E. (1905-1920), Zur Phänomenologie der Intersubjektivität. Texte aus dem<br />
Nachlass. Erster Teil: 1905-1920, Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1973<br />
Husserliana XIV: Husserl, E. (1921-1928), Zur Phänomenologie der Intersubjektivität. Texte aus dem<br />
Nachlass. Zweiter Teil: 1921-1928, Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1973<br />
Husserliana XV: Husserl, E. (1929-1935), Zur Phänomenologie der Intersubjektivität. Texte aus dem<br />
Nachlass. <strong>Dr</strong>itter Teil: 1929-1935, Den Haag, Ne<strong>the</strong>rl<strong>and</strong>s: Martinus Nijhoff, 1973<br />
Hutchins, E. (1995), Cognition in <strong>the</strong> wild, Cambridge, MA: The MIT Press<br />
Iida, F. & Pfeifer, R. (2004), “„Cheap‟ Rapid Locomotion of a Quadruped Robot: Self-Stabilization of<br />
Bounding Gait”, in: F. C. A. Groen, N. Amato, A. Bonarini, E. Yoshida & B. Kröse (eds.), Intelligent<br />
Autonomous Systems 8, Amsterdam, Ne<strong>the</strong>rl<strong>and</strong>s: IOS Press, Inc., pp. 642-650<br />
Iizuka, H. & Di Paolo, E. A. (2007a), “Toward Spinozist robotics: Exploring <strong>the</strong> minimal dynamics of<br />
behavioral preference”, Adaptive Behavior, 15(4), pp. 359-376<br />
Iizuka, H. & Di Paolo, E. A. (2007b), “Minimal Agency Detection of Embodied Agents”, in: F. Almeida<br />
e Costa, L. M. Rocha, E. Costa, I. Harvey & A. Coutinho (eds.), Advances in Artificial Life: Proc. of<br />
<strong>the</strong> 9 th Euro. Conf. on Artificial Life, Berlin, Germany: Springer-Verlag, pp. 485-494<br />
242 | P a g e
Iizuka, H. & Di Paolo, E. A. (2008), “Extended Homeostatic Adaptation: Improving <strong>the</strong> link between<br />
internal <strong>and</strong> behavioral stability”, in: M. Asada, J. C. T. Hallam, J.-A. Meyer & J. Tani (eds.), From<br />
Animals to Animats 10: Proc. of <strong>the</strong> 10 th Int. Conf. on Simulation of Adaptive Behavior, Berlin,<br />
Germany: Springer-Verlag, pp. 1-11<br />
Iizuka, H. & Ikegami, T. (2004a), “Adaptability <strong>and</strong> diversity in simulated turn-taking behavior”,<br />
Artificial Life, 10(4), pp. 361-378<br />
Iizuka, H. & Ikegami, T. (2004b), “Simulating autonomous coupling in discrimination of light<br />
frequencies”, Connection Science, 16(4), pp. 283-299<br />
Ikegami, T. & Iizuka, H. (2007), “Turn-taking interaction as a cooperative <strong>and</strong> co-creative process”,<br />
Infant Behavior & Development, 30(2), pp. 278-288<br />
Ikegami, T. & Suzuki, K. (2008), “From homeostatic to homeodynamic Self”, BioSystems, 91(2), pp.<br />
388-400<br />
Iverson, J. M. & Goldin-Meadow, S. (1998), “Why people gesture when <strong>the</strong>y speak”, Nature, 396, p. 228<br />
Izquierdo, E. & Buhrmann, T. (2008), “Analysis of a dynamical recurrent neural network evolved for two<br />
qualitatively different tasks: walking <strong>and</strong> chemotaxis”, in: S. Bullock, J. Noble, R. Watson, & M. A.<br />
Bedau (eds.), Artificial Life XI: Proceedings of <strong>the</strong> Eleventh International Conference on <strong>the</strong><br />
Simulation <strong>and</strong> Syn<strong>the</strong>sis of Living Systems, Cambridge, MA: The MIT Press, pp. 257-264<br />
Izquierdo-Torres, E. & Di Paolo, E. A. (2005), “Is an embodied system ever purely reactive?”, in: M.<br />
Capcarrere, A. A. Freitas, P. J. Bentley, C. G. Johnson & J. Timmis (eds.), Advances in Artificial<br />
Life: Proc. of <strong>the</strong> 8th Euro. Conf. on Artificial Life, Germany, Berlin: Springer-Verlag, pp. 252–261<br />
Izquierdo, E., Harvey, I. & Beer, R. (2008), “Associative Learning on a Continuum in Evolved<br />
Dynamical Neural Networks”, Adaptive Behavior, 16(6), pp. 361-384<br />
Jonas, H. (1966), The Phenomenon of Life: Toward a Philosophical Biology, Evanston, Illinois:<br />
Northwestern University Press, 2001<br />
Jonas, H. (1968), “Biological Foundations of Individuality”, International Philosophical Quarterly, 8, pp.<br />
231-251<br />
Jonas, H. (1992), “The Burden <strong>and</strong> Blessing of Mortality”, The Hastings Center Report, 22(1), pp. 34-40<br />
Kant, I. (1790), Kritik der Urteilskraft, trans. by: W. S. Pluhar, Critique of Judgment, Indianapolis, IN:<br />
Hacket Publishing Company, 1987<br />
Kelso, J. A. S. (1995), Dynamic Patterns: The Self-Organization of Brain <strong>and</strong> Behaviour, Cambridge,<br />
MA: The MIT Press<br />
Kirsh, D. (1991), “Today <strong>the</strong> earwig, tomorrow man?”, Artificial Intelligence, 47(1-3), pp. 161-184<br />
Kuhn, T. S. (1962), The Structure of Scientific Revolutions, Chicago, IL: University of Chicago Press<br />
243 | P a g e
LaFrance, M. (1982), “Posture mirroring <strong>and</strong> rapport”, in: M. Davis (ed.), Interaction Rhythms:<br />
Periodicity in communicative behavior, New York, NY: Human Sciences Press, pp. 279-298<br />
Langton, C. G. (1989), “Artificial Life”, in: C. G. Langton (ed.), Artificial Life: Proceedings of an<br />
Interdisciplinary Workshop on <strong>the</strong> Syn<strong>the</strong>sis <strong>and</strong> Simulation of Living Systems, Santa Fe Institute<br />
Studies in <strong>the</strong> Sciences of Complexity, vol. 4, Redwood City, CA: Addison-Wesley, pp. 1-47<br />
Langton, C. G. (1995) (ed.), Artificial Life: An Overview, Cambridge, MA: The MIT Press<br />
Legr<strong>and</strong>, D. (2006), “The bodily self: The sensori-motor roots of pre-reflexive self-consciousness”,<br />
Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 5(1), pp. 89-118<br />
Levinas, E. (1979), Le temps et l’autre, trans. by: R. A. Cohen, Time <strong>and</strong> <strong>the</strong> O<strong>the</strong>r, Pittsburgh, PA:<br />
Duquesne University Press, 1987<br />
Levine, J. (1983), “Materialism <strong>and</strong> Qualia: The Explanatory Gap”, Pacific Philosophical Quarterly, 64,<br />
pp. 354-361<br />
Lindblom, J. & Ziemke, T. (2003), “Social Situatedness of Natural <strong>and</strong> Artificial Intelligence: Vygotsky<br />
<strong>and</strong> Beyond”, Adaptive Behavior, 11(2), pp. 79-96<br />
Luhmann, N. (1984), Soziale Systeme: Grundriß einer allgemeinen Theorie, Frankfurt, Germany:<br />
Suhrkamp Verlag<br />
Luisi, P. L. (2003), “Autopoiesis: a review <strong>and</strong> reappraisal”, Naturwissenschaften, 90, pp. 49-59<br />
MacLennan, B. J. (1992), “Syn<strong>the</strong>tic Ethology: An Approach to <strong>the</strong> Study of Communication”, in: C. G.<br />
Langton, C. Taylor, D. Farmer & S. Rasmussen (eds.), Artificial Life II, Redwood City, CA:<br />
Addison-Wesley, pp. 631-658<br />
Marcel, A. J. (1992), “The personal level in cognitive rehabilitation”, in: N. von Steinbuchel, E. Poppel &<br />
D. von Cramon (eds.), Neuropsychological Rehabilitation, Berlin, Germany: Springer, pp. 155-168<br />
Maturana, H. R. (1978), “Biology of Language: The Epistemology of Reality”, in: G. Miller & E.<br />
Lenneberg (eds.), Psychology <strong>and</strong> Biology of Language <strong>and</strong> Thought: Essays in Honor of Eric<br />
Lenneberg, New York, NY: Academic Press, pp. 27-63<br />
Maturana, H. R. (1988), “Reality: The Search for Objectivity or <strong>the</strong> Quest for a Compelling Argument”,<br />
The Irish Journal of Psychology, 1(9), pp. 25-82<br />
Maturana, H. R. (2002), “Autopoiesis, Structural Coupling <strong>and</strong> Cognition: A history of <strong>the</strong>se <strong>and</strong> o<strong>the</strong>r<br />
notions in <strong>the</strong> biology of cognition”, Cybernetics & Human Knowing, 9(3-4), pp. 5-34<br />
Maturana, H. R., Mpodozis, J. & Letelier, J. C. (1995), “Brain, Language <strong>and</strong> <strong>the</strong> Origin of Human<br />
Mental Functions”, Biological Research, 28, pp. 15-26<br />
Maturana, H. R. & Varela, F. J. (1980), Autopoiesis <strong>and</strong> Cognition: The Realization of <strong>the</strong> Living,<br />
Dordrecht, Holl<strong>and</strong>: Kluwer Academic Publishers<br />
244 | P a g e
Maturana, H. R., & Varela, F. J. (1987), The Tree of Knowledge: The Biological Roots of Human<br />
Underst<strong>and</strong>ing, Boston, MA: Shambhala Publications<br />
McCarthy, J. & Hayes, P. J. (1969), “Some philosophical problems from <strong>the</strong> st<strong>and</strong>point of artificial<br />
intelligence”, in: B. Meltzer & D. Michie (eds.), Machine Intelligence 4, Edinburgh, UK: Edinburgh<br />
University Press, pp. 463-502<br />
McClell<strong>and</strong>, J. L., Rumelhart, D. E. & <strong>the</strong> PDP Research Group (1986), Parallel Distributed Processing:<br />
Explorations in <strong>the</strong> Microstructure of Cognition. Vol. 2: Psychological <strong>and</strong> Biological Models,<br />
Cambridge, MA: The MIT Press<br />
McGann, M. (2007), “Enactive <strong>the</strong>orists do it on purpose: Toward an enactive account of goals <strong>and</strong> goaldirectedness”,<br />
Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 6(4), pp. 463-483<br />
McMullin, B. (2004), “Thirty Years of Computational Autopoiesis: A Review”, Artificial Life, 10(3), pp.<br />
277-295<br />
Meltzoff, A. N. (1995), “Underst<strong>and</strong>ing <strong>the</strong> Intentions of O<strong>the</strong>rs: Re-Enactment of Intended Acts by 18-<br />
Month-Old Children”, Developmental Psychology, 31(5), pp. 838-850<br />
Meltzoff, A. N. & Borton, R. W. (1979), “Intermodal matching by human neonates”, Nature, 282, pp.<br />
403-404<br />
Meltzoff, A. N. & Moore, M. K. (1977), “Imitation of facial <strong>and</strong> manual gestures by human neonates”,<br />
Science, 198, pp. 75-78<br />
Meltzoff, A. N. & Moore, M. K. (1983), “Newborn infants imitate adult facial gestures”, Child<br />
Development, 54, pp. 702-709<br />
Meltzoff, A. N. & Moore, M. K. (1989), “Imitation in newborn infants: exploring <strong>the</strong> range of gestures<br />
imitated <strong>and</strong> <strong>the</strong> underlying mechanisms”, Developmental Psychology, 25, pp. 954-962<br />
Meltzoff, A. N. & Moore, M. K. (1997), “Explaining facial imitation: A <strong>the</strong>oretical model”, Early<br />
Development <strong>and</strong> Parenting, 6, pp. 179-192<br />
Merleau-Ponty, M. (1945), Phénomènologie de la perception, trans. by: C. Smith, Phenomenology of<br />
perception, New York, NY: Routledge & Kegan Paul, 1962<br />
Merleau-Ponty, M. (1960), “Les relations avec autrui chez l‟enfant”, Paris, France: Cours de Sorbonne,<br />
trans. by W. Cobb, “The Child‟s Relations with O<strong>the</strong>rs”, in: M. Merleau-Ponty (1964), The Primacy<br />
of Perception And O<strong>the</strong>r Essays on Phenomenological Psychology, <strong>the</strong> Philosophy of Art, History<br />
<strong>and</strong> Politics, J. M. Edie (ed.), Evanston, IL: Northwestern University Press, pp. 96-155<br />
Miller, G. (2005), “What is <strong>the</strong> biological basis of consciousness?”, Science, 309(5731), p. 79<br />
Millikan, R. G. (1989), “Biosemantics”, The Journal of Philosophy, 86(6), pp. 281-297<br />
Moreno, A. (2002), “Artificial Life <strong>and</strong> Philosophy”, Leonardo, 35(4), pp. 401-405<br />
245 | P a g e
Moreno, A. & Etxeberria, A. (2005), “Agency in Natural <strong>and</strong> Artificial Systems”, Artificial Life, 11(1-2),<br />
pp. 161-175<br />
Mossio, M. & Taraborelli, D. (2008), “Action-dependent perceptual invariants: From ecological to<br />
sensorimotor approaches”, Consciousness <strong>and</strong> Cognition, 17, pp. 1324-1340<br />
Murray, L. & Trevar<strong>the</strong>n, C. (1985), “Emotional regulations of interactions between two-month-olds <strong>and</strong><br />
<strong>the</strong>ir mo<strong>the</strong>rs”, in: T. M. Field & N. A. Fox (eds.), Social perception in infants, Norwood, NJ: Ablex<br />
Publishing, pp. 177-197<br />
Nadel, J., Carchon, I., Kervella, C., Marcelli, D. & Réserbat-Plantey, D. (1999), “Expectancies for social<br />
contingency in 2-month-olds”, Developmental Science, 2(2), pp. 164-173<br />
Nagel, T. (1974), “What is it like to be a bat?”, Philosophical Review, 83(4), pp. 435-450<br />
Nagel, E. (1977), “Teleology revisited: Goal-Directed Processes in Biology”, The Journal of Philosophy,<br />
74(5), pp. 261-279<br />
Newell, A. & Simon, H. A. (1976), “Computer Science as Empirical Enquiry: Symbols <strong>and</strong> Search”,<br />
Communications of <strong>the</strong> Association for Computing Machinery, 19(3), pp. 113-126<br />
Noë, A. (2004), Action in Perception, Cambridge, MA: The MIT Press<br />
Nolfi, S. & Floreano, D. (2000), Evolutionary Robotics: The biology, intelligence, <strong>and</strong> technology of selforganizing<br />
machines, Cambridge, MA: The MIT Press<br />
O‟Regan, J. K. & Noë, A. (2001), “A sensorimotor account of vision <strong>and</strong> visual consciousness”,<br />
Behavioral <strong>and</strong> Brain Sciences, 24(5), pp. 939-1031<br />
Oyama, S. (2009), “Friends, Neighbors, <strong>and</strong> Boundaries”, Ecological Psychology, 21(2), pp. 147-154<br />
Pascal, F. & O‟Regan, J. K. (2008), “Commentary on Mossio <strong>and</strong> Taraborelli: Is <strong>the</strong> enactive approach<br />
really sensorimotor?”, Consciousness <strong>and</strong> Cognition, 17(4), pp. 1341-1342<br />
Pentl<strong>and</strong>, A. (2007), “On <strong>the</strong> Collective Nature of Human Intelligence”, Adaptive Behavior, 15(2), pp.<br />
189-198<br />
Petitmengin, C. (2006), “Describing one‟s subjective experience in <strong>the</strong> second person: An interview<br />
method for <strong>the</strong> science of consciousness”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences¸ 5(3-4), pp.<br />
229-269<br />
Pfeifer, R. (1996), “Building „Fungus Eaters‟: Design Principles of Autonomous Agents”, in: P. Maes, M.<br />
J. Mataric, J.-A. Meyer, J. Pollack & S. W. Wilson (eds.), From Animals to Animats 4: Proc. of <strong>the</strong><br />
4 th Int. Conf. on Simulation of Adaptive Behavior, Cambridge, MA: The MIT Press, p. 3-12<br />
Pfeifer, R., Lungarella, M. & Iida, F. (2007), “Self-Organization, Embodiment, <strong>and</strong> Biologically Inspired<br />
Robotics”, Science, 318, pp. 1088-1093<br />
Pfeifer, R. & Scheier, C. (1999), Underst<strong>and</strong>ing Intelligence, Cambridge, MA: The MIT Press<br />
Piaget, J. (1962), Play, <strong>Dr</strong>eams, <strong>and</strong> Imitation in Childhood, New York, NY: Norton<br />
246 | P a g e
Piaget, J. (1967), Biologie et connaissance, Paris, France: Editions Gallimard; transl. as: Biology <strong>and</strong><br />
Knowledge: An Essay on <strong>the</strong> Relations between Organic Regulations <strong>and</strong> Cognitive Processes,<br />
Chicago, IL: The University of Chicago Press, 1971<br />
Povinelli, D. J. & Eddy, T. J. (1996), “What young chimpanzees know about seeing”, Monographs of <strong>the</strong><br />
Society for Research in Child Development, 61, pp. 1-152<br />
Prinz, J. (2006), “Putting <strong>the</strong> Brakes on Enactive Perception”, Psyche, 12(1), pp. 1-19<br />
Quinn, M. (2001), “Evolving communication without dedicated communication channels”, in: J. Kelemen<br />
& P. Sosik (eds.), Advances in Artificial Life: Proc. of <strong>the</strong> 6 th Euro. Conf. on Artificial Life, Berlin,<br />
Germany: Springer-Verlag, pp. 357-366<br />
Quinn, M., Smith, L., Mayley, G. & Husb<strong>and</strong>s, P. (2003), “Evolving controllers for a homogeneous<br />
system of physical robots: structured cooperation with minimal sensors”, Phil. Trans. R. Soc. Lond.<br />
A, 361, pp. 2321-2343<br />
Racine, T. P., Leavens, D. A., Susswein, N. & Wereha, T. J. (2008), “Conceptual <strong>and</strong> Methodological<br />
Issues in <strong>the</strong> Investigation of Primate Intersubjectivity”, in: F. Morganti, A. Carassa & G. Riva (eds.),<br />
Enacting Intersubjectivity: A Cognitive <strong>and</strong> Social Perspective on <strong>the</strong> Study of Interactions,<br />
Amsterdam, Ne<strong>the</strong>rl<strong>and</strong>s: IOS Press, pp. 65-79<br />
Rizzolatti, G., Fogassi, L. & Gallese, V. (2001), “Neurophysiological mechanisms underlying <strong>the</strong><br />
underst<strong>and</strong>ing <strong>and</strong> imitation of action”, Nature Reviews Neurscience, 2, pp. 661-670<br />
Rohde, M. (2008), Evolutionary Robotics Simulation Models in <strong>the</strong> Study of Human Behaviour <strong>and</strong><br />
Cognition, unpublished D.Phil. <strong>the</strong>sis, Brighton, UK: University of Sussex<br />
Rohde, M. & Di Paolo, E. A. (2007), “Adaptation to sensory delays: An evolutionary robotics model of<br />
an empirical study”, in: F. Almeida e Costa, L. M. Rocha, E. Costa, I. Harvey & A. Coutinho (eds.),<br />
Advances in Artificial Life: Proc. of <strong>the</strong> 9 th Euro. Conf. on Artificial Life, Berlin, Germany: Springer-<br />
Verlag, pp. 193-202<br />
Rohde, M. & Di Paolo, E. A. (2008), “Embodiment <strong>and</strong> Perceptual Crossing in 2D: A Comparative<br />
Evolutionary Robotics Study”, in: M. Asada, J. C. T. Hallam, J.-A. Meyer & J. Tani (eds.), From<br />
Animals to Animats 10: Proc. of <strong>the</strong> 10 th Int. Conf. on Simulation of Adaptive Behavior, Berlin,<br />
Germany: Springer-Verlag, pp. 83-92<br />
Roy, J.-M., Petitot, J., Pachoud, B. & Varela, F. J. (1999), “Beyond <strong>the</strong> Gap: An Introduction to<br />
Naturalizing Phenomenology”, in: J. Petitot, F. J. Varela, B. Pachoud & J.-M. Roy (eds.),<br />
Naturalizing Phenomenology: Issues in Contemporary Phenomenology <strong>and</strong> Cognitive Science,<br />
Stanford, CA: Stanford University Press, pp. 1-80<br />
Ruiz-Mirazo, K. & Moreno, A. (2004), “Basic Autonomy as a Fundamental Step in <strong>the</strong> Syn<strong>the</strong>sis of<br />
Life”, Artificial Life, 10(3), pp. 235-259<br />
Russel, J. (1996), Agency: Its role in mental development, Hove, UK: Taylor & Francis<br />
247 | P a g e
Russel, S. & Norvig, P. (2002), Artificial Intelligence: A Modern Approach, 2 nd ed., Prentice Hall<br />
Sartre, J.-P. (1943), L’Être et le néant, trans. by H. E. Barnes, Being <strong>and</strong> Nothingness: An essay on<br />
phenomenological ontology, Oxon, UK: Routledge, 2003<br />
Savage-Rumbaugh, S., Fields, W. M., Segerdahl, P. & Rumbaugh, D. M. (2005), “Culture prefigures<br />
cognition in Pan/Homo Bonobos”, Theoria: An International Journal for Theory, History <strong>and</strong><br />
Foundations of Science, 20(54), pp. 311-328<br />
Savage-Rumbaugh, S., Fields, W. M. & Taglialatela, J. P. (2001), “Language, Speech, Tools <strong>and</strong> Writing:<br />
A Cultural Imperative”, Journal of Consciousness Studies, 8(5-7), pp. 273-292<br />
Schwier, C., van Maanen, C., Carpenter, M. & <strong>Tom</strong>asello, M. (2006), “Rational imitation in 12-monthold<br />
infants”, Infancy, 10, pp. 303-311<br />
Shanon, B. (1998), “What is <strong>the</strong> function of consciousness?”, Journal of Consciousness Studies, 5(3), pp.<br />
295-308<br />
Spector, L. & Klein, J. (2006), “Trivial Geography in Genetic Programming”, in: T. Yu, R. Riolo, & B.<br />
Worzel (eds.), Genetic Programming Theory <strong>and</strong> Practice III, Springer US, pp. 109-124<br />
Steels, L. (1994), “The artificial <strong>life</strong> roots of artificial intelligence”, Artificial Life, 1(1-2), pp. 75-110<br />
Steiner, P. & Stewart, J. (2009), “From autonomy to heteronomy (<strong>and</strong> back): <strong>the</strong> Enaction of Social Life”,<br />
Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 8(4), pp. 527-550<br />
Stewart, J. (1992), “Life = Cognition: The epistemological <strong>and</strong> ontological significance of Artificial<br />
Life”, in: F. J. Varela & P. Bourgine (eds.), Toward a Practice of Autonomous Systems: Proc. of <strong>the</strong><br />
1 st Euro. Conf. on Artificial Life, Cambridge, MA: The MIT Press, pp. 475-483<br />
Stewart, J. (1996), “Cognition = <strong>life</strong>: Implications for higher-level cognition”, Behavioral Processes, 35,<br />
pp. 311-326<br />
Stewart, J. (2000), “From Autopoiesis to Semantic Closure”, Annals of <strong>the</strong> New York Academy of<br />
Sciences, 901, pp. 155-162<br />
Stewart, J. (in press), “Foundational issues in enaction as a paradigm for cognitive science: From <strong>the</strong><br />
origin of <strong>life</strong> to consciousness <strong>and</strong> writing”, in: J. Stewart, O. Gapenne & E. A. Di Paolo (eds.),<br />
Enaction: Towards a New Paradigm for Cognitive Science, Cambridge, MA: The MIT Press<br />
Thelen, E. & Smith, L. (1994), A Dynamic Systems Approach to <strong>the</strong> Development of Cognition <strong>and</strong><br />
Action, Cambridge, MA: The MIT Press<br />
Thompson, E. (2001), “Empathy <strong>and</strong> Consciousness”, Journal of Consciousness Studies, 8(5-7), pp. 1-32<br />
Thompson, E. (2004), “Life <strong>and</strong> <strong>mind</strong>: From autopoiesis to neurophenomenology. A tribute to Francisco<br />
Varela”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 3(4), pp. 381-398<br />
Thompson, E. (2005), “Sensorimotor subjectivity <strong>and</strong> <strong>the</strong> enactive approach to experience”,<br />
Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 4(4), pp. 407-427<br />
248 | P a g e
Thompson, E. (2007), Mind in Life: Biology, Phenomenology, <strong>and</strong> <strong>the</strong> Sciences of Mind, Cambridge, MA:<br />
The Belknap Press of Harvard University Press<br />
Thompson, E. & Stapleton, M. (2009), “Making Sense of Sense-Making: Reflections on Enactive <strong>and</strong><br />
Extended Mind Theories”, Topoi, 28(1), pp. 23-30<br />
Thompson, E. & Varela, F. J. (2001), “Radical embodiment: neural dynamics <strong>and</strong> consciousness”, Trends<br />
in Cognitive Sciences, 5(10), pp. 418-425<br />
<strong>Tom</strong>asello, M. (1988), “The role of joint attentional process in early language development”, Language<br />
Sciences, 10, pp. 69-88<br />
<strong>Tom</strong>asello, M. (1999), The cultural origins of human cognition, Cambridge, MA: Harvard University<br />
Press<br />
<strong>Tom</strong>asello, M. (2000), “Primate Cognition: Introduction to <strong>the</strong> Issue”, Cognitive Science, 24(3), pp. 351-<br />
361<br />
<strong>Tom</strong>asello, M. (2001), “Cultural Transmission: A View From Chimpanzees <strong>and</strong> Human Infants”, Journal<br />
of Cross-Cultural Psychology, 32(2), pp. 135-146<br />
<strong>Tom</strong>asello, M., Call, J. & Hare, B. (2003), “Chimpanzees underst<strong>and</strong> psychological states – <strong>the</strong> question<br />
is which ones <strong>and</strong> to what extent”, Trends in Cognitive Sciences, 7(4), pp. 153-156<br />
<strong>Tom</strong>asello, M., Kruger, A. C. & Ratner, H. H. (1993), “Cultural learning”, Behavioral <strong>and</strong> Brain<br />
Sciences, 16, pp. 495-552<br />
<strong>Tom</strong>asello, M., Call, J., Warren, J., Frost, G. T., Carpenter, M., & Nagell, K. (1997), “The ontogeny of<br />
chimpanzee gestural signals: A comparison across groups <strong>and</strong> generations”, Evolution of<br />
Communication, 1(2), pp. 223-259<br />
Torrance, S. (2005), “In search of <strong>the</strong> enactive: Introduction to special issue on enactive experience”,<br />
Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 4(4), pp. 357-368<br />
Trevar<strong>the</strong>n, C. B. (1979), “Communication <strong>and</strong> cooperation in early infancy: A description of primary<br />
intersubjectivity”, in: M. Bullowa (ed.), Before Speech, Cambridge, Cambridge University Press, pp.<br />
321-347<br />
Trevar<strong>the</strong>n, C. & Hubley, P. (1978), “Secondary intersubjectivity: Confidence, confiding <strong>and</strong> acts of<br />
meaning in <strong>the</strong> first year”, in: A. Lock (ed.), Action, Gesture <strong>and</strong> Symbol: The Emergence of<br />
Language, London, UK: Academic Press, pp. 183-229<br />
van Gelder, T. (1998), “The dynamical hypo<strong>the</strong>sis in cognitive science”, Behavioral <strong>and</strong> Brain Sciences,<br />
21(5), pp. 615-665<br />
van Gelder, T. (1999), “Wooden iron? Husserlian phenomenology meets cognitive science”, in: J. Petitot,<br />
F. J. Varela, B. Pachoud & J.-M. Roy (eds.), Naturalizing Phenomenology: Issues in Contemporary<br />
Phenomenology <strong>and</strong> Cognitive Science, Stanford, CA: Stanford University Press, pp. 245-265<br />
249 | P a g e
van Gelder, T. & Port, R. F. (1995), “It‟s About Time: An Overview of <strong>the</strong> Dynamical Approach to<br />
Cognition”, in: R. F. Port. & T. van Gelder (eds.), Mind as Motion: Explorations in <strong>the</strong> Dynamics of<br />
Cognition, Cambridge, MA: The MIT Press, pp. 1-43<br />
Varela, F. J. (1976), “Not One, Not Two”, The Co-Evolution Quarterly, 12, pp. 62-67<br />
Varela, F. J. (1979), Principles of Biological Autonomy, New York, NY: Elsevier North Holl<strong>and</strong><br />
Varela, F. J. (1984), “The creative circle: Sketches on <strong>the</strong> natural history of circularity”, in: P.<br />
Watzlawick (ed.), The Invented Reality: How do we know what we believe we know? Contributions<br />
to constructivism, New York, NY: W. W. Norton & Company, pp. 309-323<br />
Varela, F. J. (1991), “Organism: A meshwork of selfless selves”, in: A. I. Tauber (ed.), Organisms <strong>and</strong><br />
<strong>the</strong> Origins of Self, Dordrecht, Ne<strong>the</strong>rl<strong>and</strong>s: Kluwer Academic Publishers, pp. 79-107<br />
Varela, F. J. (1992), “Autopoiesis <strong>and</strong> a Biology of Intentionality”, in: B. McMullin & N. Murphy (eds.),<br />
Proc. of Autopoiesis <strong>and</strong> Perception: A Workshop with ESPRIT BRA 3352, Dublin, Irel<strong>and</strong>: Dublin<br />
City University, pp. 4-14<br />
Varela, F. J. (1996a), “The early days of autopoiesis: Heinz <strong>and</strong> Chile”, Systems Research, 13(3), pp. 407-<br />
416<br />
Varela, F. J. (1996b), “Neurophenomenology: A Methodological Remedy for <strong>the</strong> Hard Problem”, Journal<br />
of Consciousness Studies, 3(4), pp. 330-349<br />
Varela, F. J. (1997), “Patterns of Life: Intertwining Identity <strong>and</strong> Cognition”, Brain <strong>and</strong> Cognition, 34(1),<br />
pp. 72-87<br />
Varela, F. J. (1999), “The Specious Present: A Neurophenomenology of Time Consciousness”, in: J.<br />
Petitot, F. J. Varela, B. Pachoud & J.-M. Roy (eds.), Naturalizing Phenomenology: Issues in<br />
Contemporary Phenomenology <strong>and</strong> Cognitive Science, Stanford, CA: Stanford University Press, pp.<br />
266-317<br />
Varela, F. J. & Shear, J. (1999), „First-person Methodologies: What, Why, How?‟, Journal of<br />
Consciousness Studies, 6(2-3), pp. 1-14<br />
Varela, F. J., Maturana, H. R. & Uribe, R. (1974), “Autopoiesis: The organization of living systems, its<br />
characterization <strong>and</strong> a model”, BioSystems, 5, pp. 187-196<br />
Varela, F. J, Thompson, E. & Rosch, E. (1991), The Embodied Mind: Cognitive Science <strong>and</strong> Human<br />
Experience, Cambridge, MA: The MIT Press<br />
Velmans, M. (2007), “Where experiences are: Dualist, physicalist, enactive <strong>and</strong> reflexive accounts of<br />
phenomenal consciousness”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 6(4), pp. 547-563<br />
Von Foerster, H. (1960), “On Self-Organizing Systems <strong>and</strong> <strong>the</strong>ir Environments”, in: M. C. Yovits & S.<br />
Cameron (eds.), Self-Organizing Systems, London, UK: Pergamon Press, pp. 31-50<br />
250 | P a g e
Von Foerster, H. (1973), “On constructing a reality,” in: W. F. E. Preiser (ed.), Environmental Design<br />
Research, vol. 2, Stroudsburg, PA: Dowden, Hutchinson <strong>and</strong> Ross, pp. 35-46<br />
Von Foerster, H. (1976), “Objects: Tokens for (Eigen-)Behaviors”, ASC Cybernetics Forum, 8(3-4), pp.<br />
91-96<br />
von Glasersfeld, E. (1984), “An Introduction to Radical Constructivism”, in: P. Watzlawick (ed.), The<br />
Invented Reality: How do we know what we believe we know? Contributions to constructivism, New<br />
York, NY: W. W. Norton & Company, pp. 17-40<br />
von Uexküll, J. (1934), “Streifzüge durch die Umwelten von Tieren und Menschen: Ein Bilderbuch<br />
unsichtbarer Welten”, trans. by: C. H. Schiller, “A stroll through <strong>the</strong> worlds of animals <strong>and</strong> men: a<br />
picture book of invisible worlds”, in: C. H. Schiller (ed.), Instinctive Behavior: The Development of a<br />
Modern Concept, New York, NY: International Universities Press, 1957, pp. 5-80. Also appeared in:<br />
Semiotica, 89(4), 1992, pp. 319-391<br />
Vygotsky, L. S. (1934/1978), Mind in society: The development of higher psychological processes,<br />
Cambridge, MA: Harvard University Press<br />
Webb, B. (2001), “Can robots make good models of biological behaviour?”, Behavioral <strong>and</strong> Brain<br />
Sciences, 21, pp. 1033-1050<br />
Weber, A. & Varela, F. J. (2002), “Life after Kant: Natural purposes <strong>and</strong> <strong>the</strong> autopoietic foundations of<br />
biological individuality”, Phenomenology <strong>and</strong> <strong>the</strong> Cognitive Sciences, 1(2), pp. 97-125<br />
Wheeler, M. (1997), “Cognition‟s Coming Home: <strong>the</strong> Reunion of Life <strong>and</strong> Mind”, in: P. Husb<strong>and</strong>s & I.<br />
Harvey (eds.), Proc. of <strong>the</strong> 4 th Euro. Conf. on Artificial Life, Cambridge, MA: The MIT Press, pp. 10-<br />
19<br />
Wheeler, M. (2005), Reconstructing <strong>the</strong> Cognitive World: The Next Step, Cambridge, MA: The MIT<br />
Press<br />
Wheeler, M. (2008), “Cognition in Context: Phenomenology, Situated Robotics <strong>and</strong> <strong>the</strong> Frame Problem”,<br />
International Journal of Philosophical Studies, 16(3), pp. 323-349<br />
Williams, P., Beer, R. & Gasser, M. (2008), "Evolving referential communication in embodied dynamical<br />
agents", in: S. Bullock, J. Noble, R. Watson & M. Bedau (eds.), Artificial Life XI: Proceedings of <strong>the</strong><br />
11th Int. Conf. on <strong>the</strong> Simulation <strong>and</strong> Syn<strong>the</strong>sis of Living Systems, Cambridge, MA: The MIT Press,<br />
pp. 702-709<br />
Winograd, T. (1972), Underst<strong>and</strong>ing Natural Language, New York, NY: Academic Press<br />
Winograd, T. & Flores, C. F. (1986), Underst<strong>and</strong>ing Computers <strong>and</strong> Cognition: A New Foundation for<br />
Design, Norwood, NJ: Ablex Publishing Corporation<br />
Wood, R. & Di Paolo, E. A. (2007), “New Models for Old Questions: Evolutionary Robotics <strong>and</strong> <strong>the</strong> „A<br />
Not B‟ Error”, in: F. Almeida e Costa, L. M. Rocha, E. Costa, I. Harvey & A. Coutinho (eds.),<br />
251 | P a g e
Advances in Artificial Life: Proc. of <strong>the</strong> 9 th Euro. Conf. on Artificial Life, Berlin, Germany: Springer-<br />
Verlag, pp. 1141-1150<br />
Zahavi, D. (1996), Husserl und die transzendentale Intersubjektivität: Eine Antwort auf die<br />
sprachpragmatische Kritik, trans. by: E. A. Behnke, Husserl <strong>and</strong> Transcendental Intersubjectivity: A<br />
response to <strong>the</strong> Linguistic-Pragmatic Critique, A<strong>the</strong>ns, OH: Ohio University Press, 2001<br />
Zahavi, D. (1999), Self-Awareness <strong>and</strong> Alterity: A Phenomenological Investigation, Evanston, Ill.:<br />
Northwestern University Press<br />
Zahavi, D. (2001), “Beyond Empathy: Phenomenological Approaches to Intersubjectivity”, Journal of<br />
Consciousness Studies, 8(5-7), pp. 151-167<br />
Zahavi, D. (2005), Subjectivity <strong>and</strong> Selfhood: Investigating <strong>the</strong> First-Person Perspective, Cambridge,<br />
MA: The MIT Press<br />
Ziemke, T. (1999), “Rethinking Grounding”, in: A. Riegler, A. von Stein & M. Peschl (eds.),<br />
Underst<strong>and</strong>ing Representation in <strong>the</strong> Cognitive Sciences, New York, NY: Plenum Press, pp. 177-190<br />
Ziemke, T. (2001), “Are Robots Embodied?”, in C. Balkenius, J. Zlatev, H. Kozima, K. Dautenhahn & C.<br />
Breazeal (eds.), Proc. of <strong>the</strong> 1 st Int. Workshop on Epigenetic Robotics: Modeling Cognitive<br />
Development in Robotic Systems, Lund University Cognitive Studies, 85, pp. 75-93<br />
Ziemke, T. (2003), “What‟s that thing called embodiment?”, in: R. Alterman & D. Kirsh (eds.), Proc. of<br />
<strong>the</strong> 25 th Annual Conf. of <strong>the</strong> Cognitive Science Society, Mahwah, NJ: Lawrence Erlbaum, pp. 1305-<br />
1310<br />
Ziemke, T. (2007), “What‟s <strong>life</strong> got to do with it?”, in: A. Chella & R. Manzotti (eds.), Artificial<br />
Consciousness, Exeter, UK: Imprint Academic, pp. 48-66<br />
252 | P a g e