02.11.2012 Views

Reduction and Elimination in Philosophy and the Sciences

Reduction and Elimination in Philosophy and the Sciences

Reduction and Elimination in Philosophy and the Sciences

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Rosen 1997) call objectual reduction. Ra<strong>the</strong>r, we might<br />

start out by treat<strong>in</strong>g our computational practice as<br />

constitut<strong>in</strong>g a term-<strong>in</strong>troduc<strong>in</strong>g “procedural” <strong>the</strong>ory Tp.<br />

Such a <strong>the</strong>ory would conta<strong>in</strong> not only st<strong>and</strong>ard ma<strong>the</strong>matical<br />

terms <strong>and</strong> quantifiers, but also terms (A1, A2, ...)<br />

nam<strong>in</strong>g algorithms <strong>and</strong> quantifiers (∀X1, ∀X2, ...) rang<strong>in</strong>g<br />

over such entities.<br />

The question as to whe<strong>the</strong>r Tp commits us to <strong>the</strong><br />

existence of a non-ma<strong>the</strong>matical class of entities<br />

correspond<strong>in</strong>g to <strong>the</strong> range of <strong>the</strong> procedural quantifiers<br />

can accord<strong>in</strong>gly be formalized by ask<strong>in</strong>g whe<strong>the</strong>r it is<br />

possible to <strong>in</strong>terpret Tp over a purely ma<strong>the</strong>matical <strong>the</strong>ory<br />

Tm ⊆ Tp. In particular, we can ask whe<strong>the</strong>r Tp is<br />

conservative over Tm for purely ma<strong>the</strong>matical statements<br />

<strong>and</strong> also whe<strong>the</strong>r it is possible to formulate Tm <strong>in</strong> a manner<br />

such that it is able to derive appropriate <strong>in</strong>terpretations of<br />

results of types II) <strong>and</strong> III).<br />

Demonstrat<strong>in</strong>g <strong>the</strong> former fact is likely to be<br />

straightforward as it requires only that Tm is able to prove<br />

<strong>the</strong> correctness of various algorithms (<strong>in</strong> <strong>the</strong> sense of note<br />

2) relative to some means of represent<strong>in</strong>g <strong>the</strong>m which<br />

need not reflect <strong>the</strong>ir <strong>in</strong>tensional properties such as<br />

runn<strong>in</strong>g-time. However, construct<strong>in</strong>g an <strong>in</strong>terpretation<br />

which is also capable of account<strong>in</strong>g for results <strong>in</strong><br />

algorithmic analysis will most likely require that we attend<br />

to <strong>the</strong> details of how we make reference to <strong>in</strong>dividual<br />

algorithms <strong>in</strong> practice. Reflection on this topic suggests<br />

that: 1) <strong>the</strong> only l<strong>in</strong>guistic means we have of referr<strong>in</strong>g to<br />

<strong>in</strong>dividual algorithms is via expressions of <strong>the</strong> form “<strong>the</strong><br />

algorithm implemented by mach<strong>in</strong>e m” 7 ; 2) we generally<br />

take it to be possible to refer to <strong>the</strong> same algorithm by<br />

referr<strong>in</strong>g to different mach<strong>in</strong>es. As a consequence, Tp is<br />

likely to conta<strong>in</strong> many statements of <strong>the</strong> form imp(m1) =<br />

imp(m2), <strong>and</strong> imp(m1) ≠ imp(m2) (where imp(⋅)) is <strong>in</strong>tended<br />

to formalize “<strong>the</strong> algorithm implemented by m”).<br />

This latter observation illustrates why it is unlikely<br />

that <strong>the</strong> <strong>in</strong>terpretation of an algorithmic name like<br />

MERGESORT can be taken to be identical to any particular<br />

mach<strong>in</strong>e. 8 If Tp is to reflect <strong>the</strong> grammatical structure of<br />

statements like those <strong>in</strong> II) <strong>and</strong> III), this suggests that we<br />

7 The o<strong>the</strong>r option is to treat algorithms as correspond<strong>in</strong>g to <strong>the</strong> denotations of<br />

programs -- i.e. l<strong>in</strong>guistic descriptions of procedures given over a formal programm<strong>in</strong>g<br />

language. However, reference to algorithms via this route arguably<br />

collapses <strong>in</strong>to reference via mach<strong>in</strong>es as each program will be <strong>in</strong>terpretable as<br />

a mach<strong>in</strong>e via an appropriate form of operational semantics.<br />

8 For if we take A = M for a fixed M <strong>the</strong>re will generally be no way of def<strong>in</strong><strong>in</strong>g<br />

imp so that <strong>the</strong>se identity <strong>and</strong> non-identity statements are satisfied. More<br />

generally, such a proposal will entail that <strong>the</strong> computational properties of A are<br />

identical to those of M. But this will generally be unacceptable s<strong>in</strong>ce mach<strong>in</strong>es<br />

possess a variety of “artifactual” properties which we generally do not attribute<br />

to algorithms -- e.g. have a fixed number of states, hav<strong>in</strong>g exact (as opposed<br />

to asymptotic) runn<strong>in</strong>g-time, etc.<br />

60<br />

Algorithms <strong>and</strong> Ontology — Walter Dean<br />

must take <strong>the</strong> values of imp(⋅) to be equivalences classes<br />

of mach<strong>in</strong>es under a def<strong>in</strong>ition of computational<br />

equivalence ≈ def<strong>in</strong>ed over a suitable class of mach<strong>in</strong>es<br />

M. 9 Such a def<strong>in</strong>ition would ideally serve to analyze <strong>the</strong><br />

mean<strong>in</strong>g of statements of <strong>the</strong> form<br />

(M) mach<strong>in</strong>es m1 <strong>and</strong> m2 implement <strong>the</strong> same algorithm<br />

<strong>in</strong> a manner which additionally satisfied all statements of<br />

algorithmic identity <strong>and</strong> non-identity conta<strong>in</strong>ed <strong>in</strong> Tp. On<br />

this proposal, <strong>the</strong> susta<strong>in</strong>ability of (A) will rest on <strong>the</strong><br />

availability of such a def<strong>in</strong>ition of equivalence. If such a<br />

def<strong>in</strong>ition could be given, we would have shown how it was<br />

possible to contextually reduce procedural discourse to<br />

ma<strong>the</strong>matical discourse (aga<strong>in</strong> <strong>in</strong> <strong>the</strong> sense of Burgess &<br />

Rosen). The ontological status of algorithms could<br />

accord<strong>in</strong>gly be taken to be that of (neo)-Fregean abstracts<br />

over M relative to ≈.<br />

Literature<br />

Burgess, John & Gideon Rosen 1997 A subject with no object,<br />

Oxford: Clarendon Press.<br />

G<strong>and</strong>y, Rob<strong>in</strong> 1980 “Church’s <strong>the</strong>sis <strong>and</strong> pr<strong>in</strong>ciples for mechanisms”<br />

<strong>in</strong> Jon Barwise,<br />

H. J. Keisler, <strong>and</strong> K. Kunen (eds.) The Kleene Symposium, Amsterdam:<br />

North-<br />

Holl<strong>and</strong>, 123–148.<br />

Gurevich, Yuri 1999 “The sequential ASM <strong>the</strong>sis.” Bullet<strong>in</strong> of <strong>the</strong><br />

EATCS, 67, 93-125.<br />

Hopcroft, John & Jeffrey Ullman 1979 Introduction to Automata<br />

Theory, Languages, <strong>and</strong> Computation Boston: Addison-Wesley.<br />

Knuth, Donald 1973 The art of computer programm<strong>in</strong>g, volumes I-I<br />

I I. Boston: Addison Wesley.<br />

Moschovakis, Yiannis 1998 “On <strong>the</strong> found<strong>in</strong>g of a <strong>the</strong>ory of algorithms”<br />

H. G. Dales & G. Oliveri, (eds.), Truth <strong>in</strong> ma<strong>the</strong>matics, 71–<br />

104. Oxford: Clarendon Press.<br />

Rogers, Hartley 1967 Theory of Recursive Functions <strong>and</strong> Effective<br />

Computability.<br />

9 This fact is recognized by Moschovakis (who identifies algorithms as equivalence<br />

classes of computational models known as recursors) but it is ultimately<br />

denied by Gurevich. Even for Moschovakis, however, <strong>the</strong> question rema<strong>in</strong>s<br />

whe<strong>the</strong>r his chosen notion of equivalence ei<strong>the</strong>r 1) serves to analyze <strong>the</strong><br />

mean<strong>in</strong>g of statements of <strong>the</strong> form (M) <strong>and</strong> 2) <strong>in</strong>duces identity questions on<br />

algorithms which are consistent with those reflected by Tp.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!