06.09.2021 Views

Mind, Body, World- Foundations of Cognitive Science, 2013a

Mind, Body, World- Foundations of Cognitive Science, 2013a

Mind, Body, World- Foundations of Cognitive Science, 2013a

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

2.10 Architectures against Homunculi<br />

We have described an algorithm for calculating an input-output mapping as a<br />

sequence <strong>of</strong> operations or behaviours. This description is misleading, though,<br />

because the notion <strong>of</strong> sequence gives the impression <strong>of</strong> a linear ordering <strong>of</strong> steps.<br />

However, we would not expect most algorithms to be linearly organized. For instance,<br />

connectionist cognitive scientists would argue that more than one step in an algorithm<br />

can be carried out at the same time (Feldman & Ballard, 1982). As well, most<br />

algorithms <strong>of</strong> interest to classical cognitive scientists would likely exhibit a markedly<br />

hierarchical organization (Miller, Galanter, & Pribram, 1960; Simon, 1969). In<br />

this section, I use the notion <strong>of</strong> hierarchical organization to motivate the need for an<br />

algorithm to be supported by an architecture.<br />

What does it mean for an algorithm to be hierarchical in nature? To answer<br />

this question, let us again consider the situation in which behavioural measurements<br />

are being used to reverse engineer a calculating black box. Initial experiments<br />

could suggest that an input-output mapping is accomplished by an algorithm<br />

that involves three steps (Step 1 Step 2 Step 3). However, later studies<br />

could also indicate that each <strong>of</strong> these steps might themselves be accomplished by<br />

sub-algorithms.<br />

For instance, it might be found that Step 1 is accomplished by its own fourstep<br />

sub-algorithm (Step a Step b Step c Step d). Even later it could be<br />

discovered that one <strong>of</strong> these sub-algorithms is itself the product <strong>of</strong> another sub-subalgorithm.<br />

Such hierarchical organization is common practice in the development<br />

<strong>of</strong> algorithms for digital computers, where most programs are organized systems <strong>of</strong><br />

functions, subfunctions, and sub-subfunctions. It is also a common characteristic <strong>of</strong><br />

cognitive theories (Cummins, 1983).<br />

The hierarchical organization <strong>of</strong> algorithms can pose a problem, though, if an<br />

algorithmic account is designed to explain a calculating device. Consider our example<br />

where Step 1 <strong>of</strong> the black box’s algorithm is explained by being hierarchically<br />

decomposed into the sub-algorithm “Step a Step b Step c Step d.” On closer<br />

examination, it seems that nothing has really been explained at all. Instead, we have<br />

replaced Step 1 with a sequence <strong>of</strong> four new steps, each <strong>of</strong> which requires further<br />

explanation. If each <strong>of</strong> these further explanations is <strong>of</strong> the same type as the one to<br />

account for Step 1, then this will in turn produce even more steps requiring explanation.<br />

There seems to be no end to this infinite proliferation <strong>of</strong> algorithmic steps that<br />

are appearing in our account <strong>of</strong> the calculating device.<br />

This situation is known as Ryle’s regress. The philosopher Gilbert Ryle raised it<br />

as a problem with the use <strong>of</strong> mentalistic terms in explanations <strong>of</strong> intelligence:<br />

Must we then say that for the hero’s reflections how to act to be intelligent he must<br />

first reflect how best to reflect to act? The endlessness <strong>of</strong> this implied regress shows<br />

46 Chapter 2

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

Saved successfully!

Ooh no, something went wrong!