The Mind Doesn't Work That Way: The Scope and Limits of Computational Psychology (review)

As has been his wont in recent years, Jerry Fodor offers here a statement of deepest pessimism about the possibility of doing cognitive science except in a very limited class of subdomains. F is of course justly celebrated for at least two major ideas in cognitive science: The language of thought (Fodor 1975) and the modularity hypothesis (Fodor 1983). However, the form in which these ideas have been taken enthusiastically into the lore of the field differs in some important respects from the form in which F couched them and in which he still believes. As I hope to show, the tension between F’s actual views and those generally attributed to him plays a major role in the position he advocates here.

Here is a summary of F’s argument, as best as I can reconstruct it. The central issue is the problem of ‘abduction’: how one determines the truth of a proffered proposition and its consistency with one’s beliefs. The chief obstacle to successful abduction is that meaning is holistic: One must potentially check the proffered proposition and inferences from it against one’s entire network of belief/knowledge. The resulting combinatorial explosion makes it impossible to reliably fix new beliefs and plan new actions within a traditional Turing-style computation. For F, this casts serious doubt on the computational theory of mind, which presumes Turing-style ‘symbolic’ computation over the syntactic form of mental representations.

F dismisses a number of proposed solutions to the problem of abduction. Connectionist style computation, he maintains, is actually a step backward, since it cannot even capture the characteristic free combinatoriality of thought, an essential feature of the language of thought hypothesis. Here I concur; Marcus 2001 offers an extended argument to this effect. F also argues that a system of heuristics is unsatisfactory since one needs to perform an abduction to determine which heuristic to apply. I find this argument less convincing; we’ll return to it below.

It is worth mentioning that when F speaks of ‘Turing-style computation’, it is not clear whether he intends to include massively parallel ‘symbolic’ computation. Such computation is perhaps mathematically equivalent to serial Turing-style computation, but it is quite different in practical terms. Certainly the brain’s form of computation is massively parallel, whether connectionist or symbolic or some combination thereof. It is interesting therefore to ask if such computation is of any practical help in solving the combinatorial explosion of abduction; F does not address this question.

F’s dismissals of connectionism and heuristics, however, are just warmups for his principal line of attack. This is aimed against the thesis of ‘massive modularity’ proposed by such people as Pinker (1997) and Cosmides and Tooby (1992): the idea that the entire mind (or most of it, anyway) is made up of innate domain-specific modules. Most everybody seems to consider this a natural extension of F’s modularity hypothesis. What is frequently overlooked is that the final chapter of Fodor 1983 argues at length that, unlike input systems, thought cannot be modular, because of meaning holism.

F sticks to his guns in the present book. He observes that if massive modularity were correct, it would keep abduction under control; each module would have access only to a small subset of one’s knowledge and thus would avoid the combinatorial explosion of dealing with everything at once. He then argues that, on a variety of grounds, massive modularity cannot be correct. Therefore, he concludes, abduction remains a crushing objection to the computational theory of mind.

F then trains his armaments on evolutionary psychology. Pinker and Cosmides and Tooby, among many others, have urged that a proper theory of the mind/brain ought to include a reasonable story of how it got to be the way it is; hence considerations of evolutionary plausibility ought to play a role. In general, natural selection tends to come...