The Mind Doesn't Work That Way:
The Scope and Limits of Computational Psychology
by Jerry Fodor
MIT Press/Bradford Books, 126 pp., £15.95 (hbk.)
ISBN 0 262 06212 7
Published 29 September 2000

Reviewed by Keith Sutherland
Times Higher Education Supplement
10-11 March 2001

A few years ago the Royal Society organised a short meeting on the 'problem of consciousness' at their headquarters in Carlton House Terrace. Most of the speakers -- including Jeffrey Gray from the Institute of Psychiatry and psychologists Max Velmans and Stevan Harnad -- could be positioned on the 'mysterian' wing of consciousness studies. Mysterians believe that there would still be 'something missing' even after a complete description of brain states or the mental modules beloved by cognitive science.

Jerry Fodor, the author of The Mind Doesn't Work That Way, could also be classified as a consciousness mysterian, but then so would his opponent Steven Pinker, who's How the Mind Works is the explicit target of this short polemic. However, not content with one mystery, Fodor claims that the real mystery of psychology is how to extend the dominant computational model to account for the holistic properties of cognition ('abduction' to the cognoscenti and the artificial intelligentsia; 'common sense' or 'flexibility' to the rest of us).

Fodor acknowledges that artificial intelligence has been moderately successful in its attempts to model domain-specific cognitive processes. The dominant theory (which Fodor helped to establish) is the fusion of nativism -- the mind as composed of a series of innate modules -- and computationalism -- thinking as essentially a syntactic operation on the local 'database' that goes to make up the module under consideration. According to this theory, local mental processes are operations defined on syntactically structured mental representations that are much like sentences (the 'language of thought'). The emphasis on syntax presupposes the primary assumption of rationalist psychology -- that beliefs, desires, thoughts and the like have logical forms (and are thus amenable to computational modelling).

But hang on a minute, says Fodor, the modularity/computational assumption may be the sine qua non of classical AI, but unfortunately the Mind Doesn't Work That Way. The computational theory of mind presupposes that syntactic mental processes are insensitive to context-dependent properties of mental representations, and this is plainly not the case. If thinking is a mechanical syntactic operation on a local database (informational encapsulation), then how do we account for the properties of a thought 'that are sensitive to which belief systems it's embedded in' (Fodor's italics)?

Unfortunately (at least from the point of view of AI and computational modelling), human cognition is characterised by what philosophers call abductive (holistic) inference. Abductive reasoning is known in the jargon of artificial intelligence as the 'frames problem' and no one has yet figured out how to solve it. According to Fodor, Chicken Little was right all along -- 'abduction really is a terrible problem for cognitive science'.

Some philosophers and scientists approach the problem by arguing that the apparent success of abductive reasoning is just an illusion -- all we have is local approximations to global processes, and perhaps the problem of calculating these approximations is solved heuristically, case by case. But Fodor dismisses this as a circular argument. The computational theory of mind can't be patched up with quick heuristic fixes, as the inference about which local heuristic to employ are themselves often abductive.

Fodor argues that the failure of classical artificial intelligence -- even with heuristic fixes -- to produce successful simulations of routine commonsense cognitive operations is 'scandalous'. But he is even more dismissive of the claims of the new kid in town -- connectionism -- as a competing theory. Connectionist networks fail to provide either a plausible account of the causal consequences of logical form or a plausible account of abductive inference. Here Fodor starts to sound off in the style of the recently deceased Auberon Waugh -- connectionist networks are 'simply hopeless -- it must be the sheer magnitude of their incompetence that makes them so popular'.

But Fodor is not a newspaper columnist, he is a philosopher by profession, best known as one of the original pioneers of the fusion of Turing's computational theory of mind with Chomskian nativism. His two books The Language of Thought (1975) and The Modularity of Mind (1983) have been immensely influential, and yet in his current book he argues that the really interesting aspects of human cognition are neither modular nor syntactic (or at least, in the latter case, possibly not). This leads him to claim that 'so far, what our cognitive science has found out about the mind is mostly that we don't know how it works' and he even refers to his 1983 book as adopting an 'unhelpful policy'. Is Fodor undergoing a Wittgensteinian epistème -- will scholars end up discussing the 'early' and the 'late' Fodor -- or are there other factors at work?

Although Fodor claims that this book is primarily a light-hearted and polemical attack on the New Synthesis (the fusion of the computational theory of mind and the 'massive modularity' thesis with a Darwinian theory of origins), he admits that he is 'worried half to death' about the failure of classical cognitive science, and explains the 'pervasive good cheer' of writers like Steven Pinker and Henry Plotkin with the aid of the psychoanalytic concepts of suppression and deep denial. Otherwise how could Pinker admit that we are far from being able to build a serviceable robot, but still claim that we know, more or less, how the cognitive mind works?

One of the reasons that the Faustian pact that constitutes cognitive science has been such a flop is that no-one can even agree what the discipline is meant for (science or engineering). On the one hand the human and social sciences (psychology, anthropology, linguistics, neuroscience, philosophy etc) are interested in understanding how the mind works as an end in itself or in order to improve the human condition. On the other hand computer science and artificial intelligence are engineering disciplines -- 'AI was generally supposed to be about engineering, not about science; and certainly not about philosophy'. Although Marvin Minsky and Aaron Sloman may like to think they are researching how the mind works, their paymasters (mostly in the military) are more interested in ensuring that the next generation of Cruise missiles hits Saddam's bunker, rather than the hospital next door [but see note]. A similar confusion might be imagined if we were to merge literary criticism and quantum physics (oops, Alan Sokal got there first).

An example of this confusion can be found in Fodor's repeated claim that Alan Turing's principal contribution to cognitive science was his discovery that 'thinking was syntactic'. But how and when did Turing make this 'discovery'? Although it may have amused him to dream up parlour games between human subjects and robots, the bosses at Bletchley Park were more concerned with cracking the German Enigma codes. The electronic computer was invented for this purpose, and Turing machines by definition solve problems using local syntactic processes only. The technopomorphic projection of this architecture on to human subjects has been the prime cause of the confusion in cognitive science.

To return to our Royal Society meeting, Stevan Harnad's take-home message was that even if we are never likely to crack the problem of consciousness, there is still plenty of bread-and-butter work left to do in cognitive science. Fodor endorses this, arguing that we should 'plug on at the problems about the mind that we do know how to think about'. Fair enough, but why then bother to write this book?

The answer, of course, is to deflate the 'relentless cheerfulness' of the Pinkers of this world and to undermine the tedious and hubristic claims that we are about to conquer the 'last frontiers of human understanding' etc. etc. According to Fodor, the main achievement of forty years of cognitive science has been to 'throw light on how much dark there is.' Such refreshing outpourings of humility are few and far between and this book should be on all Lent term reading lists for this reason alone.

But the other value of the book is that it shows how philosophy is still an important discipline. No doubt the proponents of classical AI and connectionism like to see their work judged on purely technical grounds, but to Fodor the arguments in AI reflect an epistemological dispute that goes back at least as far as Kant and Hume ('you can drop out for a couple of centuries and not miss a thing') and ultimately to Plato and Aristotle. Fodor is proud to nail his own philosophical colours to the mast as an Enlightenment rationalist and describes the theories of his empiricist opponents as 'appalling'. Fodor's claim that bona fide abductive inferences are nonlocal, hence noncomputational, by definition closely parallels Roger Penrose's argument that human cognition involves a noncomputational element. Most people working in cognitive science are quick to dismiss Penrose as a deranged Platonist -- 'out of his microtubules' -- who should eff off and mind his own business (mathematical physics), but it's difficult to dismiss the Fodor polemic so easily as this is an attack from a respected (fifth-) columnist within the camp.

Fodor's other target is the growing arrogance and imperialism of neo-Darwinism and evolutionary psychology. To Fodor, drawing on Chomsky's argument that the development of natural language cannot be accounted for by Darwinian processes, nativism does not necessarily imply adaptationism. Fodor describes the argument that cognitive nativism necessarily implies Darwinism as an indication of the politicisation of science. He claims that 'some radical reorganisation of global cognitive structure' must have occurred in the process of getting from the minds of apes to homo sapiens, and it is hard to see how this could be attributed to Darwinian processes. I don't know whether this will lead to any premature rejoicing in the Bible Belt, but it would be amusing if the Amazon page for Fodor's book starts to attract similar review puff to the books by creationists and intelligent design theorists.

One final caveat on an otherwise enjoyable and informative read. MIT Press/Bradford Books became the leading publisher in cognitive science partly on account of the press's exacting editorial and production standards. However this book does not appear to have been anywhere near a proof reader or copy editor and the style oscillates wildly between the witty polemic of the broadsheet columnist and dense philosophical jargon. (The publishers also appear to have decided that the book did not merit a proper index.) All this is unfortunate, as it deserves to be widely read beyond the ivory towers of the philosophical community. The young Turks at Bradford Books have failed to live up to the high editorial standards of the founders, the sadly-missed Harry and Betty Stanton.

Keith Sutherland is executive editor, Journal of Consciousness Studies

[2005-12-12:  Responding to Sloman's objection that the claim he worked for the military is false and libellous, Sutherland writes, "My unreserved apologies to Aaron Sloman for an unintended libel, caused by a combination of sloppy journalism and the very tight word-limit of THES reviews. What I meant to say was that, as an engineering discipline, much of the research in artificial intelligence has been financed by the military. I was quite wrong to personalise it in this way and apologise without reservation."]