From Weak to Strong Functionalism, and Back

Although it is customary to begin the discussion of functionalism with an introduction of the notion of Turing machines, such a construal of it, engaging the apparatus of mathematics at the level of the doctrine’s most fundamental concepts, has the disadvantage of obscuring some of its aspects which deserve highlighting before we proceed with a more detailed analysis. It is well to remark at the outset that the very idea of the mind’s being a functionally organized structure can easily be formulated without recourse to a particular theory of computability or, indeed, without any mention of the computer metaphor. What seems to underlie all the technical jargon of various functionalist avowals, whether they employ the language of neural networks or the formalism of Ramsey sentences, is the basic intuition that mental states are in fact or should in principle be defined on the basis of causal connections which relate them to other mental states, on the one side, and to stimuli and responses, on the other. One virtue of this informal definition is that it shows that functionalism springs from the two apparently conflicting sources we have discussed above, namely identity theory and logical behaviorism; it combines the behaviorist insight that mental states are individuated by virtue of their causes and effects with the physicalist claim that they are situated in the brain and cannot be reduced to behavioral dispositions. It is actually very tempting to see in functionalism a synthesis, as it were, of these two broadly materialist views. And my use of Hegelian dialectics in this connection is not merely stylistic either; it points to a disconcerting fact about all syntheses, including functionalism: a combination of two good ideas may not be such a good idea, after all. Just as the proponents of functionalism will argue that their doctrine takes the best of both philosophical worlds, so its opponents will claim that it is open to all the devastating charges made against behaviorism and identity theory, and possibly some new ones. At any rate, our final verdict will depend on some pretty general philosophical convictions. Another virtue of my loose construal of the functionalist thesis is that it does not commit the functionalist to any particular notion of computability or research program in AI, thereby accommodating for future developments in mathematics and computer science. It is true that by 1960, when Putnam first suggested the position, Church’s thesis, asserting the equivalence of the intuitive notion of what a human being can in principle calculate with Turing-machine computability, had been so firmly established that its employment in a philosophical theory must have been entirely uncontroversial. However, the advent of connectionism and other mathematical results have since made the agreement on that point less general. By my informal