Psychology is concerned with how the mind forms concepts. Philosophy is interested too, but more interested in forming the right concepts. To get a theory of mind right, a concept itself, one of the eliminative materialists, Stephen Stich, argues that one has to give an accurate description of the concept, or the body of tacit knowledge, that “underlies our quotidian practice.” Once we have that the challenge to folk psychology is underway.

In the psychological literature, the most widely known challenge to the assumptions of our traditional philosophical analysis (the Socratic folk psychology in which categorizations in the mind have one-to-one match-ups in the physical view) is the model provided by Eleanor Rosch and her colleagues. On this model, mental structures that underlie our judgments do not exploit tacitly known necessary and sufficient conditions for category membership, “or anything roughly equivalent,” Stich adds.

The Rosch model, with its emphasis on idealized descriptions about “prototypes” and their similarities to other categorizations, says that categorizations are made without meeting conditions that the commonsense view, exemplar theory, would predict. Categorizations are instead determined by “tacit similarity knowledge”.

Though this model does not talk about intentional states like beliefs and desires, it is one more example how the commonsense paradigm does not make concrete one-to-one match-ups with the world. Stich suggests that there is no underlying concept motivating categorization judgments, as the folk theory predicts. He argues that subjects construct “various different sorts, on the fly” in response to a situation where concept-formation generally happens. Hence there are no one-to-one match-ups, so the traditional Socratic method of proposing definitions and then searching for intuitive counter-examples will have to be dropped.

If our introspective accounts about whether a mental state has the property of qualia, or the propositional “that p”, are motivated by the sorts of prototypes of the Rosch model, as opposed to exemplars or tacit theories, or if the judgments we make result spontaneously as subatomic particles do in quantum theory, then the conditions that they intuitively fall under a target concept will also be dropped.

In other words, the “classical structure” of defining the things we talk about has less weight, as Stich says. He proposes that the traditional Socratic method will have to be superseded by a cognitive science approach instead. (In Kripke’s example of gold being some sort of yellow substance, but not fool’s gold, we are also reminded that the scientific approach of defining it in terms of its chemical composition with the atomic number 79 allowed us to speak of rigid designators, and this is analogous to the way that cognitive science should be able to talk about mental states.)

The Rosch model is not the only model in support of this thesis. There are various computational models in support, and several papers on the relationship between connectionism and eliminativism are in support of the mature approach. This pluralism might be thought to be evidence against the idea that one explanation of cognitive phenomena is needed.

But this is precisely Stich’s point. “For if different paradigms within cognitive science use different notions of representation, then there isn’t going to be a theory of mental representation of the sort we have been discussing. There will be lots of theories.” It will make little sense to ask which one is the right theory, since each theory exploits a different branch of cognitive science, but we can be sure that the traditional method is a stagnant, degenerating research program.

But it does look like what we see in cognitive science is the beginning of an argument for eliminativism, since cognitive theories and folk theories make incompatible claims, and the folk theories are not reducible to the phenomena cognitive science exploits.

Paul and Patricia Churchland have emphasized the mismatch between the sentential structure of propositional attitudes on the one hand, and the actual neurological structures of the brain on the other hand. Whereas the former involves discrete symbols and a combinatorial syntax, the latter involves action potentials, spiking frequencies and spreading activation. As the Churchlands have argued, it is hard to see where in the brain we are going to find anything that even remotely resembles the sentence-like structure that appears to be essential to beliefs and other propositional attitudes.

Stich has emphasized, on the other hand, that folk psychology individuates beliefs by virtue of their semantic properties, e.g., we taxonomize states like beliefs by virtue of what they are about. However, according to Stich, there are a host of reasons for rejecting a semantic taxonomy for scientific psychology.

Semantic taxonomies ignore causally salient aspects of cognitive states, involve a high degree of vagueness, and break down in the case of the mentally ill or the very young. In place of the semantic individuation method adopted by folk psychology, Stich argues for a syntactic taxonomy that is based upon the causally relevant syntactic or physical properties of a given cognitive state.