The cycle’s engagement with artificial intelligence is pervaded by folk-psychological language: the AI “knows,” the AI “believes,” the AI “understands” or fails to understand. Eliminative materialism is the philosophical challenge to take that language seriously as a theory rather than a transparent description—to ask whether the folk vocabulary is the right vocabulary for what these systems are doing, or whether it is being stretched to cover something structurally different. The machines that Edo Segal describes in [YOU] on AI—systems that write, reason, and produce what looks for all the world like understanding—are, by Churchland’s analysis, the strongest empirical argument his thesis has yet received: intelligence without beliefs, competence without the folk-psychological furniture.
The fluency-authority decorrelation that the cycle identifies as the signature hazard of the AI transition is, through Churchland’s lens, precisely what eliminativism predicts: a system that produces outputs indistinguishable from belief-guided behavior without the beliefs that were supposed to make such behavior reliable. If folk psychology were true, a system that produces correct-seeming assertions would have to have correct beliefs behind them; if folk psychology is false, the outputs and the beliefs can come entirely apart, and we have built systems that demonstrate the coming-apart at scale.
Churchland’s 1981 paper “Eliminative Materialism and the Propositional Attitudes,” published in the Journal of Philosophy, is the canonical statement of the thesis. The paper opens with the claim that folk psychology constitutes “a radically false theory, a theory so fundamentally defective that both its principles and its ontology will eventually be displaced, rather than smoothly reduced, by a completed neuroscience”—and it proceeds with remarkable clarity through the distinction between reduction and elimination, the inductive argument from the graveyard of folk theories, and the response to the self-refutation objection.
The self-refutation charge—pressed most forcefully by Jerry Fodor—holds that eliminativism cannot be believed if there are no beliefs to believe it with, and that the thesis therefore undermines itself. Churchland’s reply is that this objection smuggles the folk-psychological framework into the very form of the complaint, assuming that assertion must consist in expressing a belief—which is precisely what is under dispute. He offers the historical parallel: a seventeenth-century vitalist could object that without vital spirit no living creature could argue for eliminativism about vital spirits, and the objection would be formally valid but philosophically empty, since vitalism turned out to be false and life turned out not to require any spirit.
The distinction between reduction and elimination. Temperature was reduced to mean molecular kinetic energy: heat survived, explained and relocated. Phlogiston was eliminated: it was not relocated but abandoned, along with the entire theoretical framework that required it. Churchland argues that folk psychology is headed for elimination rather than reduction because its central posits—beliefs with propositional structure, desires with logical form—do not appear to map onto anything the brain contains. A theory is eliminated when its core concepts refer to nothing; folk psychology’s core concepts may refer to nothing the brain is built to contain.
The inductive argument from folk-theory history. Every folk theory that humanity has subjected to mature scientific scrutiny—folk physics, folk biology, folk cosmology, folk medicine—has turned out to be radically mistaken in its fundamentals. Folk psychology is a folk theory addressing the most complex object in the known universe. By the plainest induction from the history of ideas, the probability that folk psychology alone happens to be correct is vanishingly small. The elimination of other folk theories was achieved not by refuting them in debate but by developing a better science that made their vocabulary unnecessary.
The machines as existence proof. For decades the self-refutation charge carried its force partly because every known thinking thing was a person described by folk psychology. The symbolic AI tradition made matters worse by building its architecture on folk-psychological principles, reinforcing the assumption that intelligence requires beliefs. The arrival of large neural networks—systems that perform cognitive tasks at human or superhuman levels without containing anything resembling a propositional attitude—breaks the spell. The machines do not prove Churchland’s thesis; they make the key assumption of the self-refutation argument look like an assumption rather than a necessary truth.
Plasticity as the path to the new vocabulary. Eliminativism is not merely a negative claim; Churchland argued that if folk psychology is displaced, we could—and eventually will—learn to experience ourselves through the categories of a mature neuroscience. The plasticity of mind means that the eliminativist future is a genuine human possibility: as the new vocabulary becomes familiar, perception itself will shift, and the phenomenology we currently describe in folk-psychological terms will be re-experienced through the richer, truer framework. The machines, in remaking the cognitive habits of those who work with them, are perhaps already beginning this transformation by routes Churchland could not have anticipated.