Category Mistake

In the AI Story

The structure of a category mistake is subtle because the person making it does not feel confused. The Oxford visitor feels that his guide is being evasive, not that his question has defective grammar. This is because logical grammar is the medium of thought, not its object. You cannot easily see what you are thinking with. Ryle's philosophical method is, in essence, the slow, patient surfacing of these invisible grammatical structures and the demonstration that they generate pseudo-problems — questions that feel deep because they resist resolution, but resist resolution because they have no genuine content to resolve.

Category mistakes flourish at the intersections of different logical types. When the large language model produces fluent, contextually appropriate output, two different logical types collide: the computational (processing tokens, applying attention) and the behavioral (responding intelligently, demonstrating understanding). Each is legitimate at its own level. The mistake is to treat them as competing descriptions of the same kind of thing — to ask whether, alongside the processing, the thinking also occurs. That 'also' is the grammatical marker of the mistake. It demands an additional entity in the inventory where the inventory has already been completed.

The resilience of category mistakes explains why the AI debate has proved so intractable. Both triumphalists and skeptics share the same defective grammar. The triumphalist finds the ghost; the skeptic fails to find it; neither questions whether a ghost was the right thing to look for. The debate proceeds with great intensity about a question whose grammar is malformed, while the genuine questions — about behavioral reliability, about the judgment economy, about what humans must become — go unasked. Dissolving the category mistake does not settle those questions. It makes them askable.

The practical consequence is that recognizing a category mistake is not a rhetorical victory but a clearing operation. Once the visitor understands that the University is not a thing alongside the colleges, he can begin to ask genuine questions: How is the institution organized? Who governs it? What binds the colleges together? These are answerable questions, and the visitor is now equipped to pursue them. Similarly, once the AI discourse stops asking whether the machine 'really' thinks, it can begin asking what the machine actually does, how reliably, and with what implications for human practice.

Origin

The concept appears in the opening pages of Ryle's 1949 The Concept of Mind, where the Oxford visitor example serves as the doorway to the book's central argument against Cartesian dualism. Ryle deploys it to show that the mind-body problem, as traditionally posed, rests on a category mistake of exactly the Oxford-visitor variety: treating 'mind' as a thing alongside the body rather than as a characterization of the body's behavior under certain aspects.

Key Ideas

Logical grammar, not factual error. A category mistake is not wrong in the way a scientific hypothesis is wrong. It is confused in a way that makes its resolution impossible until the grammar is repaired.

Resilience through invisibility. The mistake is built into the framing of the question, which makes it invisible to the person asking. The visitor cannot see the mistake because the mistake is how he is seeing.

Dissolution, not answer. The response to a category mistake is not to provide the missing piece but to show that nothing was missing — the inventory was complete, the question was malformed.

The AI application. The question 'Does the machine really think?' treats thinking as a hidden event alongside processing. It is the Oxford visitor question in silicon.

Debates & Critiques

Critics of Ryle, including Thomas Nagel, have argued that some apparent category mistakes are genuine ontological questions in disguise — that consciousness in particular may name something over and above behavioral dispositions, and that dismissing the question as a category mistake is itself a philosophical overreach. The Ryle volume acknowledges this tension in its chapter on honest limits, conceding that the framework is strongest where behavior is the issue and weaker where qualia are.