Gilbert Ryle

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI argues that the machine is finally a mirror—that the most important questions it poses are questions about consciousness, judgment, and what it means to understand, not questions about hardware and benchmarks. Ryle enters this inquiry as the philosopher who showed that these very questions are likely to be malformed. When one party insists the model “really understands”—that behind the fluent text there must be genuine comprehension, a grasping of meaning—and another insists the model “understands nothing”—that behind the text there is only statistics, an empty mechanism—both parties share the same Cartesian assumption: that there is a fact about a hidden inner state, present or absent, full or empty, and that settling AI's status means peering behind the behavior to check. This is the ghost, resurrected in silicon. Ryle's move is not to take a side but to question whether the hunt is coherent.

His distinction between knowing how and knowing that cuts to the heart of what makes large language models strange. The model does not store and retrieve facts the way a database does. It has no table of propositions it looks up. What it has is a capacity—a knowing-how whose exercise is the production of appropriate continuations. It knows how to produce text that includes “Paris” after “the capital of France is”—a skill, a disposition, a competence—not a propositional belief that Paris is the capital of France. This is why hallucination is not a bug in a knowing-that system but a feature of a knowing-how system whose competence is the production of plausible text regardless of whether that text is true: the model can produce the rules of a game it cannot reliably play, state a fact in one breath and contradict it in the next.

The intellectualist legend—Ryle's name for the fallacy that intelligent practice derives from prior theoretical knowledge—is being played out in the AI transition in the domain of education. An educational system built on the premise that intelligence means possessing propositional knowledge has invested heavily in exactly the currency the machine now makes cheap. Ryle's regress argument shows that this investment was philosophically mistaken before AI made it economically disastrous: intelligent practice was never the application of a prior theory but the exercise of formed dispositions, and the machine has demonstrated this by achieving genuine knowing-how without ever possessing the knowing-that that the intellectualist legend insisted was foundational.

Ryle is also the philosopher of the limits that the machine throws into sharpest relief. His dispositional analysis of mind—the proposal that mental concepts describe tendencies to behave rather than inner episodes—takes the machine seriously in a way no other philosophical framework does: a trained neural network is, literally, nothing but a vast structure of dispositions, billions of numbers encoding tendencies to respond. But the machine also delivers the strongest available evidence that something in the neighborhood of felt experience is real and separate from behavioral disposition—that Ryle may have dissolved the ghost without honoring whatever is genuinely staged in the inner theater he cleared.

Origin

Ryle was born in Brighton in 1900, the son of a physician, and educated at Brighton College and the Queen's College, Oxford, where he read classics before turning to philosophy. He joined the Oxford faculty in the 1920s and spent his entire career there, becoming one of the central figures in what came to be called Ordinary Language philosophy—the movement that treated philosophical puzzles not as deep mysteries requiring speculative solutions but as confusions generated by the systematic misuse of language, to be dissolved by careful attention to how words actually work. His colleagues included J. L. Austin and, later, P. F. Strawson; his most important intellectual adversary was the entire tradition of Cartesian dualism that he regarded as having generated most of the mind-body problem.

The Concept of Mind appeared in 1949 to both acclaim and controversy. Its method was unusual: rather than arguing for a positive theory of mind, Ryle applied a kind of philosophical pathology, identifying the specific grammatical errors that had produced the dualist picture and showing how each of them dissolved under careful analysis. The category mistake—treating a concept as belonging to the wrong logical type, as when a visitor to Oxford asks where the University is after being shown all the colleges—was his central diagnostic tool. The ghost in the machine was his name for the result of systematically mis-categorizing mental concepts as names for inner events occurring in a private substance alongside the physical body.

Ryle served as editor of Mind, the leading philosophical journal, from 1947 to 1971, and his influence on British philosophy of the mid-twentieth century was pervasive. He died in 1976, thirty years before a program could hold a conversation. He never imagined a large language model. But the disease he diagnosed—the Cartesian grammar that makes us look for a hidden tenant behind the behavior, a ghost performing the understanding behind the linguistic output—is exactly the disease we now suffer in the machine's presence, because it was never a disease about machines. It was always a disease about how we talk, and how our talk traps us.

Key Ideas

The category mistake. The most expensive error in thinking about AI is not a mistake of fact but a mistake of type. We observe, correctly, that a model produces fluent and contextually apt text, and then we conclude—illegitimately—that it must therefore possess the inner mental life that, in a human being, would accompany such text. The category mistake is to treat “understanding” as the name of an inner occurrence whose presence or absence is a further fact beyond everything the system can do—like asking where the University is after being shown all the colleges. Ryle's tool does not hand victory to either side of the AI debate; it indicts the shared assumption that there is an inner mental thing to be present or absent behind the behavior at all.

Knowing how and knowing that. The distinction between practical competence exhibited in performance and propositional knowledge statable in a sentence is Ryle's most directly applicable idea for the AI moment. Large language models possess knowing-how in abundance—the skill of producing appropriate continuations—without possessing knowing-that in any robust sense. This explains both their uncanny competence at tasks that are themselves matters of skill (drafting, coding, translating) and their characteristic failures: they can be confidently, fluently wrong, because their competence is the skill of plausibility rather than the possession of truth.

The intellectualist legend and education. The persistent fallacy that intelligent practice derives from prior theoretical knowledge generates an infinite regress: applying a rule intelligently requires further rules about how to apply rules, and so on. Intelligent practice was always the ancestress of theory, not its application. AI has made the educational consequences of this fallacy acute: systems built to produce knowing-that in unlimited quantities have exposed the entire apparatus of propositional education as an investment in the currency the machine now makes cheap.

Dispositional analysis. To say a person is intelligent, knows French, believes it will rain—these are not reports of inner occurrences but attributions of dispositions: structured tendencies to behave in certain ways under certain conditions. The dispositional analysis of mind takes the machine seriously as a minded thing, because a trained neural network is literally nothing but dispositions. But it also reveals the machine's specific deficit: human dispositions are integrated into a single continuing perspective, anchored by a body to a world that pushes back, held together by stakes and consequences. The machine's dispositions are frozen at training and contradiction-tolerant, produced by a process whose formative power accrued to no one.

The ghost in the machine. Ryle coined this phrase as philosophical ridicule. The “ghost in the machine” was his name for the Cartesian picture of a non-physical mind inhabiting a physical body—a picture he regarded as not wrong in the way a scientific hypothesis can be wrong but confused in the way that the Oxford visitor's question about the University is confused. The phrase has had a strange afterlife: it became a band's name, a film's title, a cliché for anything eerie and disembodied. But Ryle coined it as a weapon, and the abusiveness was the point. He did not think there was a ghost. He thought “the ghost” was what you get when you take perfectly ordinary facts about people and mislocate them into an invisible inner chamber.