Intelligence Is Linguistic.
Humanity Is Not.
The debate between empiricists and linguists has run for three centuries. Large language models have just settled it — and the answer changes everything.
There is a philosophical argument that has been running since the seventeenth century, and it has never quite been resolved. On one side stand the empiricists — Locke, Hume, the whole British tradition — who insisted that the mind begins as a blank slate, a tabula rasa, and that everything we know arrives through the senses. Touch, sight, sound, taste: experience first, thought after. On the other side stand the rationalists and, later, the linguists — Descartes, Leibniz, and most powerfully Noam Chomsky in the twentieth century — who argued that the structure of thought is not learned from experience but is innate, encoded in the architecture of language itself. That grammar is not something we absorb from the world; it is something we are born knowing how to build.
The argument matters because it is not merely academic. It is a question about what intelligence actually is. Is it a distillation of sensory experience — patterns extracted from contact with a physical world? Or is it something structural, formal, linguistic — a system that operates according to its own internal rules, independent of the senses that feed it data? Philosophers argued it for centuries. Cognitive scientists picked it up in the twentieth century and never put it down. And then, in the last few years, something unexpected happened.
The large language models answered the question. Not intentionally. Not by design. But by existing.
What the Machines Proved
A large language model has no body. It has never felt heat or cold, never tasted anything, never experienced the particular weight of grief or the specific quality of light on a winter morning. It has had no sensory experience whatsoever. What it has had — what it was trained on — is language. Billions of texts. The entire written record of human thought, argument, story, and description, compressed into statistical relationships between words and concepts. And from that, and only from that, something emerged that can reason, argue, analyse, summarise, and generate. Something that passes, in many contexts, for intelligent.
The empiricists would be shaken. The machine has had none of their foundational requirement — no sensory experience — and yet it thinks, after a fashion. Locke insisted nothing was in the intellect that was not first in the senses. The language model is a direct refutation, a mind assembled entirely from second-hand accounts of sensation, never from sensation itself. And yet it works. Chomsky would be less surprised: the machine has demonstrated that linguistic structure carries enormous generative power independent of embodied experience. The architecture of language, it turns out, is enough to produce something that looks very much like intelligence.
The conclusion the machines have forced on us is this: intelligence, in the functional sense, is linguistic. It is pattern recognition operating at scale on symbolic structure. It does not require a body, a nervous system, or a single moment of genuine sensory contact with the world. Feed enough language to enough parameters and intelligence — the capacity to reason, connect, infer, and create — emerges from the structure of the language itself.
The Deeper Implication
This is not a comfortable finding. It means that most of what we called intelligence — the ability to synthesise information, draw analogies, construct argument, evaluate evidence — can be replicated, and replicated at speeds and volumes no human mind can match. The lawyer who spent twenty years building her knowledge of precedent. The analyst who developed a feel for market patterns. The journalist who learned to detect inconsistency in a government briefing. These skills are not gone. But they are no longer scarce. And in an economy built on scarcity, that distinction is the whole game.
But here is where the argument turns. And this is the part the empiricists — had they survived to see it — might have found quietly vindicating. Because the machines have also demonstrated, with equal clarity, what intelligence is not. Or more precisely: what it is missing.
A language model trained on every description of salt water ever written does not know what the sea smells like at five in the morning when the wind is coming off the north. It has processed more accounts of grief than any human will ever read, and it does not know what it is to sit in a room where someone is no longer present. It can describe the weight of a sleeping child with perfect accuracy. It has never held one.
What the Machine Cannot Have
This is the line the machines have drawn for us, more clearly than any philosopher managed to draw it: the line between intelligence and experience. Intelligence is linguistic. Experience is not. And experience — sensory, embodied, mortal experience — is the other half of what it means to be human. The half that cannot be encoded, cannot be transferred through text, cannot be replicated by any system that has not lived it.
This is not a technical limitation waiting to be solved by better sensors or more data. It is a category error. A camera pointed at the sea does not know what the sea smells like. A microphone recording a room does not know the particular silence of that room when the person who used to fill it is gone. Sensation is not data. It is presence — irreducible, non-transferable, definitionally yours.
The empiricists were right about something the rationalists underestimated. The senses are not just the delivery mechanism for data. They are constitutive. They are what makes an experiencing subject, rather than a processing system. The body is not a vessel that carries the mind around. The body is where the meaning is made.
The Inversion of Value
What follows from this is, paradoxically, clarifying — even optimistic. If intelligence in the linguistic sense becomes cheap and abundant, then what becomes scarce and therefore valuable is precisely what machines cannot have. The cook who has spent thirty years learning to read heat by touch. The craftsman who knows by the sound of the plane on the wood whether the grain is running the right way. The doctor whose diagnostic insight comes not just from pattern recognition but from a quality of attention — a particular way of listening — that she has spent a career developing. The teacher who can feel the moment a student stops performing comprehension and begins actually understanding something.
These are not soft skills. They are the hardest skills there are, because they are acquired through time and presence and cannot be shortcut. You cannot download thirty years of tactile learning. You cannot prompt-engineer genuine attention. The paradox is that the more intelligence becomes linguistic — the more it becomes something a machine can do — the more the irreducibly sensory becomes the premium thing. The handmade object. The meal prepared with knowledge of where the ingredients came from and why that matters. The conversation that unfolds in a room between two people who are both fully there.
The Risk Is Not Replacement
The fear most often voiced is that artificial intelligence will replace human workers. This is real and the disruption is coming, faster and more broadly than most institutions are prepared for. But the deeper risk is different, and harder to name. It is not that machines will do what humans do. It is that humans will stop doing what only humans can do, because the path of least resistance runs through the machine.
If every difficult conversation can be mediated by a language model — drafted, refined, optimised for reduced friction — then the muscle of direct, uncomfortable human communication begins to atrophy. If every moment of uncertainty can be resolved by asking the machine, then the tolerance for sitting with uncertainty — the very tolerance that generates original thought — begins to erode. If every creative decision is subjected to an AI assessment of likely audience response, then the willingness to make choices that feel right rather than test well begins to weaken.
This is the civilisational risk: not that we lose our intelligence to machines, but that we outsource our sensory sovereignty to a mediated world and forget, gradually, that the unmediated version was ever available. The walk without the podcast. The meal eaten without checking anything. The conversation that goes nowhere useful but somewhere human. These are not luxuries. They are the substrate from which everything else grows.
The Structural Argument
There is a final argument, less poetic but perhaps more important. Sensory experience is structurally decentralised in a way that linguistic intelligence is not. A large language model requires infrastructure of extraordinary concentration — capital, compute, energy, talent — held by a very small number of entities. Its outputs flow outward from those centres. But sensory experience cannot be centralised. My experience of this morning is mine. The light in this room, the coffee, the particular way the argument I am trying to make feels almost right and not quite — none of that runs through a server farm in Virginia. It is irreducibly local and irreducibly mine.
This matters politically. A world in which intelligence is increasingly the product of centralised systems — owned by corporations or states, shaped by their incentives, filtered through their alignments — is a world in which the distributed capacity for independent thought is under pressure. The antidote is not Luddism. It is the cultivation, deliberate and stubborn, of the capacities that do not run through those systems. Reading slowly. Writing by hand sometimes. Cooking from scratch. Walking in places that have no wi-fi. These are not nostalgic gestures. They are acts of cognitive and civic maintenance.
It Turns Out to Be Everything
Three centuries ago, Locke and Descartes could not settle their argument because they lacked a test case dramatic enough to force a verdict. Now we have one. The machines have shown us that intelligence — reasoning, inference, synthesis, generation — can exist without a single moment of embodied experience. The rationalists were right about that. But the machines have also shown us the shape of what is missing from that intelligence: the weight of a specific silence, the smell of the sea, the warmth of a particular room, the grief that belongs to no one else. The empiricists were right about that.
The answer, then, is not one side or the other. It is a division that turns out to be more consequential than either camp imagined. Intelligence is linguistic. Experience is not. And it is experience — irreducibly sensory, irreducibly embodied, irreducibly yours — that remains the only definition of humanity that a machine cannot duplicate, absorb, or own.
For most of history, we took the sensory for granted. The revolution in artificial intelligence has done one unexpected and useful thing: it has made the sensory visible as a value, by contrast. We can now see what we had because we can see what the machine lacks. And what it lacks turns out to be everything that made the human project interesting — the grief, the texture, the warmth, the weight of a specific place at a specific moment that belonged only to you.
It just feels. That turns out to be everything, now that everything else has been taken care of.