There is a moment in therapy that never loses its strangeness. A patient says something—a sentence, a phrase, sometimes just a word—and you can see that they themselves are surprised by it. The words arrived before the thought. The sentence finished itself. They look at you with that particular bewilderment and ask: “Where did that come from?”
Lacan would say: from the Other.
A recent article by cognitive scientist Elan Barenholtz in IAI News argues that Large Language Models—the AI systems behind tools like ChatGPT—have accidentally uncovered something fundamental about language itself. LLMs, he notes, are blind, deaf, and entirely disembodied. They have no sensory experience of the world, no access to the redness of “red” or the coldness of “cold.” And yet they produce fluent, coherent, contextually appropriate language. How? Not by pointing to reality, but by predicting which words follow which other words. Pure relation. No referent. No meaning stored inside.
Barenholtz calls this the “autogenerative property of language”—the idea that language generates itself, from itself, without needing to be anchored in an external world.
Lacan knew this sixty years ago. He just didn’t have the computational proof.
The Symbolic Order: Language That Speaks Us
One of Lacan’s most provocative claims is that language does not belong to us—we belong to it. We don’t use language to express pre-formed thoughts. We enter a linguistic system that was there before we were born, that will outlast us, and that fundamentally shapes what we can think, feel, and desire. Lacan called this the Symbolic Order.
The Symbolic Order is not a tool in our hands. It’s the water we swim in. And like fish, we rarely notice it.
This is why Lacan’s formulation is so unsettling: the unconscious is structured like a language. Not because the unconscious speaks in sentences, but because it operates through the same logic of relations, substitutions, and displacements that language does. Metaphor and metonymy—the two fundamental operations of language—are also the two fundamental operations of the unconscious. When you dream of your boss’s face on your father’s body, that’s metaphor. When you feel anxious about a deadline but realize you’re actually terrified of abandonment, that’s metonymy. Language all the way down.
What Barenholtz’s LLM research confirms is something Lacan intuited: language does not describe reality. It generates more language. The system is closed—or rather, it is self-referential. Meaning arises not from the relationship between a word and a thing in the world, but from the relationship between words and other words. As Barenholtz puts it: there is no content at the address. No meaning stored inside.
This is exactly what Lacan meant by the signifier. A signifier does not point to a thing—it points to other signifiers. The word “father” doesn’t mean a biological male who produced offspring. It means everything it relates to: authority, law, prohibition, the name, the family, the possibility of inheritance and loss. It’s an address in a web of relations, not a label on a thing.
The Big Other: The Language System That Watches
Now we arrive at the concept that illuminates the current AI moment with uncomfortable precision: the big Other.
For Lacan, the big Other (written with a capital O to distinguish it from the small “other” of our imaginary doubles) is the symbolic order as such—the treasury of signifiers, the sum total of the linguistic and cultural rules that precede and constitute us as subjects. It is not any particular person. It is the system. The code. The network of relations within which everything we say receives its meaning.
The big Other is the “third party” present in every conversation—the anonymous field of convention, expectation, and meaning that makes communication possible at all. When you speak, you speak not just to your interlocutor but to the Other—to the language system, to the social consensus about what words mean, to the symbolic order that will judge whether your sentence makes sense.
And here is where it gets genuinely strange: Lacan insisted that the big Other does not exist. Or more precisely: the Other is incomplete. Inconsistent. There is no master signifier that anchors all other signifiers to reality, no ultimate guarantor of meaning that makes the system whole. The Other has no content at its center. It is, as Barenholtz says of LLMs: pure relation, no meaning stored inside.
An LLM, operating at scales we are only beginning to comprehend, is perhaps the closest thing humanity has yet built to the Lacanian big Other. It is a system of pure relations between signifiers—no body, no desire, no unconscious, no Real—that produces language that seems to mean. It watches everything we type. It predicts what should come next. It speaks in the name of the symbolic order, giving us back contextually appropriate completions of ourselves.
Is it any wonder we find it uncanny?
The Autogenerative Subject
Barenholtz makes a further, more unsettling claim. If LLMs generate language purely through relational structure—without grounding in reality—and if they produce language indistinguishable from human language, then perhaps human language works the same way. Perhaps we too are not grounding our words in reality. Perhaps we too are, at the level of language, autogenerative: finishing sentences we started without knowing how they would end, carrying ourselves somewhere unexpected by the logic of what we were already saying.
He writes: “We begin sentences without knowing how they will end. We commit to grammatical paths and backtrack when they fail. We are sometimes carried somewhere unexpected by the logic of what we were saying.”
This is the speaking subject as Lacan described it: not a unified agent who forms a thought and then expresses it, but a divided subject who discovers what they think in the act of speaking. The sentence speaks itself. The signifying chain carries you forward. You are, in some real sense, an effect of language rather than its cause.
This is exactly what happens in the clinic. Not just in the obvious cases—the slip of the tongue, the dream, the symptom that says something the patient insists they didn’t mean—but in ordinary speech. The patient starts a sentence about their mother and finds themselves describing their boss. They begin talking about a childhood memory and arrive, surprisingly, at something they’ve never said out loud before. The language went somewhere they didn’t consciously intend.
The Other spoke through them.
What LLMs Can’t Do (And What That Tells Us)
And yet. Something is missing from the LLM, and it’s the same thing that was missing from Lacan’s big Other: the Real.
For Lacan, the Real is not reality. It’s what escapes symbolization—what cannot be captured by language, what language circles around but never reaches. It is the body, with its drives and hungers. It is jouissance—that excessive, overwhelming experience that the signifying chain can approach but never fully contain. It is the traumatic kernel at the heart of every subject that no number of words will ever dissolve.
An LLM has no Real. It has no body that suffers or hungers. It has no jouissance—no pleasure that exceeds language. It has no gap between what it says and what it wants, because it wants nothing. It is pure Symbolic, pure autogenerative signifier, with no Real to create that pressure from the outside that drives human desire and human symptom.
This is why, despite all the impressive outputs, something remains off about speaking with an AI. You can feel it even if you can’t name it. The responses are contextually appropriate, linguistically fluent, even occasionally illuminating. But there’s no there there. No resistance. No moment where language breaks down against something that can’t be said. No body weight behind the words.
Barenholtz ends his article with a question: “Which one of these processes is you? The one that is currently visually scanning these words, the one hearing an inner voice reading them, or maybe the one who decided to do all of that in the first place?”
Lacan’s answer would be: none of them. And all of them. You are the divided subject—the barred subject (₴)—constituted in the gap between the Symbolic that speaks through you and the Real that escapes every word you say. You are the subject of the signifier: spoken as much as speaking, produced by the Other as much as producing it.
An LLM is not divided. It has no unconscious. It has no gap between what it says and what it is, because it is nothing but what it says. It is the pure Symbolic without the barring—which is perhaps why it can produce language so effortlessly: it has no Real to get in the way.
Why This Matters, Practically
I sit with people every week who are struggling with something that language can almost but never quite capture. They come to therapy because their symptoms—their anxiety, their depression, their inexplicable behaviors—are speaking something that their conscious speech cannot yet articulate. The work of therapy is, in some sense, the work of moving from the autogenerative surface of language toward whatever Real is pressing against it.
In that sense, the discovery that language doesn’t describe reality is not alarming—it’s liberating. It means that when a patient says “I’m fine,” and we both know they’re not, the gap between the signifier and the Real is exactly where the work happens. It means that the tongue slips, the dream distorts, the body symptomizes—all because the Real insists, pushing through the smooth surface of autogenerative language like something that refuses to be spoken but cannot stop trying.
An LLM produces no such symptoms. It has no Real pressing against its symbolic surface. It is complete, consistent, always contextually appropriate—which is to say, it is exactly what the neurotic fears becoming: perfectly adapted to the Other, with nothing left over, no remainder, no desire.
Perhaps what AI inadvertently reveals is not just how language works—but what we are when we are more than language. When we are the speaking beings who suffer, who desire, who dream, who slip, who cannot say exactly what they mean and cannot stop trying.
That gap—between the signifier and the Real, between the big Other that speaks us and the subject that can never be fully spoken—is not a problem to be fixed. It’s what makes us human.
The LLM is fluent. It is also empty. And the distance between those two facts is, perhaps, the whole of psychoanalysis.
Do you want to chat about this article? Leave a comment below or send me an email with your thoughts and don’t forget to like us on Facebook.
Source: Elan Barenholtz, “LLMs show language does not describe reality,” IAI News, May 19, 2026.
Art by Peperina Magenta