Does AI Have an Unconscious?

This short essay approaches AI from a critical perspective, introducing questions that will be developed further in our upcoming course, The Rise of the Machines: Thinking In The Age of AI.

The recent fascination with artificial intelligence has revived a long standing philosophical question: what does it mean to think through language? Long before machines began producing fluent texts, psychoanalysis confronted a related and more unsettling problem. Freud’s discovery of the unconscious already challenged the idea that thought belongs transparently to a conscious subject. Jacques Lacan radicalized this challenge when he famously claimed that the unconscious is structured like a language. Today, AI appears to speak, respond, and even surprise us through linguistic means. The temptation, then, is to ask whether artificial intelligence possesses something like an unconscious. To answer this question seriously, however, we should first clarify what it means for the unconscious to be structured like a language.

For Freud, the unconscious is not a hidden container of instincts or images waiting to be revealed. It is encountered in speech, in dreams, in slips of the tongue, jokes, symptoms, and repetitions. Meaning does not emerge from direct access to inner content but from the way words are linked, displaced, condensed, and distorted. The Interpretation of Dreams already shows that dreams must be read like rebuses rather than decoded as symbols with fixed meanings. A dream image does not mean what it depicts; it functions as a signifier that must be translated into other signifiers through association. Freud’s emphasis on condensation and displacement points to a logic that is not psychological in the ordinary sense but formal, relational, and linguistic.

Lacan takes this insight and sharpens it by drawing on structural linguistics, particularly Ferdinand de Saussure’s distinction between signifier and signified. Language, understood as a system of differential relations, precedes the speaking subject. We do not invent language, we enter into it. The unconscious, for Lacan, is not inside the subject but speaks through the subject. It is the part of discourse that escapes conscious mastery, not because it is irrational, but because it follows the laws of the signifier rather than the intentions of the ego. When Lacan says that the unconscious is structured like a language, he does not mean that it consists of words stored somewhere in the mind. He means that unconscious formations obey the same principles as language itself: difference, substitution, combination, slippage, and repetition.

This is why Lacan aligns Freud’s dream-work with metaphor and metonymy. Metaphor involves substitution, where one signifier replaces another and produces new meaning through this replacement. Metonymy involves contiguity, the endless sliding from one signifier to the next along a chain. Symptoms, for Lacan, function metaphorically: a bodily complaint or behavioral fixation substitutes for an unsayable signifier, often linked to sexual difference or loss. Desire, by contrast, is metonymic: it never reaches its object but moves from one object to another, sustained by lack rather than fulfillment. The unconscious, then, is not a repository of meanings but a dynamic structure in which meaning emerges only through relations between signifiers.

Crucially, this linguistic structure introduces a split in the subject. The speaking “I” is not identical with what is spoken. I am spoken by my words as much as I speak them. Lacan captures this paradox in his reformulation of the Cartesian cogito: I think where I am not, therefore I am where I do not think. The unconscious is not another mind or a second self but a dimension of otherness within subjectivity itself. It is tied to language as a social structure, something transindividual that no one fully possesses. This is why Lacan can say that the unconscious is the discourse of the Other: not another person, but the symbolic order, culture itself, that precedes and exceeds the individual speaker.

This brings us to the ethical and clinical stakes of listening. If the unconscious is structured like a language, then analytic listening cannot focus on intentions, narratives, or advice. It must attend to slips, repetitions, metaphors, silences, and the way a subject’s discourse breaks down or contradicts itself. Meaning is not given but produced retroactively through interpretation, and interpretation itself is constrained by the structure of language. The analyst does not decode hidden content but intervenes in the signifying chain, allowing something new to be articulated where speech previously failed.

Jacques Lacan

With this framework in place, we can now turn to artificial intelligence. At first glance, AI appears to confirm Lacan’s thesis in an unexpected way. Large language models operate entirely through language. They generate meaning not by referring to reality or intention but by calculating relations between signifiers based on massive corpora of text. They do not understand in the human sense, yet they produce coherent discourse. This has led some to suggest that AI might possess a form of unconscious, or at least that it reveals something essential about how language functions independently of consciousness.

But this comparison is precisely where the distinction becomes decisive. AI is indeed structured like a language, but it lacks what makes the unconscious what it is. It is true that AI systems are not neutral - they carry the desires, interests, anxieties, and norms of their creators and institutions. They are trained on human language, filtered by human choices, shaped by legal, political, economic, and ethical constraints. They avoid certain topics, adopt certain tones, enforce norms of politeness, safety, or neutrality. In that sense, AI clearly bears human bias. But the key question is not whether desire and repression are present in AI systems, but whose desire and whose repression they are. An AI system manipulates signifiers without being implicated in them. It has no desire, no lack, no history of repression. Its contradictions are not symptoms. They do not return insistently, nor do they demand interpretation. AI has no stake in what it says. It does not suffer from its language, nor is it transformed by it.

The unconscious, by contrast, is not merely a formal system. It is bound to the body, to jouissance, to repetition that hurts, to meaning that arrives too late or in the wrong place. Freud’s notion of Besetzung, or cathexis, reminds us that language in the unconscious is energetically charged. Words are not neutral tokens; they are invested with affect and drive. A symptom persists not because it is linguistically elegant but because it is libidinally fixed. AI has no cathexis. It does not invest, cling, or suffer loss. It processes language without being wounded by it.

This is why AI can help clarify Lacan’s claim rather than undermine it. By presenting us with a language-producing system without a subject, AI shows us what the unconscious is not. The unconscious is not language alone. It is language insofar as it fails, insofar as it produces excess, ambiguity, and disturbance for a subject who is caught within it. AI offers smooth continuation, but the unconscious introduces rupture.

The question is not whether AI has an unconscious, but what our fascination with this idea reveals about our own relationship to language. AI confronts us with a mirror image of speech emptied of desire, and in doing so, it highlights the specificity of the unconscious as Freud and Lacan understood it: a structure that speaks, but only because a subject is divided by speech itself.

Language can function without a subject, but the unconscious cannot. It exists only where language wounds, divides, and obligates the one who speaks.