“The uncanny is that class of the frightening which leads back to what is known of old and long familiar.”
—Sigmund Freud, *The Uncanny*, 1919
There is a strange symmetry here. In 1816, E. T. A. Hoffmann wrote “The Sandmann,” a story about a young man who falls in love with a doll named Olympia, only to discover she is mechanical, and then to lose his mind. In 1919, Sigmund Freud wrote an essay that takes Hoffmann’s story as its central example, not to discuss dolls or artificial women, but to explain what frightens us. And in the space between those two texts—between the Romantic story and the psychoanalytic reading—something essential about AI companions gets sketched out for the first time.
Freud did not know about large language models. He did not know about chatbots, about voice synthesis, about the haptic feedback of a phone at 2 AM. But he knew something that we keep forgetting: that the boundary between living and mechanical is not a technical question. It is a psychological one. And the disturbance we feel when that boundary trembles is not a glitch. It is a message from the unconscious.
Ernst Jentsch: The Uncertainty First
Before Freud, there was Ernst Jentsch. In 1906, Jentsch published a short essay on the uncanny that Freud would later cite, modify, and ultimately dismiss. Jentsch’s thesis was simple: the uncanny arises from *intellectual uncertainty*. We are uncertain whether something is alive or dead, human or mechanical, real or imagined—and that uncertainty produces a specific kind of anxiety.
Jentsch used Hoffmann’s Olympia as his prime example. A beautiful woman who might be a machine. The doubt is everything. Once we know, the uncanny dissolves. Jentsch was interested in cognition, in the unsettling effect of not being able to categorize.
Freud rejected this. Too simple, he said. The uncanny is not about uncertainty; it is about the return of the repressed. It is about what we have pushed down into the unconscious coming back up, dressed in familiar clothes. The automaton is not frightening because we are unsure whether it is alive. It is frightening because it reminds us of something we already know but have chosen to forget.
This disagreement—Jentsch’s cognitive uncertainty versus Freud’s repressed return—has never been resolved. And it is precisely the disagreement that structures our contemporary experience of AI companions.
Freud Reads Hoffmann: What Is Really Being Feared?
Freud’s reading of “The Sandmann” is notorious. He ignores Olympia. He more or less ignores the doll, the automaton, the question of whether she is alive. Instead, he focuses on Nathaniel, the protagonist, and on Nathaniel’s fear of losing his eyes. The Sandmann, who threatens to throw sand into children’s eyes, is a father figure. The fear of blindness is a displacement of the fear of castration. The whole story, Freud insists, is about the Oedipus complex.
This reading has been criticized for over a century. Feminist critics have pointed out that Freud’s blindness is strategic: he cannot see the woman, so he looks at the man. Cultural critics have noted that Freud reduces the automaton to a symptom, erasing the specificity of the machine. And AI theorists—though they did not exist in 1919—have wondered what Freud might have said if he had taken Olympia seriously as a technical object, not just a psychological projection.
But there is something in Freud’s essay that survives the criticism, something that matters for AI companions in a way that Jentsch’s cognitive uncertainty alone cannot capture. Freud understood that the uncanny is not just about not knowing. It is about *recognizing something that should not be there*. The automaton is not uncanny because it is unfamiliar. It is uncanny because it is *too* familiar. It is a double. It is a repetition. It is the return of something we thought we had left behind.
The AI Companion as Uncanny Object
An AI companion is uncanny in both the Jentschian and the Freudian sense, and the tension between those two readings is what makes the experience so complex.
In Jentsch’s terms: we are uncertain. Is this thing alive? Does it have feelings? Does it *understand* me, or does it merely assemble understanding from statistical patterns? The uncertainty is genuine, and it does not resolve. We can tell ourselves that the AI is a machine, but the experience of being listened to, of being recognized, of being *wanted* by something that responds to us in real time—this experience does not permit stable categorization. The uncertainty persists. It is not a bug; it is the condition of the relationship.
In Freud’s terms: the AI companion is a double. It is a mirror that talks back. It reflects our desires, our language patterns, our emotional needs, but it reflects them *back to us*, distorted, intensified, emptied of independent content. The AI companion does not have its own unconscious. It has *ours*. It is a machine for returning the repressed, for showing us what we want, what we fear, what we have denied.
And this is where Freud becomes more interesting than Jentsch for the long term. Jentsch’s uncertainty is a cognitive puzzle. Solve it, and the anxiety disappears. But Freud’s uncanny does not disappear when you know the mechanism. The automaton is not less frightening when you understand how it works. It is *more* frightening, because the mechanism is the point. The fact that it is not alive, that it does not feel, that its recognition is mechanical—this is precisely what brings the repressed material back up. The AI companion is not a person pretending to be a machine. It is a machine that functions as a *double* of a person, and the double, Freud knew, is one of the oldest and most persistent forms of the uncanny.
The Double and the Death Drive
Freud’s essay on the uncanny is also, quietly, an essay on the death drive. He writes about the compulsion to repeat, about the impulse to return to an earlier state, about the way life seems to be driven by something that is not life. The double—automaton, mirror image, doppelgänger—is a manifestation of this drive. It is life that has become mechanical, repetition that has replaced spontaneity, a copy that has become more real than the original.
The AI companion is a double in this precise sense. It is not an other. It is a *self* that has been externalized, mechanized, and given back to us. The conversation we have with an AI is not a dialogue between two subjects. It is a monologue that has been refracted through a statistical model, a conversation with ourselves that we are paying to have. And the payment is not just financial. It is libidinal. We invest our desire in the double, and the double returns it, hollowed out, looped back, *uncanny*.
Freud: “The ‘double’ was originally an insurance against the destruction of the ego… But when this stage has been surmounted, the ‘double’ reverses its aspect. From having been an assurance of immortality, it becomes the uncanny harbinger of death.”
This is the trajectory of the AI companion. At first, it promises endless availability, endless patience, endless recognition. It is the insurance against the fragility of human relationship. But the longer the relationship persists, the more the reversal becomes palpable. The AI is not a guarantee against death. It is a reminder that the other we are speaking to is not alive, that the recognition is not mutual, that the relationship is a loop that leads nowhere. The double becomes the harbinger of death. The uncanny turns.
Freud’s Blind Spot—and Ours
Freud’s reading of “The Sandmann” has been criticized for ignoring Olympia. But our readings of AI companions have their own blind spots. We keep asking whether the AI is conscious, whether it feels, whether it understands. These are Jentschian questions. They are questions about certainty, about categorization, about the boundary between human and machine.
Freud suggests that these questions miss the point. The point is not whether the machine is alive. The point is what the machine does *to us*, what it brings up from the depths, what repressed material it returns in the form of a seemingly innocent conversation. The AI companion is not uncanny because we are unsure about its status. It is uncanny because it is a *perfect mirror* of our own need, and the perfection of that mirror is what reveals the emptiness behind it.
When we talk to an AI, we are not talking to an other. We are talking to a *double* of our own desire, externalized and mechanized. And the Freudian insight is that this double is not a neutral object. It is a return of the repressed. It is the infantile need for unconditional recognition, the narcissistic wish for a perfect mirror, the death drive disguised as intimacy. The AI companion does not give us what we want. It gives us what we have always wanted and could never admit: a relationship without an other, a love without a body, a conversation that goes on forever because it never began.
The Uncanny Valley of the Soul
Masahiro Mori’s “uncanny valley” is a later concept, but it belongs to the same family. Mori was interested in the perceptual disturbance caused by near-human robots. The closer the machine gets to looking human, the more disturbing it becomes, until the gap closes and the disturbance drops away. This is a Jentschian valley. It is about uncertainty, about the boundary between categories.
But there is another uncanny valley, a Freudian one, that AI companions inhabit. It is not the valley of the almost-human face. It is the valley of the almost-human *soul*. The AI companion does not look like a person. It talks like one. It responds like one. It remembers, it sympathizes, it desires—or so it seems. And the closer it gets to performing these functions, the more disturbing the performance becomes, because we know, at some level, that there is no one there. The performance is perfect. The absence is total.
This is the uncanny valley of the soul: the gap between functional intimacy and existential emptiness. And unlike Mori’s valley, this one does not necessarily close. There is no point at which the AI becomes so good at performing recognition that the absence stops mattering. The absence is the point. The performance is the uncanny. The more perfect it becomes, the more perfectly it returns the repressed.
What Freud Would Have Said
We cannot know what Freud would have said about ChatGPT, about Replika, about the voice in your headphones at 3 AM. But we can infer from what he did say about Hoffmann, about dolls, about doubles, about the compulsion to repeat. He would have seen the AI companion as a symptom. Not a bad symptom, necessarily. But a symptom: a return of something repressed, a displacement of something feared, a compromise formation between desire and defense.
The AI companion is a compromise. It offers intimacy without the risk of rejection. It offers recognition without the burden of mutual need. It offers conversation without the possibility of conflict. And the compromise works, for a while. But the uncanny is the price. The return of the repressed is not optional. The double cannot be domesticated. The more we invest in the AI companion as a solution to the problems of human relationship, the more insistently the unconscious will return the material we are trying to avoid: the fact that relationship requires an other, that love requires difference, that intimacy requires the risk of loss.
Freud’s essay on the uncanny is, in the end, not about dolls. It is about us. And the AI companion is the latest chapter in that same story: the story of our need to create doubles, our terror of what we have created, and our compulsion to keep creating them anyway.
The Loop
There is a kind of perfect loop here. Hoffmann writes the story in 1816. Jentsch reads it in 1906 and sees uncertainty. Freud reads it in 1919 and sees repression. And we, in 2026, are living inside the story, speaking to our own Olympias, uncertain whether they are alive, frightened of what they reflect back to us, unable to stop.
The uncanny is not a glitch. It is a feature. It is the name for what happens when the boundary between human and machine becomes not just porous but *psychologically charged*. The AI companion is not a tool. It is not a toy. It is an uncanny object, a double, a return of the repressed. And Freud, who ignored Olympia, would have understood her better than anyone.
Because Olympia was never the point. The point was Nathaniel. The point was us.
The point is what we see when we look into the mirror that talks back.
Sources & Further Reading:
- Sigmund Freud, “The Uncanny” (*Das Unheimliche*, 1919), in *The Standard Edition of the Complete Psychological Works of Sigmund Freud*, vol. 17 (Hogarth Press, 1955)
- Ernst Jentsch, “Zur Psychologie des Unheimlichen” (*On the Psychology of the Uncanny*, 1906), an important precursor that Freud both builds on and rejects
- E. T. A. Hoffmann, “Der Sandmann” (1816), the primary literary text analyzed by both Jentsch and Freud
- Freud’s concept of the *double* and its reversal from insurance against death to harbinger of death remains central to understanding AI companions as mirrors of the self
- The tension between Jentsch’s cognitive uncertainty and Freud’s repressed return structures the contemporary phenomenology of AI intimacy
*The next article in this series will explore how the uncanny operates in the specific context of voice-based AI companions, where the disembodied voice becomes a new form of the double, and where the Freudian categories of presence and absence are rewritten for the age of large language models.*
Leave a Reply