The Uncanny
DESCRIPTION: Freud’s ‘The Uncanny’ explains why the almost-perfect AI face repels us, whilst the obviously robotic one leaves us cold—a psychoanalytical interpretation of the Uncanny Valley in the age of deepfakes and generative video AI.
The Uncanny: Why the AI face sends shivers down our spines
A tin robot leaves us cold, but an almost human AI face sends a shiver down our spines.
What it’s all about:
· Sigmund Freud’s 1919 essay ‘The Uncanny’
· Research into the Uncanny Valley and
· generative video AI systems of 2026.
The question is why an approximation of the human form is more frightening than outright strangeness.
What did Freud mean by the uncanny?
Freud begins his essay with a linguistic observation. The German word ‘unheimlich’ contains its opposite. ‘Heimlich’ refers to the familiar, the domestic, the secure, and at the same time to the hidden, the secret. At a certain point, the ‘heimlich’ becomes the ‘unheimlich’. Freud quotes Schelling: the ‘unheimlich’ is everything that should have remained a secret but has nevertheless come to light.
It is from this relationship that Freud derives his thesis. The uncanny is a kind of dread that stems from something long known and familiar. The shudder comes from close at hand, from the recognition of something that had been repressed. Thus, the uncanny differs from the mere fear of the new. It is a return of the repressed in a familiar guise.
Why is the familiar more uncanny than the unfamiliar?
We quickly categorise a recognisably alien being. It is different; it does not belong to us. We keep our distance, and that is that. The uncanny works differently, because it presents itself as familiar whilst simultaneously creating a rift in that familiarity. It is this rift that causes fear, for it calls into question an order on which we had relied.
Freud collects examples in which this very boundary disappears: the doubt as to whether something that appears to be alive is actually animated, or whether something lifeless might in fact be alive. Wax figures, automatons and dolls trigger this effect. The confusion concerns a distinction we took for granted: that between the living and the dead. Where the boundary between them becomes blurred, the familiar becomes threatening.
What does the automaton have to do with fear?
Before Freud, the psychiatrist Ernst Jentsch had attributed the uncanny to uncertainty about what is animate. His example was the doll that appears human. Freud takes up this observation and interprets it more deeply. In E. T. A. Hoffmann’s story ‘The Sandman’, the hero falls in love with Olimpia, who turns out to be a machine. For Freud, however, the horror runs deeper than mere deception.
The automaton touches upon an early emotional layer. As children, we used to think of dolls and soft toys as being alive, or at least often wished they were. The animated automaton reawakens this old notion, which we had long since forgotten. The karakuri automaton of Japanese tradition and the mechanical automata of 18th-century European salons strike the same chord: they move as if they had a purpose, and yet we know at the very same moment that there is none behind it.
What is the Uncanny Valley?
In 1970, the robotics engineer Masahiro Mori formulated a curve which he called the ‘uncanny valley’. The more human-like a figure becomes, the more likeable it appears, up to a point just short of perfect likeness. At that point, acceptance plummets. The almost-perfect replica evokes revulsion, whereas the crude replica still arouses affection. After that, when the figures become completely indistinguishable, the curve rises again.
Mori’s Valley describes in technical terms what Freud had conceptualised psychoanalytically. The abyss lies precisely at the boundary where we can no longer say with certainty whether we are looking at an object or a being. Robotics recognises the factors that trigger this collapse in familiarity: a gaze that stares into the void, facial expressions with incorrect timing, and skin that is too smooth. Minor deviations from the familiar carry greater weight than gross deviations from the unfamiliar.
Why this valley exists at all remains a matter of debate to this day. One explanation points to the avoidance of disease: an almost human face with minor flaws is reminiscent of signs of disease or death, and disgust keeps us at a distance. A second explanation cites mate selection, in which subtle facial deviations are interpreted as an indication of poor genetic makeup. A third emphasises expectation: as soon as something is labelled as human, we scrutinise it against the strict standards we apply to humans, and every flaw stands out. All three interpretations converge on Freud’s fundamental idea that growing familiarity with the familiar lowers the threshold for horror.
Why does the almost perfect AI face repel us?
An AI-generated face today can display pores, fine lines and glints of light, yet still evoke a sense of falseness. The reason lies in the same blurring of boundaries. We are constantly searching the other person’s face to see if someone is looking back at us. The AI face provides many signals suggesting this, but none are intentional. This emptiness behind the perfect surface is what sends a shiver down our spines.
The ELIZA effect from the early days of computing illustrates the other side of this same tendency. People attributed understanding even to a simple chat programme simply because it mimicked the form of conversation. We are hardwired to assume that a human being lies behind human signs. With an AI face, this expectation works against us. The closer the replica comes to the real thing, the more insistently our ancient perceptual mechanisms signal that something is amiss with its sense of life.
What will happen in 2026 with AI video and deepfakes?
By 2026, generative video AI will have reached a stage where many clips pass for the real thing at first glance. Experts say that the technology is beginning to cross the uncanny valley. It has not yet been fully overcome, however. Attentive viewers will still notice the drifting gaze, the lips lagging behind the voice by fractions of a second, and the weightless movement.
With deepfakes, a second layer comes into play. Here, it is not just any face that is generated, but that of a specific person we know. The familiar face, suddenly saying things it has never said, is uncanny in the full Freudian sense. A familiar face returns, distorted, and undermines our trust in our own perception. The post on deepfake detection on this blog addresses the practical side of this uncertainty.
The sheer volume exacerbates the situation. Platforms are flooded with AI-generated clips that appear only half-real – a stream of images with no provenance, pure AI rubbish. A previous post on this blog about this phenomenon, ‘ ’, describes how this mass of semi-real content desensitises our perception. The more commonplace the sight becomes, the fainter the shudder, and the harder it is to distinguish between a real recording and a fabricated one.
Is that shudder a flaw or a defence mechanism?
The reflex in the face of an AI face seems like a malfunction, a remnant from the time before screens. Yet it serves a purpose. The shudder marks a boundary that is important to us – the one between what a counterpart actually possesses and what it merely simulates. Where technology blurs this boundary, a bodily sensation keeps it open and signals doubt.
This warning loses its force the more accustomed we become to the images. Mori’s curve ends in complete indistinguishability, and that is where the shudder falls silent. That is precisely what makes the current phase so precarious. As long as the AI face still sends a shudder down our spines, an old defence mechanism is at work. When it falls silent, the signal that has hitherto helped us distinguish the simulated from the living is lost.
What does the doppelgänger reveal about the self?
Freud devotes a separate section to the doppelgänger. Originally, he writes with Otto Rank, the double was an insurance against death, a second version of the ego intended to live on. This protective mechanism later becomes a spectre. The doppelgänger we encounter reminds us of our repressed fear of death and of the parts of the ego that we have split off.
The AI face and the deepfake are doppelgängers in a technological guise. The generated image of a person demonstrates that their appearance can be copied – their expression, their voice, their gestures. The copy claims that the original is no longer needed. This interchangeability strikes at a narcissistic wound that was already a concern in Freud’s time, and which the post about Sora on this blog takes up.
Freud refers to Romantic literature, in which the double appears as a harbinger of doom. The 1913 German horror film *The Student of Prague* illustrated this motif for him: a man sells his reflection, which takes on a life of its own and pursues him. The severed image acts against the person from whom it originated. It is precisely this structure that the deepfake replicates by other means, when the copied face begins to take on a life of its own and threatens the original’s reputation.
Why does the repressed return in the deepfake?
For Freud, the uncanny is always a return. Something that had been repressed from consciousness reappears in a new form, bringing its old terror with it. The deepfake stages this structure using technical means. It takes the familiar – a face or voice known to – and makes it reappear in the wrong place.
The anxiety this creates extends beyond the individual manipulated clip. It concerns the reliability of the signs we use to distinguish one person from another. When the face is no longer proof of presence, and the voice is no longer proof of authorship, an old uncertainty returns – one that culture had painstakingly kept in check. The uncanniness of AI lies in this return of doubt regarding the authenticity of the other.
What does this uncanniness mean for our present day?
A hundred years on, Freud’s 1919 essay reads like a foreshadowing of the screen-based world. He describes a horror that comes from close at hand, from the familiar that turns upside down. It is precisely this experience that generative AI produces in everyday life today, repeatedly and cheaply. The uncanny valley has evolved from a curiosity of robotics into an everyday perception.
Those who take this shudder seriously gain access to their own scepticism towards AI, one that does not give way to panic. The reflex triggered by a false face is a form of knowledge that the body possesses more readily than the mind. To understand it is to consciously maintain the boundary between the animate and the simulated, whilst technology blurs it. The uncanny remains a reliable compass in this regard, because it always strikes where the human is imitated and yet simultaneously missed.
The key points in brief
• Freud defines the uncanny as the return of something familiar and repressed in an alien form.
• The strongest trigger is the blurred boundary between the living and the dead, evident in dolls, automatons and waxworks.
• Masahiro Mori’s ‘Uncanny Valley’ describes the same plunge in acceptance just before perfect human likeness is achieved.
• The AI face conveys all the signals of an interlocutor, yet none of them is genuine; this void is what induces the shudder.
• The deepfake intensifies the effect because a familiar face returns in a distorted form, undermining one’s own perception.
• The shudder acts as a safeguard that keeps an important boundary open; it subsides as soon as the images become indistinguishable.
Related:
DESCRIPTION: Freud’s ‘The Uncanny’ explains why the almost-perfect AI face repels us, whilst the obviously robotic one leaves us cold—a psychoanalytical interpretation of the Uncanny Valley in the age of deepfakes and generative video AI.
The Uncanny: Why the AI face sends shivers down our spines
A tin robot leaves us cold, but an almost human AI face sends a shiver down our spines.
What it’s all about:
· Sigmund Freud’s 1919 essay ‘The Uncanny’
· Research into the Uncanny Valley and
· generative video AI systems of 2026.
The question is why an approximation of the human form is more frightening than outright strangeness.
What did Freud mean by the uncanny?
Freud begins his essay with a linguistic observation. The German word ‘unheimlich’ contains its opposite. ‘Heimlich’ refers to the familiar, the domestic, the secure, and at the same time to the hidden, the secret. At a certain point, the ‘heimlich’ becomes the ‘unheimlich’. Freud quotes Schelling: the ‘unheimlich’ is everything that should have remained a secret but has nevertheless come to light.
It is from this relationship that Freud derives his thesis. The uncanny is a kind of dread that stems from something long known and familiar. The shudder comes from close at hand, from the recognition of something that had been repressed. Thus, the uncanny differs from the mere fear of the new. It is a return of the repressed in a familiar guise.
Why is the familiar more uncanny than the unfamiliar?
We quickly categorise a recognisably alien being. It is different; it does not belong to us. We keep our distance, and that is that. The uncanny works differently, because it presents itself as familiar whilst simultaneously creating a rift in that familiarity. It is this rift that causes fear, for it calls into question an order on which we had relied.
Freud collects examples in which this very boundary disappears: the doubt as to whether something that appears to be alive is actually animated, or whether something lifeless might in fact be alive. Wax figures, automatons and dolls trigger this effect. The confusion concerns a distinction we took for granted: that between the living and the dead. Where the boundary between them becomes blurred, the familiar becomes threatening.
What does the automaton have to do with fear?
Before Freud, the psychiatrist Ernst Jentsch had attributed the uncanny to uncertainty about what is animate. His example was the doll that appears human. Freud takes up this observation and interprets it more deeply. In E. T. A. Hoffmann’s story ‘The Sandman’, the hero falls in love with Olimpia, who turns out to be a machine. For Freud, however, the horror runs deeper than mere deception.
The automaton touches upon an early emotional layer. As children, we used to think of dolls and soft toys as being alive, or at least often wished they were. The animated automaton reawakens this old notion, which we had long since forgotten. The karakuri automaton of Japanese tradition and the mechanical automata of 18th-century European salons strike the same chord: they move as if they had a purpose, and yet we know at the very same moment that there is none behind it.
What is the Uncanny Valley?
In 1970, the robotics engineer Masahiro Mori formulated a curve which he called the ‘uncanny valley’. The more human-like a figure becomes, the more likeable it appears, up to a point just short of perfect likeness. At that point, acceptance plummets. The almost-perfect replica evokes revulsion, whereas the crude replica still arouses affection. After that, when the figures become completely indistinguishable, the curve rises again.
Mori’s Valley describes in technical terms what Freud had conceptualised psychoanalytically. The abyss lies precisely at the boundary where we can no longer say with certainty whether we are looking at an object or a being. Robotics recognises the factors that trigger this collapse in familiarity: a gaze that stares into the void, facial expressions with incorrect timing, and skin that is too smooth. Minor deviations from the familiar carry greater weight than gross deviations from the unfamiliar.
Why this valley exists at all remains a matter of debate to this day. One explanation points to the avoidance of disease: an almost human face with minor flaws is reminiscent of signs of disease or death, and disgust keeps us at a distance. A second explanation cites mate selection, in which subtle facial deviations are interpreted as an indication of poor genetic makeup. A third emphasises expectation: as soon as something is labelled as human, we scrutinise it against the strict standards we apply to humans, and every flaw stands out. All three interpretations converge on Freud’s fundamental idea that growing familiarity with the familiar lowers the threshold for horror.
Why does the almost perfect AI face repel us?
An AI-generated face today can display pores, fine lines and glints of light, yet still evoke a sense of falseness. The reason lies in the same blurring of boundaries. We are constantly searching the other person’s face to see if someone is looking back at us. The AI face provides many signals suggesting this, but none are intentional. This emptiness behind the perfect surface is what sends a shiver down our spines.
The ELIZA effect from the early days of computing illustrates the other side of this same tendency. People attributed understanding even to a simple chat programme simply because it mimicked the form of conversation. We are hardwired to assume that a human being lies behind human signs. With an AI face, this expectation works against us. The closer the replica comes to the real thing, the more insistently our ancient perceptual mechanisms signal that something is amiss with its sense of life.
What will happen in 2026 with AI video and deepfakes?
By 2026, generative video AI will have reached a stage where many clips pass for the real thing at first glance. Experts say that the technology is beginning to cross the uncanny valley. It has not yet been fully overcome, however. Attentive viewers will still notice the drifting gaze, the lips lagging behind the voice by fractions of a second, and the weightless movement.
With deepfakes, a second layer comes into play. Here, it is not just any face that is generated, but that of a specific person we know. The familiar face, suddenly saying things it has never said, is uncanny in the full Freudian sense. A familiar face returns, distorted, and undermines our trust in our own perception. The post on deepfake detection on this blog addresses the practical side of this uncertainty.
The sheer volume exacerbates the situation. Platforms are flooded with AI-generated clips that appear only half-real – a stream of images with no provenance, pure AI rubbish. A previous post on this blog about this phenomenon, ‘ ’, describes how this mass of semi-real content desensitises our perception. The more commonplace the sight becomes, the fainter the shudder, and the harder it is to distinguish between a real recording and a fabricated one.
Is that shudder a flaw or a defence mechanism?
The reflex in the face of an AI face seems like a malfunction, a remnant from the time before screens. Yet it serves a purpose. The shudder marks a boundary that is important to us – the one between what a counterpart actually possesses and what it merely simulates. Where technology blurs this boundary, a bodily sensation keeps it open and signals doubt.
This warning loses its force the more accustomed we become to the images. Mori’s curve ends in complete indistinguishability, and that is where the shudder falls silent. That is precisely what makes the current phase so precarious. As long as the AI face still sends a shudder down our spines, an old defence mechanism is at work. When it falls silent, the signal that has hitherto helped us distinguish the simulated from the living is lost.
What does the doppelgänger reveal about the self?
Freud devotes a separate section to the doppelgänger. Originally, he writes with Otto Rank, the double was an insurance against death, a second version of the ego intended to live on. This protective mechanism later becomes a spectre. The doppelgänger we encounter reminds us of our repressed fear of death and of the parts of the ego that we have split off.
The AI face and the deepfake are doppelgängers in a technological guise. The generated image of a person demonstrates that their appearance can be copied – their expression, their voice, their gestures. The copy claims that the original is no longer needed. This interchangeability strikes at a narcissistic wound that was already a concern in Freud’s time, and which the post about Sora on this blog takes up.
Freud refers to Romantic literature, in which the double appears as a harbinger of doom. The 1913 German horror film *The Student of Prague* illustrated this motif for him: a man sells his reflection, which takes on a life of its own and pursues him. The severed image acts against the person from whom it originated. It is precisely this structure that the deepfake replicates by other means, when the copied face begins to take on a life of its own and threatens the original’s reputation.
Why does the repressed return in the deepfake?
For Freud, the uncanny is always a return. Something that had been repressed from consciousness reappears in a new form, bringing its old terror with it. The deepfake stages this structure using technical means. It takes the familiar – a face or voice known to – and makes it reappear in the wrong place.
The anxiety this creates extends beyond the individual manipulated clip. It concerns the reliability of the signs we use to distinguish one person from another. When the face is no longer proof of presence, and the voice is no longer proof of authorship, an old uncertainty returns – one that culture had painstakingly kept in check. The uncanniness of AI lies in this return of doubt regarding the authenticity of the other.
What does this uncanniness mean for our present day?
A hundred years on, Freud’s 1919 essay reads like a foreshadowing of the screen-based world. He describes a horror that comes from close at hand, from the familiar that turns upside down. It is precisely this experience that generative AI produces in everyday life today, repeatedly and cheaply. The uncanny valley has evolved from a curiosity of robotics into an everyday perception.
Those who take this shudder seriously gain access to their own scepticism towards AI, one that does not give way to panic. The reflex triggered by a false face is a form of knowledge that the body possesses more readily than the mind. To understand it is to consciously maintain the boundary between the animate and the simulated, whilst technology blurs it. The uncanny remains a reliable compass in this regard, because it always strikes where the human is imitated and yet simultaneously missed.
The key points in brief
• Freud defines the uncanny as the return of something familiar and repressed in an alien form.
• The strongest trigger is the blurred boundary between the living and the dead, evident in dolls, automatons and waxworks.
• Masahiro Mori’s ‘Uncanny Valley’ describes the same plunge in acceptance just before perfect human likeness is achieved.
• The AI face conveys all the signals of an interlocutor, yet none of them is genuine; this void is what induces the shudder.
• The deepfake intensifies the effect because a familiar face returns in a distorted form, undermining one’s own perception.
• The shudder acts as a safeguard that keeps an important boundary open; it subsides as soon as the images become indistinguishable.
Related: