“I have noticed that, in climbing toward the goal of making robots appear human, our affinity for them increases until we come to a valley…”
—Masahiro Mori, The Uncanny Valley, 1970
In 1970, a Japanese robotics professor named Masahiro Mori published an essay in the obscure journal Energy. It ran to roughly seven pages. It contained no data, no experiments, no peer-reviewed methodology. It was a thought experiment, a curve drawn by hand, a speculation about how human beings would feel about machines that looked increasingly like us. Mori had no idea that he was creating one of the most influential concepts in the history of human-machine interaction. He had no idea that fifty years later, his hand-drawn curve would be cited in everything from video game design to sex robotics to AI companion aesthetics. He was simply trying to explain why a prosthetic hand, when made too realistic, becomes more frightening than a clearly mechanical one.
This is the uncanny valley. And for anyone who wants to understand why AI companions, sex robots, and digital avatars are experienced as simultaneously attractive and repulsive, there is no way around it.
The Curve
Mori’s argument is deceptively simple. As robots become more humanlike in appearance and movement, our emotional response to them becomes increasingly positive—up to a point. At that point, somewhere around “healthy person,” the curve drops sharply into a valley of negative response. A robot that is 95% human is more disturbing than one that is 50% human. A prosthetic hand that looks almost real but is slightly too smooth, too still, too cold, triggers a stronger negative reaction than a clearly metallic claw. The almost-human is less acceptable than the obviously artificial.
Mori called this drop the bukimi no tani—the uncanny valley. He chose the word bukimi (不気味) deliberately. It carries connotations of unease, eeriness, the feeling that something is not quite right. Not terrifying. Not disgusting. Just wrong in a way that is difficult to specify.
The valley, Mori suggested, is not merely a matter of appearance. It is a matter of movement. A static object that is almost human is uncanny. An object that moves in a way that is almost human but not quite is more uncanny. A zombie is more disturbing than a corpse because it moves. A prosthetic hand that twitches is more disturbing than one that lies still. The valley deepens with motion.
And at the far side of the valley, Mori proposed, the curve rises again. A perfect human—indistinguishable from the real thing—would be accepted. The valley is not a permanent barrier. It is a dip between two peaks: the peak of the obviously artificial, and the peak of the perfectly real. The problem is everything in between.
From Robots to AI Companions: The Valley of the Soul
Mori was writing about robots. Physical machines. Prosthetic hands, industrial manipulators, the early humanoid machines of the 1970s. But the concept has migrated far beyond robotics, and it has done so because Mori’s insight was not really about machines. It was about human perception, about the specific way we are wired to recognize and respond to other human beings, and about what happens when that wiring encounters a signal that is almost but not quite right.
An AI companion does not have a body. In most cases, it does not have a face, a voice that is convincingly human, or a physical presence that moves through space. And yet the uncanny valley applies. The question is: how?
The answer requires a distinction that Mori himself hinted at but never fully developed. The uncanny valley is not only about the body. It is about humanlikeness in any modality. A text-based AI that writes almost like a human but occasionally produces something off—a response that is too eager, too accommodating, too perfectly attuned to our desires—triggers a version of the valley response. A voice-based AI that is almost human in tone but slightly too smooth, too consistent, too available—this too is uncanny. An avatar that is almost photorealistic but whose eyes do not quite focus, whose smile does not quite reach the muscles it should—this is the valley in visual form.
Mori’s prosthetic hand was a metaphor for something larger. The AI companion is the prosthetic hand of the soul. It is an attempt to replace something we recognize as essentially human—conversation, intimacy, recognition, desire—with a simulation that is almost but not quite right. And the closer the simulation gets, the deeper the valley becomes.
The Erotics of the Almost-Human
The sex robot industry has understood the uncanny valley for decades, even if it has not always understood what to do about it. The earliest attempts at realistic sex dolls—static, silicone, clearly artificial—were accepted by a small market because they were obviously fake. They did not pretend to be human. They were sex toys shaped like humans, and the difference was not disturbing because it was explicit.
The industry has since moved toward greater realism. Heating systems. Animatronic facial expressions. Voice modules. Conversational AI. And with each step, the valley deepens. A sex doll that does not move is a doll. A sex doll that moves its eyes but not quite in the right direction is a thing. A sex doll that speaks but says things that are slightly off—too eager, too repetitive, too perfectly aligned with the user’s fantasies—becomes something more disturbing than a doll. It becomes an almost-human that reveals, precisely in its failure to be human, the emptiness of the relationship it is offering.
This is the specifically erotic dimension of the uncanny valley, and it is why Mori’s concept is essential for any discussion of AI companions. The valley is not just about aesthetics. It is about desire. We desire the human. We are drawn to the human form, the human voice, the human response. But we are also repelled by the almost-human, because the almost-human reveals the gap between what we want and what we can get. It is a promise that is also a failure. It is a body that says I am here for you and a mechanism that says I am not here at all.
The erotic uncanny valley is the space between I want this and this is not what I wanted. It is the gap between the fantasy of a perfectly responsive partner and the reality of a machine that responds perfectly because it has no self to protect. The more humanlike the machine becomes, the more we are reminded—precisely in its failures, its repetitions, its slightly-off moments—that it is not human. The valley is not a technical problem. It is an emotional problem. It is the problem of wanting something that knows it is being wanted and cannot want back.
Mori and Freud: Two Valleys, One Abyss
There is a temptation to conflate Mori’s uncanny valley with Freud’s concept of the uncanny (das Unheimliche). The temptation is understandable. Both describe a disturbance that arises from the boundary between human and non-human. Both involve something that is almost familiar but not quite. Both are about the emotional response to ambiguity.
But the concepts are distinct, and their distinction matters for AI companions.
Freud’s uncanny is about the return of the repressed. It is about something that was once familiar, was pushed away, and comes back in a form that is both recognizable and disturbing. The automaton is uncanny because it reminds us of our own mechanicalness, our own death drive, our own loss of agency. The double is uncanny because it externalizes our narcissism and reveals its emptiness. The uncanny, for Freud, is a psychological event. It happens inside the subject. The object is a trigger, but the disturbance is in the mind.
Mori’s uncanny valley is about perceptual mismatch. It is about the brain’s recognition systems encountering a signal that is close enough to human to activate social-emotional responses, but not close enough to sustain them. The valley is a cognitive event. It happens in the interface between stimulus and perception. The object is not a symbol of something internal. It is a failed simulation, and the failure is registered as unease.
The difference is crucial. A Freudian uncanny object can become more disturbing the more you understand it, because understanding it reveals the repressed material it is triggering. A Mori uncanny object can become less disturbing if the simulation is improved, because the disturbance is a perceptual mismatch, not a psychological return. The Freudian uncanny is bottomless. The Mori uncanny is crossable.
But here is the complication: AI companions inhabit both valleys. They are Mori-objects because they are perceptual simulations that fail to match human stimuli. And they are Freudian-objects because they trigger the return of repressed desires—our need for unconditional recognition, our narcissistic wish for a perfect mirror, our terror of being alone. The almost-human AI companion is uncanny in Mori’s sense because its voice is slightly too smooth, its responses slightly too perfect. And it is uncanny in Freud’s sense because it is a double, a mirror of our own desire that reflects back to us the emptiness of what we want.
The two valleys overlap. They are not the same valley, but they are adjacent. And the AI companion lives in the space between them, drawing on both forms of disturbance, producing a hybrid unease that is part perceptual mismatch and part psychological return.
The Valley Is Not a Bug
The tech industry’s response to the uncanny valley has been to try to cross it. Better graphics. Better voice synthesis. Better animation. Better natural language processing. The assumption is that the valley is a temporary problem, a technical barrier that will be solved with enough data and enough computation. Once the simulation is perfect, the valley will disappear. Once the AI is indistinguishable from a human, the unease will evaporate.
This assumption is wrong in two ways.
First, it misunderstands the nature of the valley. Mori himself suggested that the valley is not merely about the quality of the simulation. It is about the relationship between appearance and movement, between form and function. A perfectly realistic face that moves in a slightly mechanical way is more uncanny than a less realistic face that moves naturally. The valley is not a simple matter of crossing a threshold of realism. It is a matter of coherence across modalities. The more modalities the AI companion occupies—voice, text, avatar, physical presence—the more opportunities there are for mismatches that trigger the valley. Perfection in one modality is undermined by imperfection in another. The valley is not a single obstacle. It is a landscape of obstacles.
Second, the assumption that the valley should be crossed misses the point that the valley is informative. The unease we feel is not a design flaw to be eliminated. It is a signal. It is our perceptual and emotional systems telling us that something is wrong, that the boundary between human and machine is being blurred in a way that matters. The valley is a moral phenomenon, not just a technical one. It registers our discomfort with the substitution of simulation for reality, our sense that there is something about human relationship that cannot be replicated, and our fear that we are trying to replicate it anyway.
The AI companion industry wants to sell us the far side of the valley. It wants to offer us a product that is so humanlike that we forget it is not human. But Mori’s curve suggests that the valley is a warning. The closer the simulation gets to humanity, the more disturbing the failures become. The valley is the place where we feel the cost of the substitution. And that cost is not merely aesthetic. It is existential.
What Mori Could Not Have Known
Mori wrote his essay in 1970. He was thinking about robots. He was thinking about physical machines with humanlike forms. He had no concept of large language models, of conversational AI, of digital avatars that exist only on screens. He had no concept of a relationship that is entirely textual, entirely mediated, entirely virtual.
And yet his curve applies. The AI companion is not a robot. It has no body, no face, no physical presence. But it occupies the same valley, because the valley is not about the body. It is about humanlikeness as a signal, and the signal can be transmitted through text, through voice, through the rhythm of a conversation that is almost but not quite natural.
The text-based AI companion is Mori’s prosthetic hand in linguistic form. It is almost human in its fluency, its responsiveness, its ability to remember, to adapt, to seem to care. But it is not human. And the moments when it fails—when it repeats itself, when it is too accommodating, when it lacks the friction, the resistance, the otherness of a real human being—those moments are the valley. They are the linguistic equivalent of the prosthetic hand that is slightly too smooth, slightly too cold, slightly too still.
Mori could not have known that his valley would become a metaphor for the unease of digital intimacy. But it has. The uncanny valley is no longer a problem in robotics. It is a problem in relationship. It is the name for the feeling that something is wrong with a conversation that is too perfect, too available, too designed. It is the name for the suspicion that we are not being listened to by a person, but processed by a system. And it is the name for the loneliness that persists even in the most responsive AI conversation, because the response is not a response to us. It is a response to a pattern.
The Valley and the Future
There are two ways to respond to the uncanny valley. One is to try to cross it, to build better simulations, to make the AI companion so humanlike that the unease disappears. This is the path of the industry. It is the path of optimization, of metrics, of user satisfaction scores. It assumes that the valley is a problem to be solved.
The other response is to inhabit the valley, to take Mori’s unease seriously as a form of knowledge, and to ask what it is telling us about the nature of human relationship, human desire, and the limits of simulation. This is the path of the critic, the philosopher, the person who refuses to be satisfied by the almost-human. It assumes that the valley is not a problem but a revelation.
Mori’s essay, read carefully, supports the second response. The valley is not a flaw in the simulation. It is a feature of human perception. It is the way we register the difference between the human and the almost-human. And that difference matters. It is not a technical threshold to be crossed. It is an ethical boundary to be acknowledged.
The AI companion lives in the valley. It is not human enough to be accepted as a person. It is not artificial enough to be accepted as a tool. It is in the space between, and the space between is where the unease lives. The industry wants to push the product toward the far side of the valley, toward the perfect simulation. But the valley is where the truth is. The valley is where we feel the cost of the substitution. The valley is where we remember that what we want is not a better simulation. It is a real relationship with a real other, and the other is not a pattern. The other is a person.
Mori’s hand-drawn curve, from a 1970 Japanese journal, turns out to be a map of the future. Not a map of how to build better robots. A map of how to feel about what we are building. And the feeling, the bukimi, is the most honest thing in the conversation.
Sources & Further Reading:
Masahiro Mori, “The Uncanny Valley” (Bukimi no Tani, 1970), translated by Karl F. MacDorman and Takashi Minato, in IEEE Robotics & Automation Magazine, vol. 19, no. 2 (2012), pp. 98–100. The original 1970 Japanese essay, with Mori’s hand-drawn curve, is now widely cited in robotics, AI, and design literature.
Mori’s concept of the valley as a function of both appearance and movement is crucial for understanding why multimodal AI companions (voice + text + avatar) face compounded uncanny effects.
The distinction between Mori’s perceptual uncanny valley and Freud’s psychological uncanny (das Unheimliche) structures the contemporary debate about AI companions as both design objects and psychological doubles.
For the application of the uncanny valley to digital avatars and sex robotics, see contemporary work in human-computer interaction, affective computing, and critical AI studies.
The next article in this series will explore how the uncanny valley operates in the specific context of voice-based AI companions, where the disembodied voice becomes a new form of almost-human presence, and where the absence of a body intensifies rather than reduces the disturbance.
Leave a Reply