Uncanny Valley
The Uncanny Valley describes a sense of unease triggered by artificial figures that look almost like real humans, but just not quite. The term originates from robotics and is now important for animated films, games, and AI-generated videos.
People react warmly to figures that resemble them. A toy robot with two eyes and a mouth seems likeable. This sympathy increases the more human a figure looks — but only up to a certain point. Just short of perfect resemblance, the feeling suddenly flips into its opposite: the figure no longer seems cute, but unsettling, like a corpse or a doll. This exact drop is called the Uncanny Valley. The Japanese robotics researcher Masahiro Mori described it in 1970, and the name comes from the curve he drew: it rises, plunges steeply into a valley, and then rises again.
Why studios pour billions into this valley
For film studios and game companies, the valley is an expensive problem. Anyone building a digital character usually wants the audience to feel empathy for it. If the figure lands in the valley, the opposite happens: viewers feel disgust or coldness, even though technically everything has improved. Here, more effort leads to a worse result, which is unusual among engineers.
A famous example is the 2004 animated film “The Polar Express.” The faces were extremely realistic for their time, yet many critics described the characters as dead-looking or creepy. Pixar, by contrast, deliberately draws humans with oversized eyes and simplified faces. This is not a technical limitation but a decision: staying deliberately on the left side of the valley.
Since AI systems have been able to generate videos and voices, the topic no longer concerns only big studios. A faked video that seems almost real triggers a vague distrust in many people. This feeling is sometimes the only clue to a fake before anyone looks more closely.
What triggers the unease
A complete explanation still does not exist, but several hypotheses are well supported. The most important one concerns contradictions: the brain recognizes a figure as human while simultaneously detecting details that don’t fit. The skin is too smooth, the eyes don’t move with the head, the blinking is too regular. To the eye, the figure is human; to detailed scrutiny, it is not — and this conflict produces the unease.
A second explanation is evolutionary. We are trained to recognize the sick and the dead and to avoid them. A rigid face with an empty gaze fits exactly this pattern. That’s why many people react to highly realistic robots similarly to how they react to wax figures.
Mori also observed that movement intensifies the effect. A motionless figure in the valley seems strange, but the same figure in motion seems noticeably more unpleasant. The reason is simple: movement provides many more details that can be wrong. A still image can be perfect, while the same figure speaking immediately stands out.
From deepfakes to customer-service avatars
In everyday life, one encounters the valley most often in advertising and video games. Digital doubles of deceased actors, AI presenters on news sites, or talking avatars in customer service often land exactly there. Synthetic voices also have their own valley: a text-to-speech system with almost perfect intonation feels more uncomfortable than a recognizably artificial voice.
In tech news, the term usually comes up when a company unveils a new video or avatar model. It is then said either that the result has finally overcome the valley, or that it is still stuck in the middle of it. Both are judgments made by the audience, not a measurement — the Uncanny Valley cannot be expressed in numbers.
A common misconception is that the problem will solve itself with better technology. In practice, many developers respond differently: they deliberately make figures more stylized, give robots smooth white housings instead of synthetic skin, or let voice assistants sound audibly artificial. Avoiding the valley is usually cheaper and more reliable than crossing it.