In Search of The Uncanny Valley

We investigate one of the great challenges facing technological representations of the human form.

Feature by Michael Shea | 01 Mar 2016

Have you ever wondered why most computer-animated films, particularly the early ones, tend to feature human-like characters, but not humans? Toys, cars, bugs, ants, robots, monsters, rats and penguins. They're all nice enough, but why not use plain old humans instead? The answer is one of the greatest challenges facing all kinds of representations of the human form in technology. It's called The Uncanny Valley.

In 1970, Japanese roboticist Masahiro Mori wrote an article entitled The Uncanny Valley, in which he argued that as human likenesses appear more realistic there is an increasing level of comfort and familiarity up to a certain level. Yet at the point when a human facsimile is almost fully convincing it becomes repellent. This sudden dip in our emotional response is the uncanny valley.

This idea drew on Freud’s notion of the uncanny, applying it to robots and all manner of humanlike objects. Freud believed that the contrast between the expectation of something familiar and the realisation that the thing is strange causes an intense feeling of alienation or cognitive dissonance.

Mori used the example of prosthetic limbs to explain this idea: "Some prosthetic hands attempt to simulate veins, muscles, tendons, finger nails and finger prints, and their colour resembles human pigmentation... but this kind of prosthetic hand is too real and when we notice it is prosthetic, we have a sense of strangeness… In this case the appearance is quite humanlike, but the familiarity is negative. This is the uncanny valley."

How this affects animators

The reason why animators working with CGI are reluctant to use human characters is that they don't want to make their audience uncomfortable. While this might seem a little over-cautious, there is plenty of evidence of the pitfalls of ignoring the uncanny valley.

The 2004 animated film The Polar Express was widely criticised for the unsettling appearance of its animated characters, which had been created using 3D motion capture techniques before being digitally 'skinned'. The end product was described by critics as "lifeless", "disconcerting", even "nightmarish" and "horrifying". The problem was that the characters were quite realistic, yet not fully convincing, leading them to fall head first into the uncanny valley.

For the time being, the preferred solution is to use non-human or highly stylised human characters in computer-animated films. It remains to be seen whether future technology will allow animators to create characters so aesthetically convincing as to put the audience at ease and finally vault the uncanny valley. It is also quite possible that a computer-generated person that was indistinguishable from the real thing might be the most uncanny thing of all.

The field of humanoid robotics – more so than any other – is concerned with the problem of the uncanny valley and what the solution might be, if any exists. As in animation, one solution is to sidestep the problem altogether by creating non-human robots or highly stylised representations of humans (think ASIMO). Yet there are some who are determined to confront the uncanny valley or even to utilise it as part of their research.

A team of researchers at Indiana University recorded responses to images of people, androids and CGI characters with varying facial proportions. Their aim was to assess whether different kinds of humanlike aesthetics could produce the uncanny effect. They found that as ratings of humanlike and alive increased so did participants sensitivity to changes in the images. In other words it is more unsettling to view an image of a person with minor flaws than an image of a cartoon, robot, or animal, even though the latter is not at all realistic.

(Continues below)


The Skinny on robots:

Ana MAtronic Scissor Sister and robo-expert Ana Matronic on Robot Takeover

Pepper the robot Pepper: “The first robot that can read human emotions”


A further example of the uncanny valley being put to use is in the work of roboticist Hiroshi Ishiguro. He created an android based on his own appearance called Geminoid. The android's appearance, down to the smallest detail, is modelled on its creator, with silicone skin and skin textures painted manually to match MRI scans and photographs of Ishiguro. The finished product is profoundly creepy.

Once completed, Geminoid was used in experiments where it was used in conversations between adults and children, including between Ishiguro and his own daughter. The intention was to determine to what degree human co-presence could be achieved remotely via the android. Ishiguro’s daughter was unsurprisingly reluctant to interact with her father via his doppelganger. Geminoid’s functional limitations (being confined to its chair) and its movements and expressions (which never quite match those of Ishiguro) make it no match for the original in his daughter’s opinion.

Of course the potential for this kind of unsettled response is presumably why Ishiguro’s daughter was chosen for the experiment. The other child in the case study, who was very young, was quite confused by his experience. He sometimes said that the man (Ishiguro) was a robot, but later he thought it was a real person wearing a strange mask, saying, “He should take off his strange mask that he keeps wearing.” Masks, dolls and mannequins are all textbook examples of the uncanny.

Making robots realistic

The American roboticist and former Disney 'Imagineer' David Hanson used to build animatronic devices for use in theme parks. He gained worldwide attention in the mid-2000s when he began creating robotic human heads. Ignoring warnings of tumbling into the uncanny valley he endeavoured to create robotic heads that were as realistic as possible, using electroactive polymers, or 'frubber', to create convincing fake skin, along with around 30 servomotors to give a full range of expressions. 

Hanson defended the unsettling uncanny effect of the heads, arguing that avoiding the uncanny effect is a matter of aesthetics: "Extremely abstract robots can be uncanny if the aesthetic is off, as can cosmetically atypical humans... However, if the aesthetic is right, any level of realism or abstraction can be appealing."

As humanlike technology develops in sophistication the uncanny valley will continue to be a problem which provokes and inspires future generations of artists and scientists. The uncanny valley, with its influence ranging from the next animated blockbuster to prosthetics and human-robot interaction, will continue to unsettle us.