The critically acclaimed television series 30 Rock has one episode in which variety show host Tracy Jordan plans to create a pornographic video game. Frank Rossitano, a writer in this fantasy of what happens behind the scenes at a Saturday Night Live–like comedy show, informs Jordan that the game would surely flop because of something called the uncanny valley. He even produces a graph to demonstrate why failure is all but inevitable.
The uncanny valley has kindled debate among roboticists for more than 35 years—and more recently computer graphics jocks have joined this ongoing discussion about whether their creations will end up scaring people. Envisaged in 1970 by Japanese roboticist Masahiro Mori, the concept implies that whereas cartoonish or other abstract human figures draw immediate empathy, robots or animations that appear similar to humans (but not identical) provoke a sense of unease.
In an otherwise human-looking robot or animation, a stilted arm or eye movement—or perhaps a halting gesture in landing a kiss in Tracy Jordan’s porn game—creates an eerie sensation. This feeling is represented by a sharp dip on a graph, an “uncanny valley” in which the observer’s comfort level with the artificial character diminishes drastically. In Mori’s view, a full ascent out of the valley occurs only when robots become indistinguishable from humans.
Mori recommended that robot designers should avoid queasy reactions altogether by building robots that are not human facsimiles, an idea that has sometimes been adopted as a de facto design principle among roboticists. Despite his warnings, designers have ignored his entreaty. They can now build robotic heads or bodies covered by skin realistic enough to at least briefly fool humans [see “Android Science,” by Tim Hornyak; Scientific American, May 2006]. The silicone layer that produces this convincing effect has, in fact, become the stuff of $6,500 sex dolls.
As robots have begun to catch up with their masters, researchers have started to ask whether the uncanny valley actually exists. Mori’s graph was not based on experimental data—and recent studies mapping out responses to humanlike robots have produced conflicting results. David Hanson of Hanson Robotics in Richardson, Tex., has found that people’s varying reactions to an anthropomorphic robot or animation does not depend on the level of realism; instead it hinges on whether a robot’s appearance has an inherently creepy aesthetic. Frankenstein’s monster elicited a repellent reaction not because of an overwhelming human likeness but simply because the monster was just plain ugly. Intentionally steering clear of realistic human forms, as Mori suggested, offers no protection. “A Disney villain or cartoon can be very abstract and still unsettling,” Hanson says.
The uncanny valley may not be an exact representation of people’s perceptions of the bizarre, but a few studies provide some basis for Mori’s intuitions. As researchers have conducted experiments in search of the valley, they have found that as a robot or animation grows more realistic, the latitude that designers have to change the size of, say, the eyes or the head decreases greatly. “As you come closer to building robots that look more human, there is a narrowing of the range of forms that would still be acceptable,” says Karl MacDorman, an Indiana University professor, who attributes these reactions to innate aversions to traits that might be linked with ill health or lack of fertility.
The science of aesthetics is about to grow more complex as humans choose to morph themselves. “How will we behave when people are not quite right, not because of behavioral or physical problems but because of behavioral or physical enhancements?” asks Jamais Cascio, a consultant to the Institute for the Future. Prosthetics and genetic engineering may affect appearance; even now the work of cosmetic surgeons can yield an unease reminiscent of the valley. One blogger placed Madonna—after requisite face-lifts, Botox injections and photo retouching—at the exact spot on Mori’s graph once occupied by the handicapped, which is near the bottom of the valley (but usually replaced today by a prosthetic hand for political correctness). Robots, humans, who knows, maybe even Mickey Mouse, all appear headed for Mori’s big pothole.
The Creepy Express
Hollywood is aware of the potential for the near human to alienate audiences. After working as a lead animator on The Polar Express, a movie criticized for its characters’ high creep quotient, Kenn McDonald and his colleagues at Sony Pictures Imageworks watched the film about five times and, in subsequent projects, decided to animate the rapid, small eye movements called saccades. “If you get the eyes right, everything else is icing on the cake,” McDonald says. Without invoking roboticist Masahiro Mori’s uncanny valley specifically when designing characters, Pixar Animation Studios does not stray far from Mori’s advice. “If your goal is to create realistic humans,” says Pixar production designer Ralph Eggleston, “you run the risk of the audience being distracted when things aren’t just right, instead of having people pay attention to the story.”
Note: This article was originally published with the title, "Into the Uncanny Valley".