In June 2006 at the ATR Intelligent Robotics and Communication Laboratories in Keihanna, Japan, reporters and scientists gathered for the unveiling of a major new project by Dr. Hiroshi Ishiguro. Once everyone had arrived, an assistant pulled back a curtain to reveal…another Dr. Ishiguro? Certainly the second figure had a very strong resemblance to Dr. Ishiguro, wearing the same glasses and dressed in the same clothing. Seated in a chair, the duplicate was rocking one foot back and forth, blinking and adjusting itself. It looked around and then, in ordinary Japanese, introduced itself; it was named Geminoid HI-1.
For the reporters, up to that point virtually the only clue that Geminoid was an android had come from knowing that Ishiguro is a prominent roboticist. Ishiguro’s creation is more a puppet than an android, strictly speaking; Ishiguro speaks and acts through it via the Internet. As well as transmitting his voice, a motion-capture system allows Ishiguro to project the movements of his mouth and upper body onto Geminoid. The android itself is built of silicone and steel, and based on casts taken from Ishiguro’s body. Regular, small actions such as blinking are controlled by autonomous programs.
The strikingly realistic robot has since been met largely with wonder and admiration, which could mark success for Ishiguro in more ways than the obvious. Although Ishiguro’s earlier android projects were only a little less realistic, they tended to disturb viewers. This is consistent with a 1970 hypothesis by Dr. Masahiro Mori, another Japanese roboticist. Although not yet well-investigated by science, Mori’s “Uncanny Valley” theory holds that as a simulation of a human being’s appearance and/or motion becomes increasingly accurate, there is very suddenly a point at which humans’ interest in the creation turns into utter repulsion.
Ishiguro was inspired to develop a mechanical double after becoming tired of his long commute from the little town of Keihanna to a teaching position at Osaka University. He sees the android double as an improvement on videoconferencing, allowing not just the speaker’s image and voice to be transmitted but also his or her presence. In stark contrast with the Western fear that androids could become strong enough to overpower human beings, the Japanese forsee a future in which humans and androids work together amicably and productively.
However, the Uncanny Valley effect may prove to be an impediment to human-android interactions as androids come to resemble humans more and more closely. It’s an issue that Ishiguro wants to help resolve. One of his early robots was based on casts of his four-year-old daughter. It was capable of only basic movements, and thus was not quite lifelike. Ishiguro’s daughter was so terrified by it that she refused to set foot in Ishiguro’s lab after seeing it. Later on, Ishiguro made a robot copy of newscaster Ayako Fujii; despite being equipped with a much more intricate system of motion, it was still described as “creepy”. Ishiguro’s double is even more of an improvement, and most observers have been amazed and intrigued rather than unnerved. This may indicate that he has found the level of detail necessary to cross the Valley.
So why might there be an Uncanny Valley? There are a number of theories regarding its cause, all of them tentative since the existence of the Valley itself is not yet verified. One idea is that empathy for clearly nonhuman entities is based upon the recognition of human characteristics in an irrefutably different context. The human mind recognizes the subject as an obvious nonhuman, and then is attracted to it by the presence of human qualities.
The popularity of anthropomorphism is a testament to the validity of this part of this theory. Cartoon people and animals are a prominent example; the mind instinctively labels them nonhumans, but then finds reason to identify with their portrayal as creatures who think and feel in the same way that we do. Conversely, the theory holds that a response to a nearly-human-looking entity is exactly the reverse. The human mind’s first instinct is to label it ‘humanlike’ and only then notice the nonhuman characteristics of it. This causes the feeling of disgust and alienation. If the mind sees something as a human being, we want it to both look and act just as a human being does. This may have ties to evolutionary psychology and the maintenance of the species’ gene pool.
Another possible explanation for the existence of an Uncanny Valley is that it is an extension of the natural human tendency to fear death. A 2005 study by Karl F. MacDorman explores this possibility, which ties into Mori’s earlier reference point of ‘corpse’ as the lowest point in the Valley. After viewing images of almost-human-looking beings, participants tended to feel disturbed and to want to cling to opinions that comforted them. This is just what social psychologists believe happens during ‘terror management’, or the process of dealing with the reality of dying and its implications.
The whole Uncanny Valley theory, however, is still controversial. Some reject it completely, arguing that humanlike robots were not realistic enough in the 1970s for the effect to have been measured. David Hanson, a noted American roboticist and sculptor, considers the entire theory “pseudoscientific” and believes that it is futile to reduce ‘realism’ to a single axis on a graph. Likewise, psychologist Sara Kiesler of Carnegie Mellon University argues that there is evidence both to support the theory and to refute it.
A domain in which the concept of the Uncanny Valley – unproven or not – has become very important is CGI animation. Again, traditional two-dimensional cartoons are far enough removed from real-life human beings for the Valley not to be a problem; when animation aims at realism, though, is where it could be possible to fall into the Valley. Some believe that the box-office failure of 2001’s Final Fantasy: The Spirits Within can be partially attributed to the Uncanny Valley effect. The film required an enormous amount of work since its human characters were created from scratch. For instance, protagonist Dr. Aki Ross was animated in so much detail that it took an hour and a half to create each individual frame in which she appeared. After four years and hundreds of millions of dollars, Final Fantasy opened to mixed reviews. “At first it’s fun to watch the characters,” wrote Peter Travers in Rolling Stone, “[b]ut then you notice a coldness in the eyes, a mechanical quality in the movements.” Despite the immense amount of effort and money put into it, the film didn’t even make a tenth of its production cost in American and Canadian theaters combined. It was a sharp disappointment, and the film’s production company, Square Pictures, was put into bankruptcy.
There have been two strategies employed in CGI animation more recently to avoid going the way of Final Fantasy. One is to deliberately choose an exaggerated, cartoonish look for characters in order to avoid the potential pitfalls of the Uncanny Valley by staying well to the left of it on Mori’s graph. This is precisely what Pixar did for 2004’s The Incredibles, whose characters are recognizably human but not much more realistic than a two-dimensional cartoon. The other strategy is to do exactly what Hiroshi Ishiguro did: to “jump the Valley” by working backwards from real human beings. In the Lord of the Rings trilogy, for example, the character of Gollum was animated based on the captured recorded motions and voice of actor Andy Serkis. Moral of the story: it is very difficult to pull off creating a realistic animated human being from scratch – and perhaps even foolish to try.
Overall it seems clear that there is something causing the anxiety epitomized by Hiroshi Ishiguro’s daughter and the uneasiness experienced by many viewers of computer-animated human beings. Whether this is all based in an Uncanny Valley of aesthetics and movement or based in something else, it will likely have to be thoroughly explored and resolved if humanoid robots are ever required to become a part of human society. Masahiro Mori, although of course not at all skeptical of his own theory, agreed with the need for further study in order “to know what is human [and] to establish the design methodology for creating familiar devices through robotics research”.