The “uncanny valley”

There is a feeling we humans get when confronted with human-like artifacts. The more human-like the artifact the more we tend to like it. Think dolls or mechanical robots. But when these artifacts start to look much more like human our liking wanes and we start to get a creepy feeling.

This phenomenon is called the “uncanny valley” and has befuddled researchers in affective computing and robotic design. Making robots look human is something of a dogma when we imagine the robots of the future. Given the literature that precedes and inspires current research, we expect robots to evolve into androids, the Marias of Metropolis or the Rachaels of Blade Runner. But will they ever? And if they do, are we going to accept them?

Would you date her?

The “uncanny valley” phenomenon could potentially spell the end of android evolution. To look into the causes of this, an international team led by Ayse Pinar Saygin of the University of California, San Diego (see journal citation below) made an experiment scanning the brains of 20 subjects aged 20 to 36 while they were looking three different scenarios: (a) a human, (b) a mechanical-looking robot, (c) a human-like robot.

Interpreting the results from the fMRI scans the researchers suggested that the cause for the “uncanny valley” is a mismatch between at least two neural pathways, that of recognizing a human-like face and that of recognizing the robotic movement. Humans do not move like robots. So when someone looking like human moves like a robot it makes us feel that “something is wrong”.

The results of the experiment reminded me of Capgrass Syndrome, the personality disorder syndrome where patients feel that their world is populated by mechanical impostors of their family and friends. Interestingly, research by Vilayanur S. Ramachandran in 1997 hypothesized that Capgrass Syndrome is caused by a mismatch too. This time it was between the pathways that make the patient recognize the face of a loved one and those that evoke an emotion about that person. Failing to feel anything about the one you see creates the creepy feeling that there is “something wrong” about that person; that he is not “real”.

The connection between Capgrass Syndrome and the UncannyValley phenomenon runs deep into the culture of AI. The paranoid feeling of doubles is a common theme in Philip Dick’s work, which informs our contemporary techno-cultural milieu. Indeed Rick Dechard’s (played by Harrison Ford) dilemma in Blade Runner is to decide if Rachel is “real” or “artificial”. Can he really “love” Rachel? The Turing Test could be seen also as the statement par excellence that blurs the borders between “real” and “artificial” on the basis of emotional perception from the human observer. If Rachel speaks and moves like a human, then she is. Or isn’t she?

The only “cure” for the uncanny valley phenomenon would be to retune our perceptual systems into accepting mechanical-like motion from human-like artifacts. But this may not be so straightforward. Human brains are haphazardly evolved objects that guide our actions, our imaginations, our horrors and our triumphs, conditioned by the most powerful emotion that nature has ever invented: fear.

Journal Reference: A. P. Saygin, T. Chaminade, H. Ishiguro, J. Driver, C. Frith. The thing that should not be: predictive coding and the uncanny valley in perceiving human and humanoid robot actions. Social Cognitive and Affective Neuroscience, 2011; DOI: 10.1093/scan/nsr025