The Uncanny Valley

Last Tuesday, Jodi Forlizzi came from Carnegie Mellon to speak about design research. During the discussion, the term "uncanny valley" came up. This term refers to a phenomenon experienced, particularly in robotics, where as an artifact or other visual representation approaches near-human appearance viewers tend to experience strong feelings of strangeness and revulsion. Exactly why this happens is not entirely clear (and that it has a scientific basis at all is a matter of heavy debate). I have a couple of thoughts on possible causes, though.

Humans excel at judging others quickly based upon their physical appearance. Take our ability to independently detect red and green colors. This trait was evolved relatively recently and only a few animals (including 8% of human males) don't have it at all or experience it only in some limited capacity. The common explanation for the existance of red-green color differentiation is that it facilitates detection of edible food (think about how bananas change color as they ripen). An alternative theory, though, is that it improves our ability to analyze other people - specifically to decect subtle changes in skin tone, as when someone's face becomes flushed. This is just one mechanism that allows us to evaluate people at a glance, but countless others exist that help us to detect people with poor genes, mental disorders, or a sprained ankle. As robots grow more human-like in appearance, we begin to evaluate their visual traits as though they are actually human. Because we aren't good enough to replicate humans perfectly, the resulting robots fail various tests, signaling to us that they might have mental disorders or be dangerous in some other way. This signal manifests itself as a feeling of revulsion and helps us to avoid undesirable encounters (or potentially desirable encounters with helpful robots, but we don't get close enough to find out).

A second possibility draws on a concept presented in Scott McCloud's Understanding Comics - The Invisible Art. A single concept can be represented in different ways. A face, for example, can be represented by a detailed photographic likeness or by a variety of abstractions, all the way down to simply the word "face". McCloud contends that because we don't typically spend much time looking at our own faces, that our mental images of ourselves are fairly abstract; thus, we have an easier time relating to faces lacking detail, because they tend to resemble our mental self-image more closely than a detailed picture. Our mental representations of other people, on the other hand, tend to resemble their actual faces which we experience on a daily basis in all of their glorious detail. Consequentially, we tend to associate detailed faces with the general category of "other". The result is that, in comics, protagonists tend to have fairly plain faces to make it easier for readers to associate with them, while antagonists are often fairly detailed to enhance their other-ness. If this is the case, then our revulsion at human-like robots might be related to their detail-enhanced other-ness.

What does this mean for design? Within robotics, it clearly pays to be careful when making humanoid robots. On a more general note, I suppose it might be beneficial to err on the side of simplicity. In cases where details are not integral to the design, it may be best to leave them out to avoid giving people something to hate.