Large Image Display: The Uncanny Valley

The image above, is a large scale version of the work of Karl MacDorman. It is reproduced here under the the terms of the GNU Free Documentation License. This is a direct translation from Japanese, of the original work by Japanese researcher Masahiro Mori in 1970 that led to the discovery of the uncanny valley, a problem that has plagued efforts to create artificial human or humanoid visages. The above graph is all the more remarkable in that it was originally based on pure hypothesis - not on any experimental data.

However, in the years since it was put forward, a number of researchers, the most notable of which is probably Karl himself. He is an associate professor in the human-computer interaction program at the School of Informatics at Indiana University and has been working with the uncanny valley hypothesis, using human participants, for the past four years. As such, he has been in a fairly unique position to fill in the experimental data behind the curve, and find out how much of it is genuine. The answer seems to be both 'quite a lot', and 'nothing'.

The existence of increasing attraction and affection for machines and animals exhibiting humanlike qualities as the graph predicts, is quite real, as is the sudden and steep drop-off, once you reach about 95% of the way to human-ness. It is as if some natural instinct is triggered, that warns us away from things that seem almost like a potential ally or mate, but aren't. Yet, at the same time there is no one single cause of this, and Karl has identified at least two dozen different triggers, each of which can be mapped on a graph of their own.

As the man himself put it back in 2006: "If you are interacting with an android and the timing of its speech and gestures is off, this will be uncanny for a different reason than if its eyes are too far apart. This in turn is uncanny for a different reason than if part of its body is open, exposing wires and motors. I have identified about 10 possible causes for the uncanny valley, and I am sure there are many more."

He was certainly right on the 'many more' aspect, and there is so much we need to get right to bridge the valley, and create avatars / androids that feel naturally human, that it seems an insurmountable task. Yet, the more we learn about the challenges presented, the closer we get to solving them. Ultimately, it is quite clear that we will conquer the valley, and that it is conquerable. It's just not going to be easy to do so.

Internal Links

Link: The Uncanny Valley

Link: Bypassing the Uncanny Valley with Hands

Link: Large Image Display: Animatrix: Final Flight of the Osiris: Its all in the Eyes

External Links

Link: The Uncanny Valley: Effect of Realism on the Impression of Artificial Human Faces

Link: Our uncanny ability to spot a fake