The Robot Mirror Problem
I just watched Honda’s “Living with Robots” clip and realized that there might be a quite peculiar challenge unique to robots: the mirror problem.
The mirror test is used as a method of testing self-awareness in animals. Some higher animals, including us humans, are quite good at recognizing their mirror image. And so will autonomous machines have to be if they want to successfully navigate the real world. In particular if they are household aids where mirrors are a quite common sight.
However, unlike for most humans deciding whether they look in a mirror or not might not be as easy for robots. Humans, unless they are identical twins, almost never encounter another person that looks exactly like them. For robots encountering “mirror” copies of themselves is rather likely. In that case they need to decide whether they look at another robot or at a mirror.
For humans this is simple: we just wink. If we aren’t looking in a mirror the person opposite to us will have problems to wink exactly at the same time. Not so for a robot. If its twin is fast enough it might succeed in tricking the robot into thinking it deals with a mirror.
This scenario could be thought of being rather unlikely. And it certainly is for household robots. However, if battlefield robots are starting to encounter each other more often a behavior like this will emerge rather sooner than later. And this friend-foe detection will not be a trivial problem to solve. Instead of camouflage print we might see intricately patterned raggery on robots…
