Great article in the April Wired magazine. They report on a study (full PDF available) by some Stanford researchers who found that people who are operating human-like robots attribute the behavior to the robot a lot more than people who operate non-human-like robots, who see the robot as extensions of themselves.
This may sound esoteric, but it has lots of important implications. People remotely operating military robots that are human-like might become overly violent because they can attribute the behavior to the robot (rather than themselves). On the other hand, it might be useful to use human-like robots for search and rescue to minimize the amount of traumatic stress felt by the operator (because they can dissociate themselves from the work). Our brains are often fooling us by anthropomorphizing objects. It would be nice if we can learn from it and take advantage when appropriate.