This might be an extremely stupid question, but does it make sense to apply the concept of mirror neurons to ANNs?
See e.g. "Mirror neurons and imitation: a computationally guided review" at http://pacherie.free.fr/COURS/MSC/Oztop-Kawato-Arbib-2006.pd... for an overview.
The reason why you don't see so much of this type of literature is because the dominance of virtual AI: Google, Siri, etc. Hardware is expensive and learning with hardware is slow. However, the type of AI required is IMHO much more interesting. I think we really understand how our brains work if we have an AI competing at the Olympics.