Another advantage is that neuronal nets in most cases allow for fairly strainforward incorporation of additional information (side channels, lower layers). If this is combinaed with the algorithm learning is does allow fairly general integrative capabilities.