Are Better Machine Training Approaches Ahead?

Why unsupervised, reinforcement and Hebbian approaches are good for some things, but not others.