“Achieving AI that goes beyond human capabilities will require reinforcement learning and other techniques, not just imitation learning.”