All of the major labs are experimenting with incorporating classic reinforcement learning into their architectures. If it works things will become very, interesting, very quickly.