When will reinforcement learning learn an OS?

Everyone gets excited about reinforcement learning. Chess. Go. Driving. Etc.

How far in the future will reinforcement learning learn an operating system? Anyone have a pointer to current art trending that way?