Reinforcement Learning: a Subtle Introduction