The pros and cons of reinforcement learning