From Go to StarCraft to Dota, artificial intelligence researchers are creating reinforcement learning systems that can defeat human experts at complicated games.