How is reinforcement learning better?