Reinforcement learning is hard to apply to real-world problems, but one cannot deny the success that a company such as OpenAI has.