Would be cool to try to train it using the technique from https://blog.openai.com/deep-reinforcement-learning-from-hum...