出處:https://technews.tw/2021/06/16/deepmind-says-reinforcement-learning-is-enough-to-reach-general-ai/