This is so key not that they learned it themselves, for some reason that doesn’t surprise, but how they learned it. Reinforcement learning seems to bring about ideal conditions for machine self-learning.