It's the reinforcement learning rather than supervised learning.