[necubi]

Games are "easy" for reinforcement learning because they have win conditions that can be automatically evaluated. Music isn't really like that.