The "overlearning the game" problem