Aligning Superhuman AI with Human Behavior: Chess as a Model System