What, can't we just reinforce-learn-bootstrap the sh*t out of this?