Corrigibility thoughts II: the robot operator — AI Alignment Forum