Rationality and overriding human preferences: a combined model — AI Alignment Forum