Using lying to detect human values — AI Alignment Forum