I’m not sure if it’s better or worse to think that the AI could genuinely feel sorry for making a mistake.