Why giving AI 'human ethics' is probably a terrible idea