Because the human needs to tell the AI whether it’s the code or the comment that’s wrong.