Ethics in a Feedback Loop: A Parable — LessWrong