Should AI systems have to identify themselves? — LessWrong