"If AI systems become appreciably better at driving cars than humans are, would it be unethical for people to drive on their own, since this increases the risk of an accident and could cost tens of thousands of lives a year? What if AI was only somewhat better, but in the aggregate would still save thousands of lives a year?"