Given their advanced capabilities, future AI systems could pose significant risks to society. Some of this risk stems from humans using AI systems for bad ends (misuse), while some stems from the difficulty of controlling AI systems “even if we wanted to” (misalignment). We can analogize both of these with
