The ethical question of whether to turn an AI system on or off depends on whether it can suffer, which raises the question of what moral obligations we might have if we create something genuinely capable of suffering—we should perhaps hesitate before building such a system.