x
Avoiding the Bog of Moral Hazard for AI — LessWrong