Superintelligence: Paths, Dangers, Strategies

by Nick Bostrom

Consider, for example, the following “reasons-based” goal: Do whatever we would have had most reason to ask the AI to do.