Superintelligence: Paths, Dangers, Strategies

by Nick Bostrom

Suppose that we stipulate that the AI should act (to do what it would be morally right for it to do) only if it was morally right for its creators to have built the AI in the first place; otherwise it should shut itself down.