Nick Bostrom's 2006 book 'Superintelligence' introduced the 'paperclip problem' as an analogy for AI existential risk