"The real risk with AI isn't malice, it's competence. A superintelligent AI will be extremely good at achieving its goals."