The AI Race Has a Fatal Endgame
If superhuman artificial intelligence cannot be reliably controlled, the danger is not merely that someone might misuse it. The deeper problem is that competitive pressure virtually guarantees that someone will eventually build it anyway.