What's in it for AI? — LessWrong