DARPA Perspective on AI — LessWrong