Towards a scale-free theory of intelligent agency — LessWrong