On Abstract Systems — LessWrong