«Boundaries», Part 3a: Defining boundaries as directed Markov blankets — LessWrong