Non-Obstruction: A Simple Concept Motivating Corrigibility — LessWrong