Coherence arguments imply a force for goal-directed behavior — LessWrong