Expected utility of control vs expected utility of actions — LessWrong