Suppose you have a Newcomb predictor. More interesting than posing one-two box problems would be to predict the subject’s actions and do them before the subject becomes aware them. Eventually, their mind may adjust to the backwards causality arrow and we will learn many things