In both cases, it would be better to train the machine learning model with a loss function that ignores the human’s objective and then adjust predictions ex post according to that objective. We ...