If agents are trained to disregard happiness, they lose the epistemic access to consequences needed ...