How do you know those people aren't still "lying to themselves"? Humans are not known for being perfect, bias-free reasoners.

Maybe we can only really calculate utility after the fact by looking at what perfect Bayesian agents do rather than mere mortals.