Abstract
If it is certain that performing an observation to determine whether P is true will in no way influence whether P is tme, then the proposition that the observation is performed ought to be probabilistically independent of P. Applying the notion of "observation" liberally, so that a wide variety of actions are treated as observations, this proposed new principle of belief revision yields the result that simple utihty maximization gives the correct solution to the Fisher smoking paradox and the two-box solution to Newcomb's paradox. Contrary intuitions are explained as arising from mistakenly treating subjective probability as a measure of the intensity of conscious assent, whereas it ought to be regarded as measuring dispositions to action.