When Are Context-sensitive Beliefs Relevant?

I begin with the question of whether self-locating beliefs can be relevant to beliefs de dicto. Can it ever be rational for an agent to change her degree of belief in a belief de dicto as a result of gaining only self-locating information? To analyze this question, I develop a formal technique for modeling the ideally rational evolution of an agent’s degrees of belief as she learns new information over time. This modeling technique improves upon traditional conditionalization-based techniques by being general enough to correctly model stories involving context-sensitive beliefs. It also makes explicit the assumptions that go into a model and leaves no question what that model says about a given story. These improvements are achieved by devoting special attention to the language over which a model is defined; the central innovation of the technique is a principle for determining whether the verdicts of a model change when the modeling language expands. As I develop the modeling technique, I apply it to a number of stories — including Arntzenius’s “Shangri-La” story involving the threat of cognitive mishap — and show that it yields intuitive results in obvious cases. I then apply the model to the Sleeping Beauty Problem and show that David Lewis’s solution to the problem is incorrect. This result also demonstrates that self-locating beliefs can be relevant to beliefs de dicto. In a class of cases I describe, it is rational for an agent who learns only self-locating beliefs to adjust her relative degrees of belief in a set of beliefs de dicto all of which are logically compatible with what she has learned. Introduction. How much do I learn when I learn what time it is, or where I am, or who I am? Beliefs about one’s spatio-temporal location in the universe and beliefs about one’s identity are often called “self-locating beliefs,” and it is uncontroversial that a self-locating belief can be relevant to other self-locating beliefs. When I learn that today is Monday, it is rational to respond by adjusting my degree of belief that tomorrow is Tuesday. But is it ever rational to adjust one’s degree of belief in a belief de dicto in response to learning a self-locating belief? Here I am adopting David Lewis’s terminology in (Lewis 1979), while altering it at one point. Lewis takes all beliefs to be self-ascriptions of properties, and therefore describes all beliefs as “beliefs de se.” Among beliefs de se are beliefs de dicto, which limit the class of possible worlds to which their subject might belong. Lewis describes beliefs de dicto as locating their subject within “logical space.” On the other hand, there are “irreducibly de se beliefs,” which indicate the