%0 Report %@ 531 %A Wesley, Leonard %T Reframing Evidential Problems %C 333 Ravenswood Ave., Menlo Park, CA 94025 %I AI Center, SRI International %D 1993 %K Evidential Problems %X Real world problems are rarely solved in one fell swoop. Rather, problem solving is an iterative process that involves acquiring and interpreting environmental information, then deciding what to do next. One alternative action might involve adjusting one's problem solving questions which a system must answer to find an acceptable solution. The desirability of pursuing such an alternative follows from the view that solving a problem often requires being able to identify and answer the ``right'' questions. In evidential reasoning (ER) systems, frames of discernments represent possible answers to the questions at hand. Asking a new question is equivalent to modifying an existing frame of discernment or constructing a new one, called ``reframing evidential problems'' here. In this paper, we describe a novel approach to deciding how to reframe problems based on minimizing an entropy characterization of the current situational knowledge. Currently, evidential knowledge-based systems (KDS) employ static frames; a user must modify frames if they are inadequate for the desired domain of application. Having some automated capability to make even simple adjustments to frames is expected to provide increased autonomy to KBS. An example of how the approach might be used in the domain of high-level computer vision is presented.