Dynamic decision making based on partial probability information

Abstract

In this paper we introduce Lazy Decision Making which is a framework for dynamic decision making under uncertainty. The key idea is to start with an imprecise description of the agent's knowledge concerning the state of nature (by means of a set of probability distributions) and to successively refine this description (i.e., the corresponding set) until a "good" decision can be made. The crucial point about this scheme is that the refinement has not to be done in an arbitrary way (which might be extremely inefficient). The algorithm assists the agent by providing him hints on the effect of a refinement of the constraints in question hence guiding him in making the relevant things more precise. © Springer-Verlag 2001.