Decision making under uncertainty

Although classical decision analysis provides a reasonable conceptual basis for rational choices (i.e., maximal expected utility), it fails to provide machine-friendly schemes for representing, updating and reasoning with uncertain knowledge. Rule-based systems, on the other hand, although computationally attractive, provide no basis for rationality or coherence. Attempts to bridge this gap have led to the development of graphical models known as Bayesian networks [Pearl 1988], which combine semantic coherence with computational manageability. This note surveys the development of Bayesian networks, summarizes their semantic basis and assesses their properties and applications.