Reference: Lehmann, H. P. Book Review Uncertainty in Artificial Intelligence: A Review in the Context of Recent Workshops. 1990.
Abstract: In building any imagined autonomous computer agent, designers have to take account of the ambiguities and inaccuracies inherent in any agent's perception of the world, and of that agent's need to act despite those uncertainties. Regardless of one's interest in AI -- robot design, expert-system development, or psychological modeling, among others uncertainty management may figure prominently. This paper reviews the field of uncertainty in AI from the vantage point of the past four Workshops in Uncertainty in Artificial Intelligence. The four volumes on which we focus are not the entire literature on the subject, but they do contain information which is relatively inaccessible to the general readership and which represents much of the active research in the field. The areas on which we concentrate our effort are probability theory, non-monotonic logic, Dempster-Shafer theory, fuzzy-set theory, and pragmatic approaches. In each case, we examine the motivation and philosophical grounding for a given school. Second, we explore how the requisite qualities and quantities of a given approach can be assessed from people. Third, we describe the representational framework offered by each approach, and the types of uncertainty intended to be modeled. Fourth, we consider the calculus each approach has for updating the agentUs belief in states of the world, given new evidence. Fifth, we discuss the nature of decision-making in each approach. Finally, we present criticisms of each approach from participants in the workshops. This review gives the reader the vocabulary and knowledge to approach future research critically.