Preferences, Utilities, and Quantitative Judgments

This paper is concerned with a major hiatus in prescriptive decision analysis, which of course includes portions of the AI preferences field when used in interactions with human beings. In such settings, all multi-attribute decision aids must elicit some sort of preference ordering, and this entails quantitative judgments. Therein lies the rub: since all quantitative judgments involve cognitive processes, they have certain properties that may or may not match the properties assumed by the corresponding model. In addition, such judgments are subject to biasing effects, such as range or centering effects, in varying degree depending on the nature of the judgment. Fortunately, there is a field of psychology, called psychophysics, that deals specifically with such judgments and their properties. This paper surveys the field and explains how it can be applied to assess the soundness of quantitative judgments made in the context of constructing preference orderings, and what general lessons might be learned. Note: this is distinct from issues of descriptive adequacy of the models, which may be entirely irrelevant in the prescriptive set-