Explaining and Refining Decision-Theoretic Choices

As the need to make complex choices among competing alternative actions is ubiquitous, the reasoning machinery of many intelligent systems will include an explicit model for making choices. Decision analysis is particularly useful for modelling such choices, and its potential use in intelligent systems motivates the construction of facilities for automatically explaining decision-theoretic choices and for helping users to incrementally refine the knowledge underlying them. The proposed thesis addresses the problem of providing such facilities. Specifically, we propose the construction of a domain-independent facility called UTIL, for explaining and refining a restricted but widely applicable decision-theoretic model called the additive multi-attribute value model. In this proposal we motivate the task, address the related issues, and present preliminary solutions in the context of examples from the domain of intelligent process control.