THE PROCESS OF DECISION-MAKING UNDER UNCERTAINTY WITH AN APPLICATION TO THE THEORY OF EXPERT SYSTEMS

HOMER KEITH HALL, Purdue University

Abstract

The broad objective of this dissertation is to examine the process of decision making under uncertainty. First, within the framework of the economics of information, a theoretical model is developed that describes rational choice as a process of gathering information in a sequence of r > 1 actions, determining an appropriate stopping time for the information collection, and then making a final decision. The formulation of an a priori strategy is described where each information gathering action, as well as the final decision, is contingent upon information signals received from previous actions. As information is collected during the execution of such a strategy uncertainty is reduced. Therefore, at any step in the sequence of information gathering actions the remaining portion of the strategy, called a substrategy, is seen to solve a subproblem of the original problem. It is shown that if a strategy is optimal for a decision problem (maximizes the expected net payoff), then any of its substrategies will be optimal for its associated subproblem. Second, the possible use of this, or some other decision-theoretical model, as a framework for computer based decision making systems is discussed. Since very little work has been done on the decision-theoretic basis of expert systems as a subject of artificial intelligence, this approach is first compared to the approach taken in most current expert systems. The major advantage is seen to be the focus on the efficiency of the decision process and therefore necessarily on the trade-off between the "correctness" and the cost of making a decision. Then, some of the computational aspects of using our theoretical model as a basis for an expert system are discussed. A method for acquiring sufficient information from a decision maker to calculate a gross optimal joint information and decision strategy (maximizing the expected gross payoff of the decision process) is discussed that does not depend upon fully specifying an outcome function.

Degree

Ph.D.

Subject Area

Economic theory

COinS