Adaptive Computer Aiding in Dynamic Decision Processes

Abstract

This report describes in brief a research program directed toward the application of adaptive computer techniques for aiding the human decision maker in dynamic decision processes. Aiding information of several types comes from the on-line acquisition of the decision maker’s value structure by a trainable computer system. A maximum-likelihood model of real-world behavior is used to predict environment-state transitions, and an expected utility model of decision-maker behavior is used to predict, evaluate, and modify or automate operator decisions. The overall system models information-acquisition strategy, as well as action choices. It is presently being implemented on an interactive minicomputer, and applied to a simulated intelligence operation involving surveillance of a mobile fishing fleet using sensors of varying costs and reliabilities. Research goals include experimental investigation of the factors which influence optimal decision aiding in complex, realistic and open intelligence-gathering and decision-making tasks. A major concern is to identify aiding techniques which best match the judgmental abilities of man with the discriminative capacity of an adaptive machine.