Simulating High-Stake Decisions

摘要

Military simulation and decision-support systems differ from other simulation and decision-support systems since they must be able to support high-stake single-instance decision making, such as combat planning and other planning tasks where expensive equipment or human life are at stake. Utility theory has studied preference models for planning in high-stake single-instance domains but does not provide methods that determine such plans efficiently. Artificial intelligence has developed planning methods that are efficient but not adequate for high-stake single-instance decision making. We discuss novel decision making methods that combine constructive planning approaches from artificial intelligence with the more descriptive approaches from utility theory to efficiently determine plans that approximate the risk-attitudes of human decision makers. This allows one to simulate decisions in military planning domains more accurately.