Skip to Main Content
This report describes three experiments on Bayesian diagnostic systems. A system simulation facility provides the dynamics of a real-time environment in which the military activities of a fictitious adversary are portrayed. On the basis of intelligence data describing events in this hostile environment, a threat-evaluation team provides diagnoses regarding the threat posed by deployments of hostile military forces. The diagnoses are in the form of posterior probability estimates. The estimates made by men are compared with those calculated on the basis of a modification of Bayes' theorem. The inputs for these calculations are P(D|H) estimates produced by the same individuals who estimated the posterior probabilities. The results encourage further research on automated Bayesian hypothesis-selection procedures in threat-diagnosis systems.