Dealing With the Expert Inconsistencies: The Sooner the Better.

In Carenini et al., 1995] we present an information based Bayesian strategy for history-taking, aimed at optimizing the evidence-gathering process. We have applied this strategy to a history-taking module developed as part of a system for patient education in the clinical domain of chronic non-organic headaches. The weakest phase in the development of our prototype application was the knowledge acquisition for the construction of the belief network that we used as our knowledge base. In this paper, we describe our additional elicitation effort , aimed at reening the numbers used in the belief network. We apply known probability elicitation techniques, based on the use of bets and lotteries. We also adapt a technique, not speciically devised for the elicitation of probabilities , to compute an analytical measure of the consistency of the assessments. The technique is in our opinion very promising, and lends itself to be applied more extensively to probability elicitation.