Probabilistic methods and the interface with statistics

There has been a long debate on the appropriate means to incorporate uncertainty into expert systems but developments in methodology and computing tools now make it feasible to construct realistic applications that use rigorous probabilistic reasoning. Moreover, by exploiting links with Bayesian statistical reasoning, techniques exist for systematically updating and criticising both quantitative and qualitative aspects of the model. The author describes the current status of work being carried out jointly with University College London and the University of Aalborg, Denmark, emphasising developments since S.C. Lauritzen and D.J. Spiegelhalter (1988). An application in the diagnosis of congenital heart disease is used for illustration