Explanation in Bayesian belief networks
Henri J. Suermondt · 1992
Reasoning under uncertainty forms a central task for most medical decision-support systems. The paradigm of Bayesian belief networks allows us to reason under uncertainty using probability theory, without forcing us to make unwarranted independence assumptions. The belief-network representation has led to a recent resurgence in the use of probability theory in decision-support systems. Providing explanations of the conclusions of decision-support systems can be viewed as presenting inference results in a manner that enhances the user's insight into how these results were obtained. The ability to explain inferences has been demonstrated to be an important factor in making medical decision-support systems acceptable for clinical use. Although many researchers in artificial intelligence have explored the automatic generation of explanations for decision-support systems based on symbolic reasoning, research in automated explanation of probabilistic results has been limited. In this dissertation, I defend the thesis that we can explain belief-network inference results by determining the influences of findings and of network structure on those results; that the explanations thus derived improve users' insight into probabilistic inferences; and that such enhanced insight can lead to improved decision making by medical practitioners. This dissertation contributes a mathematical methodology that lets us determine the separate influences--on the belief-network inference result--of individual findings, sets of findings, belief-network arcs, and chains of reasoning. This methodology results in a set of functions that can be used to generate explanations. The methodology is general; it can be applied to any belief network. I call this explanation methodology--and its computer implementation--INSITE (Insight about Network Structure and Inference Through Explanation). In an evaluation study of INSITE in the domain of anesthesia, I compared subjects who had access to a belief network with explanations of the inference results, to control subjects who used the same belief network without explanations. I show that, compared to control subjects, the explanation subjects demonstrated greater diagnostic accuracy, were more confident about their conclusions, were more critical of the belief network, and found the presentation of the inference results more clear.