Decision Models: a Theory of Volitional Explanation

This paper presents a theory of motivational analysis , the construction of volitional explanations to describe the planning behavior of agents. We discuss both the content of such explanations, as well as the process by which an understander builds the explanations. Explanations are constructed from decision models, which describe the planning process that an agent goes through when considering whether to perform an action. Decision models are represented as explanation patterns, which are standard patterns of causality based on previous experiences of the understander. We discuss the nature of explanation patterns, their use in representing decision models, and the process by which they are retrieved, used and evaluated. 1 Issues in explanation In order to learn from experience, a reasoner must be able to explain what it does not understand. When a novel or poorly understood situation is processed , it is interpreted in terms of knowledge structures already in memory. As long as these structures provide expectations that allow the reasoner to function eeectively in the new situation, there is no problem. However, if these expectations fail, the reasoner is faced with an anomaly. The world is different from its expectations. In order to learn from this experience, the reasoner needs to know why it made those predictions. It also needs to explain why the failure occurred, i.e., to identify the knowledge structures that gave rise to the faulty expectations, and to understand why its domain model was violated in this situation. Finally, it must store the new experience in memory for future use. Explanation is a central issue in this process of understanding and learning. The construction of explanations is also known as abduction, or inference to the best explanation. This process is usually viewed as the chaining together of causal inference rules in order to create a causal chain, in which a proposed set of premises is shown to be causally responsible for the event or fact being explained. However, there are two problems with this view. The rst problem is the familiar one of combi-natorial explosion of inferences. Most explanation programs create explanations by chaining together inference rules that describe the causality of the domain. For example, PAM Wilensky, 1978] used a set of planning rules connecting together typical goals and plans of people, and chained them together to form motivational explanations for actions observed in a story. However, this process is very ineecient in complex …