Interview with David Leake

Very early, there was an insight that explanation was crucial for system acceptance. I remember that in Buchannan’s and Shortliffe’s early work on medical consultation systems, the ability to create explanations was a number one requirement. Initially, it seemed that simple explanation mechanisms might be sufficient, and tracing the rules used by an expert system is straightforward. Fairly soon it became clear that what makes a good explanation depends on much more than simply tracing the systems internal processing. There are all sorts of different types of explanations, which makes it necessary to develop some deep understanding of the nature of explanation first and then to develope AI methods to reflect all those different types.