Explainable Inference in the FRANK Query Answering System

The demand for insights into how artificial intelligent systems work is rapidly growing. This has arisen as AI systems are being integrated into almost every aspect of our lives from finance to health, security and our social lives. Current techniques for generating explanations focus on explaining opaque algorithms such as neural network models. However, considering the fact that these models do not work in isolation, but are combined, either manually or automatically, with other inference operations, local explanations of individual components are simply not enough to give the user adequate insights into how an intelligent system works. It is not unusual for a system made up of fairly intuitive components to become opaque when it is combined with others to build an intelligent agent. In this paper we argue that there is the need to combine diverse forms of reasoning in order to generate explanations that span the entire chain of reasoning: not just explanations for the, so called, black-box models. Our hypothesis is that: A hybrid approach using statistical and deductive reasoning makes possible a richer form of explanation not available to purely statistical ML approaches. We explore the concepts of 'local' and 'global' explanations and show how to give users a wide range of insights, using what we term an 'explanation blanket'. We tackle this challenge using the FRANK query answering system and show that its hybrid approach facilitates this kind of reasoning with explanations. It is important to note that the evaluation of user preferences for explanation is outside the scope of this work.