Explanation through Dialogue for Reasoning Systems

Explainability and transparency are becoming more critical in logical reasoning, such as in self-driving cars and medical care, where poor decisions can cause harm and, in the worst situations, death. To ensure such systems are morally sound, reliable, and secure, they must be capable of explaining their output or procedures in a human-understandable way. In this research, a dialogue explanation framework for Rule-based reasoning systems is presented to identify and explain discrepancies between the user and the system. It allows the system to explain itself by simply asking and answering "Why?" and "Why not?" questions. The formal properties of this framework and a small user evaluation that contrasts dialogue-based explanations with the proof trees generated by the reasoning system are described.