Dialogue Explanations for Rule-Based AI Systems

Abstract

The need for AI systems to explain themselves is increasingly recognised as a priority, particularly in domains where incorrect decisions can result in harm and, in the worst cases, death. Explainable Artificial Intelligence (XAI) tries to produce human-understandable explanations for AI decisions. However, most XAI systems prioritize factors such as technical complexities and research-oriented goals over end-user needs, risking information overload. This research attempts to bridge a gap in current understanding and provide insights for assisting users in comprehending the rule-based system's reasoning through dialogue. The hypothesis is that employing dialogue as a mechanism can be effective in constructing explanations. A dialogue framework for rule-based AI systems is presented, allowing the system to explain its decisions by engaging in "Why?" and "Why not?" questions and answers. We establish formal properties of this framework and present a small user study with encouraging results that compares dialogue-based explanations with proof trees produced by the AI System.