Towards a Logical Account for Human-Aware Explanation Generation in Model Reconciliation Problems

A model reconciliation problem focuses on producing explanations for human users who have varying expectations of the AI agent. This research explores the development of a general framework for generating human-aware explanations in such problems. We face two primary challenges: creating an expressive and efficient framework that generates personalized and persuasive explanations for users, and interactively incorporating users' knowledge, beliefs, and preferences into the explanation process. We propose that a logic-based framework is well-suited for identifying, representing, and offering robust and tailored explanations to human users in model reconciliation scenarios.