DR-HAI: Argumentation-based Dialectical Reconciliation in Human-AI Interactions

Stylianos Loukas Vasileiou, Ashwin Kumar, William Yeoh

Keywords: Human-aware AI, Human-AI Interactions, Model Reconciliation, Dialogues, Argumentation

TL;DR: Explanations through logic-based argumentation

Abstract: In this paper, we introduce DR-HAI (Dialectical Reconciliation in Human-AI Interactions), a novel game-theoretic framework designed to extend model reconciliation approaches for enhanced human-AI interaction. By adopting a multi-shot reconciliation paradigm and not assuming a-priori knowledge of the human user's model, DR-HAI enables interactive dialogues to address knowledge discrepancies between explainee and explainer agents. We provide formal operational semantics for DR-HAI using logic-based argumentation and offer theoretical guarantees regarding the framework's termination and success. Furthermore, we conduct a human-user study that compares DR-HAI to single-shot reconciliation approaches, demonstrating the efficacy of our framework in improving users' understanding of AI decisions in tasks characterized by substantial knowledge asymmetry. Our findings suggest that DR-HAI offers a promising direction for fostering effective human-AI interactions.