XAIR: A Framework of Explainable AI in Augmented Reality

Explainability becomes critical as AR becomes mainstream: As augmented reality systems become integrated into daily life, users will increasingly interact with AI-driven services through AR interfaces. The paper emphasizes that explainable AI (XAI) is essential in these contexts because end-users need to understand why AI systems make particular decisions or recommendations.

A practical framework addresses three key design questions: XAIR provides designers with guidance on the fundamental questions of XAI in AR: "when" to show explanations (timing), "what" information to explain (content), and "how" to present explanations (format). This structured approach helps move beyond abstract principles to concrete design decisions.

User preferences should guide XAI design in AR: The framework was validated through a large-scale survey of 500+ end-users about their preferences for AR-based explanations, indicating that real user input—not just expert opinion—shaped the recommendations. This user-centered approach helps ensure explanations are actually useful rather than technically comprehensive but confusing.

Interdisciplinary expertise is necessary for effective XAI in AR: The research combined insights from explainable AI research, human-computer interaction (HCI), expert workshops with 12 specialists, and end-user studies. This multi-disciplinary approach recognizes that designing trustworthy AI experiences requires expertise beyond computer science alone.

The framework has practical applicability for designers: XAIR was tested with both designers (10 participants) and end-users (12 participants), demonstrating that it can actually help practitioners identify design opportunities and create more effective explanations in real AR applications.