Explainable Interfaces for Rapid Gaze-Based Interactions in Mixed Reality

Explainability improves user performance: Users who had access to explanations of how the gaze-tracking AI system worked made significantly more accurate selections (10.8% improvement in F1 score) compared to those using the system without explanations, demonstrating that transparency directly enhances interaction quality.

Users learn and adapt with XAI: Participants using the explainable interface adapted their gaze behavior over time to work more effectively with the system, suggesting that understanding why the AI makes decisions helps users develop better strategies for interacting with it.

Black-box AI creates a usability problem: While machine learning models can achieve high accuracy in gaze tracking, their "black-box" nature prevents users from understanding how to effectively use them, highlighting a tension between raw accuracy and practical usability that explainability can help resolve.

XAI enables human-AI collaboration: The research suggests that explainable AI interfaces facilitate better collaboration between humans and AI systems in mixed reality environments, moving beyond one-directional AI performance toward interactive systems where users can understand and work with the technology.

Real-time explanations are feasible: The researchers successfully built a multi-level XAI interface that provided real-time explanations during live gaze-based interactions, demonstrating that explainability doesn't require sacrificing responsiveness in interactive systems.