65 User Studies on Human-Centered Explainable AI Reviewed
A systematic review of 65 user studies evaluating Explainable AI (XAI) systems across domains proposes a taxonomy for human-centered design. The paper, published on arXiv, critiques current evaluation processes as overly technical and insufficiently focused on human users. It provides a holistic overview of XAI system properties and evaluation metrics tailored to human needs, and outlines design goals adapted to user characteristics. The review serves as a guideline for XAI developers to conduct user studies that prioritize comprehensibility and user-centric evaluation.
Key facts
- Review covers 65 user studies evaluating XAI systems.
- Current XAI evaluation processes are too technical and not human-focused.
- The paper proposes human-centered design goals for XAI.
- Design goals are adapted to users' specific characteristics.
- The study provides a taxonomy of XAI system properties and evaluation metrics.
- Published on arXiv with ID 2510.12201.
- The review aims to guide XAI developers in conducting user studies.
- Focus is on making AI decisions and predictions comprehensible to humans.
Entities
Institutions
- arXiv