Transcending XAI Algorithm Boundaries through End-User-Inspired Design

XAI research has overlooked non-technical end users: The paper argues that existing explainable AI algorithms are designed primarily for technical users, while non-technical end users—who actually need explanations to make safer decisions in high-stakes domains like healthcare, criminal justice, and finance—have been largely ignored by the research community.

End users need explanations for practical decision-making, not just technical understanding: Non-technical users require AI explanations for specific functional goals such as "making safer and better decisions" and "improving users' predicted outcomes," which differ fundamentally from the technical transparency goals that currently drive XAI algorithm development.

Four new technical problems emerge from end-user needs: The research identifies novel XAI challenges including edge-case-based reasoning, customizable counterfactual explanations, collapsible decision trees, and a new "verifiability metric" to measure whether XAI actually helps users—suggesting that current evaluation methods may not capture what matters to non-technical users.

Current XAI limitations pose safety risks in critical domains: The lack of end-user-focused explainability support may compromise safe and accountable AI use in high-stakes applications, indicating that responsible AI deployment requires rethinking how explanations are designed and evaluated beyond technical metrics.

A paradigm shift is needed in XAI research: The paper calls for moving from a technically-driven research approach to a user-centered one, requiring collaboration between human-factors researchers and technical XAI developers to create explanations that actually serve end users' real-world needs.