Holistic Explainable AI (H-XAI): Extending Transparency Beyond Developers in AI-Driven Decision Making

Explainability currently serves the wrong audience: Most AI explanation tools are designed for developers rather than the people affected by AI decisions (like loan applicants) or regulators who oversee them. H-XAI addresses this gap by making explanations accessible to multiple stakeholders with different needs.

Interactive questioning reveals hidden biases: Rather than passive explanations, H-XAI enables users and auditors to actively test hypotheses about how AI systems work by comparing actual model decisions against automatically generated random and biased baselines—helping surface discrimination that might otherwise go undetected.

AI decisions in high-stakes domains demand transparency: The paper demonstrates H-XAI's application in credit scoring and financial forecasting, showing that when AI systems mediate critical decisions affecting people's financial lives, transparency and bias detection are essential for maintaining public trust and fairness.

Combining multiple explanation approaches strengthens accountability: By integrating causality-based methods with post-hoc explanations and providing both global (system-wide) and instance-level (individual decision) insights, H-XAI creates a more complete picture of how and why an AI system makes decisions, supporting better organizational oversight.

Explainability is a sociotechnical responsibility: The framework positions AI transparency not just as a technical problem but as part of broader responsible AI practices that require alignment between developers, affected users, auditors, and organizations to ensure AI systems serve society fairly.