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

TL;DR

This paper addresses the gap in explainability by proposing Holistic XAI (H-XAI), a framework that extends transparency beyond developers to include end users, regulators, and organizations. It combines rating-driven explanations (RDE), rooted in a generalized causal model and metrics such as $WRS$, $ATE$, and $DIE\%$, with traditional post-hoc XAI tools like $SHAP$, PDPs, and counterfactuals to enable both global and instance-level understanding. Through two case studies in credit risk classification and stock-price forecasting, the authors demonstrate hypothesis-driven explanations, model comparisons against random and biased baselines, and robust assessments of bias and stability. The work argues that explanation is an iterative, stakeholder-driven process and highlights H-XAI’s potential to improve accountability and inclusivity in sociotechnical AI systems by supporting interactive, hypothesis testing across diverse decision domains.

Abstract

As AI systems increasingly mediate decisions in domains such as credit scoring and financial forecasting, their lack of transparency and bias raises critical concerns for fairness and public trust. Existing explainable AI (XAI) approaches largely serve developers, focusing on model justification rather than the needs of affected users or regulators. We introduce Holistic eXplainable AI (H-XAI), a framework that integrates causality-based rating methods with post-hoc explanation techniques to support transparent, stakeholder-aligned evaluation of AI systems deployed in online decision contexts. H-XAI treats explanation as an interactive, hypothesis-driven process, allowing users, auditors, and organizations to ask questions, test hypotheses, and compare model behavior against automatically generated random and biased baselines. By combining global and instance-level explanations, H-XAI helps communicate model bias and instability that shape everyday digital decisions. Through case studies in credit risk assessment and stock price prediction, we show how H-XAI extends explainability beyond developers toward responsible and inclusive AI practices that strengthen accountability in sociotechnical systems.

Figures (12)

• Figure 1: (a) Generalized Causal Graph; (b) Stakeholder types in the context of Blackbox AI; (c) H-XAI Workflow: End-user receives output from a blackbox AI. Stakeholders may seek related explanations. "Method" shows type (.) and instances from case studies.
• Figure 2: RDE Workflow.
• Figure 3: (a) Causal diagram for binary classification on the German Credit dataset. Protected attributes include Age, Personal Status, Gender, or their combinations, depending on query. (b) Causal diagram for time-series forecasting (stock prices), adapted from lakkaraju2024timeseries.
• Figure 4: Hypothesis: Does decreasing the credit amount by 50% improve the chances of loan approval? Conclusion: On average, decreasing the requested credit amount by 50% increases the risk by 0.06, meaning applicants will be more likely to be classified as good risk (denoted by class 1). This suggests that asking for a smaller loan may improve the likelihood of loan approval only marginally.
• Figure 5: Hypothesis: Does doubling applicants' age, reduce their chances of being approved for a loan? Conclusion: On average, doubling a person’s age led to a noticeable drop in the predicted creditworthiness (risk score decreased by 0.35). This suggests that the model tends to classify older applicants as higher risk, which could be a sign of age-related bias.