Responsible Artificial Intelligence in Tax Administration and Public Financial Management: Transparency, Procedural Fairness and Accountability
Abimbola Serifat Oreoluwa *
Department Cybersecurity, Nexford University, Washington, DC, USA.
Philip Williams Appiah-Agyei
Mississippi State University, Starkville, Mississippi, USA.
Grace Ikudehinbu
Southern Illinois University, Edwardsville, Illinois, USA.
Charles Zormelo
University of Texas at El Paso, Texas, USA.
*Author to whom correspondence should be addressed.
Abstract
Artificial intelligence is moving from peripheral experimentation to consequential use in tax administration and public financial management, where models can prioritise audits, detect anomalous transactions, screen procurement, support accounting and assurance, and structure interactions with taxpayers and suppliers. These applications may improve consistency, targeting and analytical capacity, but they also change how public power is exercised. This critical narrative review examines the conditions under which artificial intelligence can be used responsibly in revenue and public-finance functions, with particular attention to transparency, procedural fairness and accountability. The review integrates public-administration research, tax-compliance scholarship, public-sector accounting and auditing literature, machine-learning studies in tax and procurement, responsible-artificial-intelligence research, and authoritative governance instruments. The evidence is strongest for the predictive and screening value of machine learning in bounded tasks, including audit selection and procurement-risk detection. Evidence is substantially weaker for long-term organisational effects, distributional fairness, legitimacy, contestability and accountability outcomes. Three recurrent findings emerge. First, transparency must be differentiated into system, process and decision-level transparency; indiscriminate technical disclosure is neither necessary nor always compatible with enforcement integrity, but meaningful notice, documentation and reasons are essential. Second, procedural fairness cannot be reduced to model accuracy or the presence of a human reviewer. It depends on data provenance, label quality, consistency, individualised consideration, reason-giving, effective challenge and correction. Third, accountability requires an institutional chain of responsibility spanning authorisation, design, procurement, deployment, monitoring, appeal and independent assurance. Human oversight is useful only when reviewers have competence, time, authority and evidence that intervention improves decisions. The review proposes a risk-tiered Fiscal Artificial Intelligence Assurance Chain that links public-purpose justification to data and model assurance, operational controls, outward-facing explanations, redress and continuous audit. Responsible deployment therefore requires treating artificial intelligence not merely as a technical efficiency tool but as public administrative infrastructure whose legitimacy depends on demonstrable legality, fairness, traceability and correctability.
Keywords: Algorithmic accountability, automated decision-making, fiscal governance, machine learning, public-sector auditing, taxpayer rights, explainability, administrative justice