Fair, Explainable, and Accountable AI: Human Oversight in Public-Sector Decision Systems

Abstract

Public-sector artificial intelligence can shape access to services, inspections, benefits, taxation, healthcare, mobility, and other consequential decisions. Its legitimacy therefore depends on more than predictive performance. Systems must avoid unjustified discrimination, provide explanations suitable for affected people and reviewers, assign responsibility, support challenge, and preserve meaningful human control. This review analyses fairness, explainability, accountability, and participation as interconnected dimensions of AI governance. It draws on the evidence base of a systematic synthesis that retained 95 high-quality studies and governance frameworks from 2020 to mid-2025. The source literature shows that ethics and transparency are frequently discussed, yet bias mitigation, accountability mechanisms, citizen participation, and implementation procedures remain uneven. The article examines how bias enters through problem definition, data, labels, modelling, thresholds, deployment, and institutional practice. It distinguishes transparency from useful explanation and describes accountability as a chain of ownership, documentation, review, appeal, and correction. It also considers the limitations of nominal human-in-the-loop arrangements and proposes conditions for effective oversight. A practical assurance model is presented that connects principles to controls, metrics, and evidence. The review concludes that fair and accountable AI requires institutional capacity, stakeholder participation, and continuous monitoring, not a one-time technical test or public statement.