Answers are grounded in this paper and cite their sources.
Public-serving institutions (government agencies, courts, health systems, social-welfare bodies, and regulators) are adopting artificial intelligence (AI) to triage caseloads, allocate resources, detect fraud, and personalize services. Because these institutions exercise coercive and distributive authority over citizens who cannot easily exit the relationship, the stakes of getting AI wrong are qualitatively different from those in commercial settings. This paper surveys the advances, challenges, and future directions of responsible AI in the public sector, synthesizing scholarship on AI in government, algorithmic fairness, transparency, and AI ethics with practitioner-facing governance instruments. We first delimit what makes public-serving institutions distinctive: statutory mandates, due-process obligations, and diffuse accountability. We then map the state of the art across four dimensions (use cases, governance and policy instruments, fairness/transparency/accountability practices, and organizational capacity and procurement) and consolidate representative advances in a described table. We argue that the field has matured from principle proliferation toward operational risk management, exemplified by the NIST AI Risk Management Framework and the OECD AI Principles, yet a persistent gap remains between high-level commitments and verifiable practice on the ground. Core challenges include bias and equity harms, threats to legitimacy and due process, opacity and contestability, thin institutional capacity, and fragmented accountability. We conclude with a research and practice agenda emphasizing contestable-by-design systems, sociotechnical evaluation, procurement as governance, participatory oversight, and workforce capability. Responsible AI in public service is less a technical property of models than an institutional achievement sustained over time.
No takes yet. Share an insight, caveat, or question.
Unamma et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: