AI ADOPTION IN PUBLIC ADMINISTRATION: HOW GOVERNANCE CHOICES SHAPE SUCCESS AND FAILURE

Abstract

Governments everywhere are leaning on AI for big calls — who owes a welfare debt, how risky a defendant looks, how one government service talks to another. Similar technology, though, hasn't meant similar results. Some of these systems ended up ruled illegal in court. Others are still stuck in legal fights. A handful get held up abroad as models worth copying. Four real cases anchor this study: Australia's Robodebt welfare scheme, the Netherlands' SyRI fraud-detection tool, the COMPAS system that predicts reoffending in US courts, and Singapore's Smart Nation programme alongside its National AI Strategy. Together they cross three policy areas. Drawing on court rulings, an official inquiry report, investigative journalism, and government strategy papers, each case gets judged against six things — legal basis, transparency, human oversight, data-privacy protection, accountability and redress for citizens, and how things eventually turned out. A clear pattern runs through all four. The two worst failures kept citizens in the dark about how the system worked and left them to disprove it on their own. Singapore fared better largely because it had an ethics framework, a staged rollout, and review bodies in place before anything launched — not stitched together after a scandal. Whether some corners of government carry more risk than others is also worth asking. The paper closes with practical suggestions on transparency, human review, appeal routes for citizens, and ethics checks before deployment, while flagging that these four cases sit in very different legal systems, so the lessons might not travel everywhere. Keywords: Artificial intelligence; public administration; algorithmic accountability; digital governance; welfare policy; criminal justice; comparative case study.