Artificial Intelligence in Public Administration

Abstract

Governments everywhere are turning to artificial intelligence (AI) to modernise how they work. Automated systems now process benefit claims, flag suspected fraud, decide which businesses get inspected, answer citizens’ questions and, increasingly, help draft the decisions themselves. The appeal is mostly about efficiency: faster processing, lower costs, and a way to cope with caseloads that have long outgrown the staff available to handle them. But the qualities that make AI attractive to administrators (its speed, its scale and its air of objectivity) also put pressure on some of the oldest commitments of administrative law: that power is exercised for reasons, that someone answers for it, and that it stays within the law. This article explores that tension. Using a qualitative doctrinal and policy-analytical approach, it draws on administrative law scholarship, comparative regulatory instruments and a set of well-documented cases: Australia’s “Robodebt” scheme, the Dutch childcare-benefits scandal (the toeslagenaffaire) and the related SyRI litigation, Michigan’s MiDAS unemployment-fraud system, and data-driven welfare targeting in Pakistan and India. I argue that AI accelerates a long-running shift from discretionary, street-level bureaucracy to rule-bound, system-level bureaucracy, and that this shift changes the character of administrative error. Mistakes stop being individual and become systemic, and legal doctrines built to catch errors one case at a time struggle to keep up. The article examines the main uses of AI across government and analyses the resulting problems under three headings: efficiency and its limits as a governing value; accountability and the erosion of meaningful human oversight; and the rule of law, including legality, equal treatment, legal certainty and the right to be heard. It closes with practical recommendations to help public administrations capture the real benefits of AI without giving up the accountability and legal safeguards on which their legitimacy depends.