From efficiency to exclusion rethinking AI governance in public institutions

Abstract

The use of artificial intelligence (AI) in public administration has been promoted as a means to increase efficiency and reduce human bias. However, recent studies reveal that these systems can reproduce and even amplify structural inequalities, thereby undermining fundamental human rights. This article offers a critical analysis of how algorithmic decision-making impacts equality, privacy, and human dignity, based on an interdisciplinary documentary review of normative frameworks, empirical studies, and real-world cases such as the COMPAS algorithm in the United States and the child welfare fraud detection system in the Netherlands. Three critical dimensions are identified: the algorithmic reproduction of historical prejudice, the prevalence of automation bias and selective adherence by public officials, and the inadequacy of current regulatory frameworks such as the European Union’s Artificial Intelligence Act (AI Act) and the General Data Protection Regulation (GDPR). Drawing on the concept of “slow violence,” the study argues that these technologies can imperceptibly erode fundamental rights, particularly among vulnerable populations. The findings support the need to move toward a model of algorithmic governance centered on human rights, incorporating principles of transparency, accountability, public oversight, and access to effective redress mechanisms. Only through comprehensive and enforceable regulation can the risk of a new form of algorithmic exclusion in the public sector be effectively mitigated.