Fair AI in public administration: Estonia and EU law

This article examines fair and trustworthy artificial intelligence in public administration, presenting Estonia as the main case and EU law as the wider legal frame. It builds on the EquiTech project, which has been successfully co-implemented by the TalTech Department of Law and its project partners. The project materials, including training materials for public servants and data scientists, the project proposal, and related academic work, form an important empirical and normative basis for the article. The central claim is that fairness in public-sector artificial intelligence (AI) cannot be reduced to technical accuracy, general transparency, or compliance with a single legal instrument. Wachter, Mittelstadt and Russell’s work on why fairness cannot be automated is the paper’s main theoretical anchor. Their concept of contextual equality helps explain why EU non-discrimination law does not translate easily into fixed fairness metrics, even though statistical evidence remains necessary. In the Estonian setting, this problem is visible in the gap between an advanced digital state and a still developing framework for assessing bias and discrimination risks across the full lifecycle of automated decision-making. The article argues that trustworthy public-sector AI requires legal authority, data governance, bias assessment, meaningful human oversight, administrative contestability, and institutional capacity. The EU Artificial Intelligence Act is a major step in this direction, but it cannot by itself solve the legal and practical problems raised by algorithmic discrimination in public administration.