AI in Public Administration: Efficiency Without Losing Trust

How Can Automated Decision-Making Be Used in the Public Sector Without Undermining Legal Certainty and Public Trust?

AI-driven decision processes are gaining ground in Nordic welfare states, promising to simplify and streamline public administration. At the same time, they pose major challenges concerning transparency, legal certainty, and trust.

These shifts in the relationship between the welfare state and its citizens are examined in the interdisciplinary research project The Automated Administration: Governance of Automated Decision-Making in the Public Sector, led by Stefan Larsson, Associate Professor at Lund University. The project focuses on governance, law, and policy development in Sweden and Finland.

“We want to understand how the state can use AI in a way that is both efficient and fair, without citizens losing confidence in public authorities,” says Larsson.

Technology Outpacing Legislators

Both technology and legislation are evolving at breakneck speed. The EU’s AI Act has changed the playing field, and researchers are studying how this regulation affects national governance. A key challenge is the so-called pacing problem – the fact that rapid technological development creates a moving target for lawmakers.

“Legislators want to regulate AI but leave much open for later adjustments. This is understandable given howgenerative AI made a huge leap during the legislative process, but it challenges the need for legal predictability,” Larsson explains.

According to him, it is particularly interesting that the research project has taken place during such a formative period. The team has identified national differences, with Sweden and Finland pursuing different strategies.

“Finland was quick to introduce AI guidelines, while Sweden has historically been more cautious. Despite a common EU framework, the law takes different shapes in the two countries, and we are exploring what that means.”

New Methods for Physical Prototyping

The research group has worked across disciplines, combining political science, legal studies, and other social sciences. One example of methodological innovation was a workshop in Helsinki where the concept of a ‘kill switch’ in AI systems was explored through physical prototyping, led by Associate Professor Riikka Koulu. A kill switch (also called an emergency stop button) is a feature that allows an AI system to be shut down if it behaves unexpectedly.

“It was a new way of approaching legal theory. Embodying concepts with tape and sticks sparked different ideas than we get at the desk. These often human-shaped but metaphorical understandings of what AI is, and how it can or cannot be regulated, ultimately shape legislation – with all its strengths and weaknesses.”

The researchers have also studied how AI affects citizens’ interactions with the state, for example through chatbots. They have followed the establishment of Sweden’s new Payments Authority (Utbetalningsmyndigheten), which, despite its name, investigates irregularities rather than deciding on benefit payments. According to Larsson, this reflects both conceptual and practical shifts in how technology’s role is framed and described, as well as broader changes in the citizen–state relationship:

“We must not forget who public administration is meant to serve when studying trust and fairness in the meeting between technology-driven change and its legal governance.”

The project maintains close contact with public authorities and has organised several seminars to foster dialogue between research and practice. The goal is to contribute to a public administration where AI is used responsibly, with the citizen at the centre.

More information:
Project Leader: Stefan Larsson,
Associate Professor at Lund University
Email: stefan.larsson@lth.lu.se