Administration med algoritmer

Autonomi og kompetencekrav ved brug af store sprogmodeller i dansk offentlig administration

Abstract

Administering with Algorithms: Autonomy and Competence Requirements in the Use of Large Language Models in Danish Public Administration
This article presents an empirical study of administrative employees in the Danish state sector that examines which employee profiles emerge from perceived barriers to the use of large language models (LLMs) and what these profiles imply for perceived autonomy and competence requirements. Drawing on a questionnaire survey among members of HK Stat (n = 620), which is a trade union, we conduct a latent class analysis of nine barrier indicators and identify four profiles: cautious optimists, curious pragmatists, experienced critics, and skeptical rejecters. Supplementary analyses of enabling conditions and rule clarity further nuance the profiles and underpin a discussion of the consequences of these technologies for different groups of employees. The analysis is theoretically anchored in scholarship on technological well‑being, a sociotechnical lens on organizational hindrances, and the concept of professionalism. The findings show that the relationship between human expertise and AI is shaped by local rules and the division of labor: where support, time, and clear rules are present, LLM use is more often associated with autonomy and competence development; where demands and obstacles dominate, respondents report constrained autonomy and heightened competence requirements. The article concludes with implications for AI implementation and employee autonomy in the public sector.