Algorithm and Administrator : An exploration of responsible administrative practices when working with machine learning models

In light of the call for human measure in public administration (Commissie van Dam, 2020), I explore the tension between the dominant material-discursive practices around machine learning models and the human public official. Central to my exploration is what it means for a public official in executive government to act responsibly in collaboration with machine learning models. Defending that administrators have discretionary space in a digital bureaucracy, responsibility of the administrator is conceptualised based on Rouse's (2015) theory of practice. The theoretical explorations of administrative responsibility are contrasted with an empirical study. This empirical study centres on the administrative practices of responsibility during the public hearings of the Dutch parliamentary inquiry on fraud policy and service provision (Belhaj, 2024), prompted by the social benefit scandal. In the analysis, the risk classification model (risicoclassificatiemodel, RCM) is studied as an exemplar case. I ultimately provide a view of administrative practices with machine learning that hopefully prompts and enables practitioners to reflect on their going abouts with machine learning and ideas of what it means to act responsibly in public administration.