The impact of machine learning within public organizations relies on coordinated effort over the functional chain from data generation to decision-making. This coordination faces challenges due to the separation between data intelligence departments and operational intelligence. Through theory about knowledge sharing between occupational communities and a case study at a Dutch inspectorate, we explore knowledge boundaries between machine learning developers and end-users and the effects of co-creation. Our analysis reveals that knowledge boundaries are dynamic, with boundaries blurring, persisting, and emerging under the influence of co-creation. Especially the emergence of boundaries is surprising and suggests the presence of a waterbed effect. Furthermore, knowledge boundaries are layered phenomena, with some boundary types more prone to change than others. Understanding knowledge boundaries and their dynamics better can be crucial for improving the intended impact of ML for organizations.

Machine learning (ML) algorithms have now entered public decision-making surrounded by enthusiasm, for the possible positive impact they may have on services and citizens. However, their introduction brings with it numerous problems that are left in the background or not even addressed. Academic contributions are growing, and often discuss general challenges, such as a lack of transparency, a lack of accountability and the issue of discrimination. However, the wickedness of public decision-making and specific public decision-making characteristics are not fully acknowledged in the literature, and the impacts of these characteristics are underexplored. With a focus on public decision-making and Llgorithms in the public sector, in this chapter, we provide a conceptual overview based on a narrative literature review. Specifically, the chapter first offers an overview of public sector decision-making characteristics. After describing our methodology, the study offers an overview of available studies focusing on decision-making with algorithms and decision-making about algorithms. Then, implications in light of specific public sector characteristics are discussed. The main implication is the amplification of existing challenges that exist with both public decision-making and ML algorithms. Finally, some conclusions are drawn.