Stakeholder Involvement in the Implementation of Artificial Intelligence (AI) in Public Agencies. Use Case Insights from the German Labour and Social Administration.

Implementing AI-based technologies in public administrations has become a major theme on the agenda of governments all across Europe. The use of AI in social administration is particularly controversial among policy-makers, citizens, legal and technology experts. While AI applications in the public sector may fail for various reasons (e.g., financial resources, technical expertise, political majorities, etc.), this study explores the role of stakeholder involvement from the perspective of public agencies. The study addresses how the active involvement of stakeholders in the implementation process can contribute to the successful implementation of AI-based technologies in public labor and social administration. Specifically, we ask, under which conditions does the systematic integration of stakeholders, in the sense of a co-creative technology design process, help to overcome implementation deficits? To answer this question, we proceeded in three steps. First, building on the work of Krafft/Zweig (2019), we develop a refined risk matrix that applies to AI-based technologies in the public sector. The X-axis represents who is affected by an AI-based decision (individuals or collective actors). The Y-axis indicates whether the task, that is delegated to an AI, involves administrative discretion. Next, we propose different stakeholder involvement strategies depending on where a specific AI-based application is located in this novel risk matrix. Finally, we test our theoretical proposition by conducting a series of expert interviews with senior IT managers in the German labor and social administration, in the fields of AI-based employment services and child benefits (Federal Employment Agency) as well as AI-supported recourse proceedings and therapy management (insurance agency).