Policy Integration for Realists: Insights from Actor Preferences for Artificial Intelligence Regulation

Complex and wicked policy problems that affect different policy systems as well as different sectors of the economy are an increasing challenge for public governance. As a response, experts and policymakers suggest integrated policy responses, which include collaborative governance across policy sectors. A recent example of such a problem is artificial intelligence (AI). Scholars agree that AI is relevant for multiple public sector organizations, and have repeatedly called for encompassing, coherent, and effective regulation to govern its diverse uses across different fields. Nevertheless, other insights from the literature on public policy and regulatory governance suggest that regulation and policymaking are typically quite sectoral. This means that regulations usually focus on one or two sectors, while it remains very difficult to design and implement rules that efficiently span multiple policy domains. In this paper, we empirically examine how diverse actors from three different political systems (France, Germany and Switzerland), prefer the regulation of AI applications at the national level. We also investigate which types of policy issues diverse policy actors would like to see horizontally integrated into a single AI regulation, and to what extent these actors believe that such integrated and comprehensive regulatory frameworks can be realistically implemented. Our analysis draws on original elite survey data comprising over 190 organizations in the three countries, including organizations and individuals engaged in AI governance across banking, health, and social welfare subsystems, as well as actors specialized in AI development and applications. A significant share of the actors in our survey considers an encompassing or integrated approach to AI regulation important. Nevertheless, when it comes to the practical implementation of such a policy approach, most respondents believe that it is not feasible to enforce a fully integrated regulatory framework, and that sectoral policies are more realistically implementable. These insights contribute to the literature on AI governance by indicating that realistic regulation will likely focus on specific issues rather than adopting an all-encompassing approach. More generally, the paper provides policymakers with a realistic perspective on the actual feasibility of implementing horizontally integrated policy programs (e.g., collaborative governance across multiple sectors). Although decision-makers often demand such comprehensive approaches, they are very difficult to implement in practice. The results from this paper therefore support a call for modest, targeted regulation and public policy rather than complex and comprehensive programs. We would like to submit the paper to this panel: Collaborative Governance: Policymaking and Cooperation in a Complex World