Governing Artificial Intelligence in Europe

• In Event: The Global Governance of Artificial Intelligence

Abstract

The European Union (EU) AI Act represents the first major regulatory framework on artificial intelligence (AI) and is widely expected to influence global standards. The adoption of the act in 2023 follows a lengthy negotiation process. This article offers the first systematic analysis of these negotiations, focusing on bargaining success and its sources. Theoretically, we draw on spatial bargaining models and develop an argument that highlights preferences, technical expertise, and the level of engagement as determinants of bargaining success. Empirically, we utilize an original dataset that records the preferences of all EU member states on more than thirty issues debated during the negotiations, including high-risk AI, biometric surveillance, and so called foundation models. We conduct a statistical analysis using member states’ preference attainment as a measure of bargaining success. Our findings are threefold. First, we observe varying levels of engagement, with initial complacency followed by intense interest and contestation, particularly after the launch of ChatGPT in 2022. Second, the negotiations were marked by a central battle over how to balance AI innovation with AI safety. Third, while the overall distribution of gains was remarkably even, states with larger and more advanced AI industries were more influential on key issues.

Authors

• Johannes Geith, Stockholm University

• Magnus Lundgren, University of Gothenburg

• Jonas Tallberg, Stockholm University