Emerging AI Discourses and Policies in the EU: Implications for Evolving AI Governance

With the emergence of powerful generative AI technologies and the increasing prevalence of high-risk AI systems in society, the conversation around the regulation of AI has gained critical global traction. Meanwhile, policymakers and regulators are struggling to stay on top of technological advances in AI. Our work uses an interdisciplinary approach combining computational linguistics, law, and sociology to examine the developments in policy discourses around the AI Act (AIA) in the European Union (EU) and their implications for global AI policy. We base our analysis on findings from an ongoing study of multi-stakeholder feedback to the AI Act, leveraging Natural Language Inference (NLI) to examine the language used by diverse stakeholders to justify viewpoints for and against the AIA. Based on the outcomes of that analysis and engaging an AI policy perspective, we examine how those initial discourses are reflected in the amendments to the AI Act during the EU co-legislative process. The study is anchored on identified trends of contentious points in the regulation, such as the definition of AI, general principles, prohibited practices, a tiered approach to foundation models, general-purpose and generative AI, high-risk categorization as well as measures supporting innovation, such as AI regulatory sandboxes. We reflect on the implications of emerging discourses, regulatory policies, and experimental frameworks for global AI governance. Our take is that experimental regulation, such as regulatory sandboxes, can offer a window of opportunity to incorporate a global perspective and produce better-informed regulation and governance of AI.