TLDR: A research paper uses BERTopic and thematic analysis to uncover evolving AI governance themes in EU policies from 2018-2025. It finds a shift from a pre-AI Act focus on ethical AI and general risks to a post-AI Act emphasis on operationalized legal AI, regulatory enforcement, and compliance, while consistent themes include risk and data protection. The study also notes a decreased focus on environmental impacts and the future of work in recent policies.
The rapid advancement of Artificial Intelligence (AI) technologies has brought about a complex and often fragmented landscape of AI governance. In response, the European Union (EU) has emerged as a global leader, particularly with the adoption of its landmark EU AI Act in 2024. However, the EU’s approach to AI governance is shaped by a multitude of documents issued both before and after this pivotal act, including the High-Level Expert Group (HLEG)’s “Ethics Guidelines for Trustworthy AI” from 2019 and various guidelines from the European Commission.
While these EU policies are expected to align with a common vision of trustworthy AI, they often differ in their scope, areas of emphasis, and priorities. To gain a comprehensive understanding of AI governance from the EU perspective, a recent research paper delves into these key EU documents. The study aims to uncover the prevalent themes and concepts within the EU AI policy landscape, tracking the evolution of its approach to addressing AI governance since 2018.
The researchers employed a dual methodology, combining qualitative thematic analysis with quantitative topic modeling. Thematic analysis, a human-driven approach, involved domain experts manually coding a subset of documents, including the EU AI Act and the HLEG Ethics Guidelines, to identify nuanced patterns of meaning. To enhance these results and scale the analysis to a larger corpus of EU AI policy documents published post-2018, they utilized the BERTopic model, an unsupervised topic modeling technique from natural language processing. This combination allowed for both deep, expert-led insights and broader, data-driven identification of recurring topics.
Also Read:
- A Structured Approach to AI Governance: Bridging Regulation and Practical Implementation
- Navigating the Complexities of Human-AI Value Alignment: A Comprehensive Review
Key Findings: The Evolution of EU AI Governance Themes
The analysis revealed a significant shift in the AI policy discourse within the EU following the adoption of the AI Act. Before the Act, the thematic landscape was largely focused on ethical AI, data governance, and risks, particularly those impacting vulnerable groups. This period, from 2019-2020, emphasized aspirational ethical AI, raising awareness of risks, and providing a strategic roadmap for trustworthy AI.
However, the post-AI Act era shows a clear transition from these aspirational ethical principles towards operationalized legal AI. The focus has shifted to regulatory enforcement, with a strong emphasis on concrete compliance tasks such as documentation and risk assessment. The policies now distinguish between different entities involved in the AI value chain – development, deployment, and use – highlighting the need for multi-stakeholder involvement to achieve trustworthy AI. Furthermore, there’s an increased focus on specific AI use cases that pose significant risks.
Themes related to risk and data protection have remained consistently important throughout both periods, reflecting the enduring influence of the EU GDPR. Interestingly, the study also identified themes that appear to have received less attention in the post-AI Act era. These include the environmental impacts of AI and its broader effects on the future of work, suggesting a waning focus on these areas as regulatory enforcement takes precedence.
This study offers a novel perspective on EU policies, tracking how the Union’s approach to AI governance has evolved over time. By combining advanced computational methods with expert interdisciplinary analysis, it provides a holistic and temporal view of EU AI policies, concentrating specifically on AI governance and the criteria for responsible AI development and deployment. For more detailed insights, you can read the full research paper here.
