Uncovering AI Governance Themes in EU Policies using BERTopic and Thematic Analysis

The EU's AI governance approach has evolved significantly, with the research tracking how priorities and emphasis have shifted across documents from 2018 onward, suggesting that EU policy on AI is not static but continuously developing in response to emerging challenges.

Key EU documents like the AI Act and HLEG Ethics Guidelines show both alignment and differences in their scope and priorities, indicating that while the EU aims for consistency, different policy instruments may emphasize different aspects of trustworthy AI—a finding relevant to organizations trying to comply with multiple EU requirements.

The research identified prevalent themes across EU AI governance documents using both qualitative analysis and computational methods (BERTopic), revealing that the EU's approach to AI governance encompasses multiple interconnected concerns rather than focusing on a single issue, which helps stakeholders understand the breadth of the regulatory landscape.

The fragmented landscape of AI governance policies creates complexity, as noted in the paper's opening—this fragmentation means that organizations and policymakers must navigate multiple, sometimes overlapping frameworks, underscoring the need for clearer harmonization efforts in responsible AI governance.

Computational analysis of policy documents can reveal governance priorities and trends that might not be immediately obvious from reading individual policies, suggesting that data-driven approaches to analyzing policy can help stakeholders better understand the actual emphasis and direction of AI regulation.