Uncovering AI Governance Themes in EU Policies using BERTopic and Thematic Analysis
This paper analyzes the evolution of EU AI governance by combining qualitative thematic analysis with quantitative BERTopic modeling to identify and compare prevalent themes across key policy documents, ranging from the 2018 HLEG Ethics Guidelines to the 2024 EU AI Act.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the European Union (EU) as a massive, busy kitchen trying to figure out how to cook with a new, powerful, and sometimes unpredictable ingredient: Artificial Intelligence (AI).
Over the last few years, the EU has written a huge stack of cookbooks, rulebooks, and safety manuals to make sure this ingredient is used safely. But with so many documents written at different times by different groups, it's hard to see the big picture. Are they all saying the same thing? How have the rules changed?
This paper is like a team of detectives (a mix of legal experts, computer scientists, and social scientists) who decided to read through this entire stack of paperwork to find the main stories hidden inside.
The Detective Tools: Human Eyes and Robot Brains
To solve the mystery, the team used two different tools, kind of like using both a magnifying glass and a metal detector.
- The Magnifying Glass (Thematic Analysis): First, human experts read a few of the most important documents (like the big new "AI Act" and the older "Ethics Guidelines") very carefully. They highlighted key ideas and grouped them into themes, just like a teacher grading essays and writing notes in the margins. This is great for understanding the nuance and deep meaning, but it's slow and hard to do for hundreds of documents.
- The Metal Detector (BERTopic): To handle the rest of the massive pile of documents, they used a smart computer program called BERTopic. Think of this as a robot that can read thousands of pages in seconds. It doesn't "understand" the text like a human, but it's excellent at spotting patterns. It groups words that appear together often, effectively saying, "Hey, these sentences all seem to be talking about the same thing."
The team then combined these two methods. They let the robot find the patterns, and then the human experts checked the robot's work to make sure the patterns made sense and to give them proper names.
What They Found: A Story of Two Eras
The most exciting discovery was how the EU's "cooking style" changed over time. The team split the documents into two eras: Before the Big Law and After the Big Law.
Before the Big Law (The "Dreaming" Phase):
In the earlier documents (around 2019–2020), the EU was mostly talking about ethics and ideals. It was like a group of chefs discussing the philosophy of food: "We should be kind," "We must protect the vulnerable," and "Let's make sure everyone trusts our cooking." The focus was on abstract concepts like "fairness," "bias," and "trustworthy AI."After the Big Law (The "Enforcement" Phase):
Once the EU AI Act was passed in 2024, the conversation shifted dramatically. The documents stopped just dreaming and started talking about rules and enforcement. The focus moved to concrete tasks like "certification," "documentation," "risk assessment," and "monitoring."- The Shift: It went from "Let's be good" to "Here is exactly how you must prove you are good, or you will be fined."
- The Supply Chain: They also started paying attention to the whole "kitchen crew." Instead of just looking at the final dish, they started regulating everyone involved in making the AI—from the people who built the model to the people who deployed it.
What Got Lost in the Shuffle?
While the new rules are very strict about safety and legal compliance, the team noticed that some important topics started to fade away in the newer documents.
- The Environment: The impact of AI on the planet (like energy use) became less of a talking point in the newer, stricter laws.
- The Future of Work: The conversation about how AI might change our jobs or the nature of work also seemed to take a backseat to the immediate legal requirements.
The Bottom Line
This paper shows that the EU has successfully moved from talking about what AI should be, to writing the rules for what AI must do.
By using a mix of human wisdom and robot speed, the authors created a clear map of how the EU is trying to govern AI. They found that while the core goal (safety) remains the same, the approach has evolved from a gentle, ethical guide into a hard-hitting, detailed legal framework. This helps us understand not just what the rules are, but how the EU's thinking has changed as AI has become more powerful and real.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.