Key Insights:

• EU is at a crossroads with the AI Act, struggling to find a consensus on regulating transformative generative AI technologies like ChatGPT.
• Debate intensifies over self-regulation vs. strict rules for AI, with France, Germany, and Italy advocating for more autonomy in AI development.
• Time-sensitive negotiations in the EU focus on balancing innovation with safety in AI amid looming European parliamentary elections and global impact.

In the rapidly advancing world of artificial intelligence, the European Union finds itself at a crucial turning point. The bloc’s efforts to enact groundbreaking legislation governing the use of generative AI, such as the technology behind ChatGPT, have hit a significant roadblock. The heart of the matter lies in effectively regulating these so-called foundation models.

Regulating Foundation Models: A Central Challenge

Foundation models, AI systems trained on vast datasets capable of learning and adapting, stand at the center of this debate. These models, including those developed by Microsoft-backed OpenAI, are a major hurdle in the EU’s journey to pass the AI Act.

After two years of discussions and a June approval of the draft AI rules by the European Parliament, EU representatives are now in crucial talks. The outcome of these negotiations will shape the future of AI use across the continent. A key point of contention is how to regulate foundation models, with opinions varying significantly among member states.

National Interests and Regulatory Approaches

The dialogue has been notably influenced by diverging views from leading EU nations. France, Germany, and Italy have been vocal in their support for a self-regulation model for AI developers. This perspective solidified during an economic ministers’ meeting in Rome, starkly contrasts the earlier consensus favoring more stringent regulatory measures.

This proposed self-regulation approach has been met with criticism from various corners. A renowned AI researcher, Geoffrey Hinton, expressed concerns, stating that self-regulation “is likely to dramatically fall short of the required standards for foundation model safety.” EU Commissioner Thierry Breton and other European parliamentarians echo this sentiment.

On the other side of the debate, companies like Mistral in France and Aleph Alpha in Germany argue against a tiered regulatory approach targeting models based on user numbers. They advocate for regulations that focus on the products created by AI rather than the technology itself.

Other Key Issues in the Negotiation Process

The regulatory discourse extends beyond foundation models. Definitions of AI, fundamental rights impact assessments, and exceptions for law enforcement and national security are also on the table. One particularly sensitive issue is law enforcement’s use of AI systems for biometric identification in public spaces, which has proven difficult for lawmakers to agree upon.

Currently holding the EU presidency, Spain attempts to facilitate compromises, but time is short. With European parliamentary elections approaching, there’s a pressing need to resolve.

Kirsten Rulf from Boston Consulting Group highlights the critical nature of these discussions, noting; 

“The growing legal uncertainty is unhelpful to European industries. Businesses are eager for clarity as they plan for 2024.”

Mark Brakel from the Future of Life Institute observes a shift in the negotiation climate. “A few weeks ago, compromises on all key issues emerged. Now, the situation has become significantly more complex,” he says.