TL;DR:
- Recent EU AI regulation negotiations hit an impasse due to disagreements over rules for foundation models.
- The AI Act aims to regulate Artificial Intelligence with a risk-based approach.
- Foundation models, including GPT-4, pose a challenge in determining their regulation.
- EU Parliament favored a tiered approach for foundation models, but large European countries opposed it.
- Start-ups like Mistral and AI company Aleph Alpha are lobbying against extensive regulation.
- A re-evaluation of foundation model provisions is proposed, potentially jeopardizing the entire law.
- Failure to reach an agreement could hinder the EU’s global AI regulatory leadership.
Main AI News:
In the realm of European AI regulation, the recent technical meeting held on November 10th came to an abrupt halt as major EU nations called for a retraction of the proposed approach concerning foundation models. Unless this deadlock is resolved in the coming days, the entire legislative process hangs in the balance.
The AI Act, a groundbreaking bill aimed at governing Artificial Intelligence through a risk-based framework, is currently in its final phase of legislative scrutiny. Key EU institutions are engaged in trilogies, gathering to finalize the provisions of this landmark legislation.
Foundation models have emerged as a significant point of contention in these late-stage negotiations. With the proliferation of ChatGPT, a popular chatbot employing OpenAI’s potent GPT-4 model, European policymakers are grappling with the best approach to encompass this category of AI within the forthcoming law.
During the last political trilogue on October 24th, there appeared to be consensus in favor of instituting rules for foundation models following a tiered approach. This strategy, mirroring the Digital Markets Act (DMA) and Digital Services Act (DSA), was seen as a concession from the European Parliament, which had initially advocated for uniform rules governing all foundation models.
The rationale behind the tiered approach was to impose stringent obligations on leading providers, especially those outside Europe. Nevertheless, this approach has encountered mounting opposition from influential European nations.
Recently, the Spanish presidency circulated an initial draft endorsing the tiered approach for internal feedback. In response, the European Parliament’s co-rapporteurs introduced some adjustments while preserving the overall structure of the provisions.
However, at a meeting of the Telecom Working Party on Thursday, a technical body within the EU Council of Ministers, representatives from several member states, notably France, Germany, and Italy, adamantly opposed any form of regulation concerning foundation models.
Leading the charge against regulation for foundation models is Mistral, a French AI start-up challenging Big Tech. Cedric O, France’s former state secretary for digital, leads Mistral’s lobbying efforts, contending that the AI Act could spell doom for the company.
Simultaneously, Germany is under pressure from its own prominent AI company, Aleph Alpha, which is believed to have high-level connections within the German establishment. These companies all fear that EU regulation could put them at a disadvantage compared to their US and Chinese counterparts.
Despite the Spanish presidency’s efforts to broker an agreement with the European Parliament, faced with these staunch positions from political heavyweights, the Spaniards have proposed a comprehensive reevaluation of the provisions concerning foundation models.
When questioned for over an hour and a half regarding the rationale for this change of direction, arguments put forth included the view that the tiered approach might result in a ‘regulation within regulation’ and potentially stifle innovation and the risk-based approach.
Interestingly, the European Commission initially advocated for the tiered approach, positioning itself to enforce regulations on foundation models. However, the Commission did not defend this approach during the technical meeting.
Representatives from the European Parliament concluded the meeting prematurely, stating that “there was nothing else to discuss.” It is understood that regulating foundation models is a non-negotiable stance for parliamentarians, and without it, an agreement seems unattainable.
“The ball is now in the Council’s court to come up with a proposal,” revealed a parliamentary official to Euractiv, emphasizing that the presidency lacks an alternative solution to the tiered approach.
Furthermore, a growing faction within the reluctant member states opposes the AI Act in its entirety, viewing it as excessive regulation. If a resolution is not reached soon, the entire law may be in jeopardy.
Policymakers within the EU were poised to finalize a political agreement at the next trilogue scheduled for December 6th, meaning that the landing zones for the most critical aspects should have been visible by the end of November.
However, if an agreement remains elusive in December, the outgoing Spanish presidency may lose the incentive to continue the technical work. The upcoming Belgian presidency would then have only a few weeks to wrap up this complex file before the European Parliament dissolves for the EU elections next June.
Furthermore, reevaluating the approach to foundation models would necessitate a significant revision of the regulation’s governance structure and provisions for responsibilities along the AI value chain—a task that may be time-prohibitive.
When the AI Act was initially proposed in April 2021, the EU held a leading position in setting global standards for regulating Artificial Intelligence. However, as interest in AI has surged, policymakers in the US, UK, and China have become increasingly active.
Failing to reach a consensus on the EU’s AI regulations during this mandate could not only result in a loss of momentum but also diminish Brussels’ influence compared to other jurisdictions.
The Telecom Working Party is scheduled to convene again next Tuesday, and a technical meeting among EU co-legislators is set for the same day. It is understood that negotiations have now reached the highest political levels in an effort to break the deadlock.
“The AI Act is on the line now,” cautioned a third EU official to Euractiv. “It’s now or never.”
Conclusion:
The deadlock in EU AI Act negotiations reflects the struggle to strike a balance between regulating powerful foundation models and fostering innovation. The opposition from influential countries and industry players signifies the challenge of harmonizing AI regulation across Europe. This uncertainty may slow down market developments as companies await a resolution, impacting the EU’s competitive standing in the global AI landscape.
