EU lawmakers are at odds over regulating AI systems like ChatGPT, jeopardizing crucial legislation. Negotiators will meet on Friday for vital discussions before December 6th’s final talks. The main hurdle is the use of ‘foundation models’ or generative AI in the proposed EU AI Act. Foundation models, like those from Microsoft-backed OpenAI, learn from vast datasets but face disagreements in negotiations, even after the European Parliament’s approval in June.
Challenges in Regulating Generative AI: Key Discussions and Risks Ahead
Friday’s discussions will focus on foundation models, source code access, fines, and other topics. Failure to agree risks shelving the act due to time constraints before next year’s European parliamentary elections. Experts debate regulating foundation models, with some suggesting a tiered approach based on a 45 million-user threshold. Others argue even smaller models pose risks. France, Germany, and Italy prefer a self-regulation approach for generative AI makers over strict rules, creating a major obstacle.
In an October 30th meeting in Rome, France persuaded Italy and Germany to support self-regulation, disrupting previous compromises on regulating high-risk AI. Critics, including European parliamentarians, EU Commissioner Thierry Breton, and AI researchers, fear self-regulation won’t meet safety standards. Companies like Mistral in France and Aleph Alpha in Germany criticize the tiered approach, urging strict rules for products.
Legal Uncertainty in Europe: Pending Issues and Potential Consequences
Growing legal uncertainty concerns European industries planning for the coming year. Pending issues include defining AI, fundamental rights impact assessment, law enforcement exceptions, and national security exceptions. Spain, currently holding the EU presidency, proposes compromises. If a December agreement isn’t reached, Belgium’s next presidency may shelve the AI Act before European elections, posing a significant challenge. Mark Brakel, director of policy at the Future of Life Institute, notes a shift from expected compromises on key issues
