Negotiations among EU lawmakers regarding the regulation of systems akin to ChatGPT have hit a roadblock, jeopardizing pivotal legislation designed to govern artificial intelligence (AI). According to six sources divulging information to Reuters, ‘foundation models,’ particularly generative AI, have emerged as a key stumbling block in discussions over the proposed AI Act.
As negotiators convened for crucial discussions on Friday in preparation for final talks scheduled on December 6, the focus on foundation models, exemplified by those developed by Microsoft-backed OpenAI, has intensified. These AI systems are trained on extensive datasets and possess the capability to learn and perform various tasks with new data.
The European Parliament greenlit the bill in June after two years of negotiations. Now, the draft AI rules necessitate agreement through discussions among representatives from the European Parliament, the Council, and the European Commission. However, the impasse centers around key issues such as foundation models, access to source codes, and potential fines. The lack of consensus puts the AI Act at risk of being shelved due to time constraints before the upcoming European parliamentary elections.
While some experts advocate for a tiered approach, imposing regulations on foundation models with more than 45 million users, others argue that even smaller models could pose risks. France, Germany, and Italy are the primary dissenting voices, advocating for self-regulation by makers of generative AI models rather than stringent rules.
Criticism of this self-regulation approach has been voiced by European parliamentarians, EU Commissioner Thierry Breton, and numerous AI researchers. In an open letter, researchers, including Geoffrey Hinton, caution that self-regulation may fall short of ensuring the safety standards required for foundation models.
France-based Mistral and Germany’s Aleph Alpha have expressed dissatisfaction with the tiered regulatory approach, garnering support from their respective countries. Despite efforts to keep negotiations on track, legal uncertainty is growing, creating challenges for European industries looking for clarity in planning for the coming year.
Remaining issues in the negotiations include defining AI, conducting fundamental rights impact assessments, law enforcement exceptions, and national security exceptions. Disagreements persist on the use of AI systems by law enforcement agencies for biometric identification in publicly accessible spaces.
Spain, holding the EU presidency until the year-end, has proposed compromises to expedite the process. However, if a deal is not reached in December, the next presidency, Belgium, will have a limited window before the legislative process is likely shelved ahead of European elections.
Mark Brakel, director of policy at the Future of Life Institute, noted the increasing difficulty in reaching compromises, marking a challenging turn in what was once seen as an emerging consensus on key AI regulation issues.
