The ICFG report proposes a ‘CERN for AI’ – a centralised institution dedicated to developing trustworthy AI. This initiative aims to boost EU’s competitiveness in advanced AI.
On 4 September 2024, the International Center for Future Generations (ICFG) published a report on the establishment of a CERN for AI, an institution similar to the European Organization for Nuclear Research (CERN) aiming to centralise AI efforts in Europe. The ICFG report underscores the need for Europe to enhance its position in the global AI landscape as it currently falls behind international competitors in computational infrastructure, talent retention, and industry scale. While the EU took an important step in AI regulation with the EU AI Act, Europe remains at a disadvantage to the AI ecosystems in the US, UK and China. To address these challenges, the ICFG report proposes the creation of a CERN-like institution, which builds on the success of the organisation. CERN employs around 70% of all leading particle physicists globally and has become the European success story for global scientific breakthroughs with collaborative, cross-border research. President von der Leyen also called for a European AI Research Council similar to CERN in her Political Guidelines. The Draghi report highlights the organisation as a model for effective EU-level coordination and collaboration among Member States.
The report outlines several key features of a centralised AI institution. First, CERN for AI should have a portfolio approach to AI research. Rather than trying to compete directly with the AI companies that dominate the industry, CERN for AI should focus on a broad range of AI research areas to diversify the advanced AI landscape. Second, the report suggests that such an institution should heavily invest in AI computing power as a continuation of the EU Chips Act to scale its AI capabilities with a targeted €30-35 billion investment. Furthermore, to attract and retain top researchers in AI, CERN for AI should create talent and compute hubs to centralise and pool talent. The authors also address the debate over whether AI models should be open-sourced, concluding that CERN for AI could help create an evidence base to guide legislators on when open-source restrictions are necessary. Further proposed features include multi-level security, leveraging of public-private partnerships, a robust benefit-sharing system and international partnerships.
According to the report, CERN for AI could benefit the EU in various ways. First, such an institution could boost the EU economy by developing more trustworthy and advanced AI than its international competitors. It could also promote investor confidence, further driving the EU economy. Second, a centralised institution would allow for the integration of AI into critical processes and infrastructure. That way, AI systems can help strengthen cybersecurity, detect disinformation and enhance intelligence on terrorist threats. Third, trustworthy AI would ensure democratic oversight. Currently, EU data regulations do not ensure that AI cannot be misused to perform cyberattacks on European targets, for example. By participating in the development of advanced AI, the EU could gain more control of AI security rather than playing a regulatory role. This centralised role of the EU stands in contrast to the more decentralised approach proposed by the Group of Chief Scientific Advisors in their report on the European Distributed Institute for Research, EDIRAS (see SwissCore article).
The ICFG urges Europe to take targeted risks and calls on EU policymakers to act decisively to stay competitive in advanced AI development. The report suggests that by focusing on trustworthy AI, CERN for AI could position Europe as a key player in the global AI landscape.