Anne Bouverot, co-chair of the Council on Artificial Intelligence and Digital (CIANum), is advocating for the creation of a European “CERN for AI”. In an interview in mid-October, she argues that Europe has few levers to exert influence alone against the United States and China, and that corporate responsibilities need to be better assigned.
While concerns around artificial intelligence seem to be subsiding, players in the sector are continuing their forward race, and states have not made much progress despite strong positions. Between the United States, which is letting things be for now, and China, which seems to control everything, Europe has chosen the path of regulation with the AI Act. But, according to Anne Bouverot, “we have little capacity to really regulate AI”. “If Europe regulates but the players are in the United States and China, the impact of the regulation is less,” she explains, pointing to the standoffs initiated by tech giants against European regulations. A global regulation, she says, “will take a lot of time”.
Assigning responsibilities and auditing
For the chair of the board of directors of ENS, one avenue is to assign responsibilities. She finds interesting the way regulation is advancing in the United States on the effects of social networks, while acknowledging that the process has been long, because it is difficult to demonstrate responsibilities there. In the field of AI, “the attribution of responsibility in AI-related cybersecurity issues, in agent issues, is really an avenue that can allow for faster progress”.
She takes the example of agents that might have escaped and caused damage. “They were developed by company employees. The employees gave an objective to these agents. Several levels of the company’s direct responsibility can be called into question,” she states. This responsibility could have “both economic consequences and potentially criminal consequences, in the event of serious damage”.
Anne Bouverot also believes that audits are a key element, provided they are designed correctly. According to her, it is necessary to “refer to rules set in advance, with a number of control and approval mechanisms”, with independent specialists, on the model of the major firms in the financial field. This layer of rules “does not exist in the field of AI and needs to be built”. It could be developed by a hybrid entity, inspired by the IFRS foundation, bringing together professionals and state representatives. “Ideally, it should be a public-private consortium that defines this,” she says, warning against the risk of this structure being dominated by “OpenAI and Anthropic, which clearly have this desire to form an oligopoly”.
A CERN for AI to exist
Anne Bouverot acknowledges that “the reality of AI today, even though we have players in France and in Europe, is developing in the United States and China”. Europe has high-performance and decarbonized energy sources, but only has “3% of the world’s data centers”. It also lacks leading players: “Once we’ve said we have Mistral, what else do we have?”
To retain its researchers, who often leave for DeepMind, Anthropic or OpenAI, she champions the idea of a “CERN for AI”, a non-profit structure bringing together “150, 200 or even 300 people”. Its resources would be used “essentially to pay for computing power”, which represents “the most significant cost”. This structure should be European, because “no country in Europe can afford this alone”. She also mentions the need to have “GPUs”, these processors used to train models, and European cloud players.
Anne Bouverot believes that “the next two years are absolutely key” for Europe to demonstrate its sovereignty in models and computing power, in agreement with Arthur Mensch, co-founder of Mistral, but she warns that “it will take us much longer”. In the meantime, Mistral and CERN have signed a “declaration of intent to promote AI for science and to strengthen Europe’s scientific and technological capabilities”.
