In the past weeks, France has emerged as a showstopper in the negotiations on the world’s first comprehensive AI law, taking an uncompromising stance in rejecting binding rules for the most powerful models.
The AI Act is a landmark legislative proposal to regulate Artificial Intelligence following a risk-based approach. Since the public launch of ChatGPT last year, which is based on OpenAI’s GPT-4, EU policymakers have been scratching their heads on regulating these powerful ‘foundation’ models.
However, with the support of Germany and Italy, France has opposed any binding obligation for foundation model providers, risking jeopardising the entire AI regulation since this position is not acceptable to the European Parliament.
Europe’s three largest countries are pushing for ‘mandatory self-regulation’, namely in the form of codes of conduct. Leading AI start-ups like the French Mistral AI and the German Aleph Alpha share this position.
“The emergence of European players, including Mistral AI in France, has shifted the perspective”, Agata Hidalgo, European Affairs Manager at France Digitale, told Euractiv, explaining the main objective of the French government has become “to regulate foundation models without limiting [French] innovation capabilities.”
France’s shifted position
France has historically been pushing for the implementation of trustworthy AI systems, exemplified by the Confiance.ai collective, financed by public funds. However, during the summer, France’s position has shifted toward a more free-market approach.
In particular, Arthur Mensch, founder and CEO of Mistral AI, publicly told French President Emmanuel Macron on the main stage of the largest French tech event, VivaTech, that “the current regulation, [the AI Act], at the European Parliament poses a significant risk of hindering innovation”, asking him what he will do to fix this.
A couple of weeks later, dozens of CEOs of Europe’s industry heavyweights like Airbus, Siemens and Mistral AI signed a letter sounding the alarm against the European Parliament’s approach toward regulating foundation models.
Alleged conflicts of interests
Euractiv understands that Cédric O, who worked from March 2019 to May 2022 as digital state secretary, was one of the main drivers of the letter and France’s current stance on foundation models.
As a public official, O advocated favourably for strong EU digital regulations and, in the context of France’s rotating presidency of the EU Council, negotiated the Digital Services Act and the Digital Markets Act, two landmark regulations of Big Tech.
O has been at the centre of public polemics, as detractors accuse him of having favoured regulation while he was state secretary but of having become a fierce opponent since he joined Mistral AI as a co-founding advisor and chief lobbyist.
In a public response to the claims against him, O laments that personal attacks are the “decay of public debate” and argues that he has not changed positions since he was in office.
During the last days of his office, the French presidency proposed regulating versatile AI systems like ChatGPT but not the underlying model, like GPT-4.
What to regulate
For Mistral’s founder Mensch, the debate on regulating foundation models has been mischaracterised as Mistral AI against the world, whilst only last week, tens of associations have expressed a similar position to give European start-ups a ‘fighting chance’.
Instead, Mistral AI leaders stress that the debate should focus on the proprietary versus open-source approaches of AI models. As Mistral AI is open source, they fear it is trying to unlock the market against Big Tech’s closed models.
Yet Internal Market Commissioner Thierry Breton openly stated on 24 November in La Tribune that Big Tech “and the startup Mistral do not defend the general interest” but their own.
At the same time, stakeholders like the European Digital SME Alliance have stressed the importance of regulating foundation models to avoid the bulk of the AI Act’s obligations falling on the downstream economic operators and risking hampering the whole EU industry.
Errors of the past
The French government contends that Europe must innovate before regulating and ensure it does not repeat the mistakes of the last technological revolution of the 1990s.
French member of parliament Éric Bothorel shares this view and told Euractiv, “One should not consider that if Europe missed certain turning points in technology, it is trailing the race. When it comes to artificial intelligence, the game is still open”.
Yet, opponents argue that the French position is strangely similar to US OpenAI’s position on the AI Act.
For these, taking a market-free approach in the EU would be a trojan horse for the EU’s competitiveness and democracy, replicating the errors of the last technological revolution, which eventually led to the creation of Big Tech and major issues in terms of data protection, illegal content online or online market competition.
[Edited by Luca Bertuzzi]