France bets big on open-source AI – POLITICO

France has a dream: to make a name for itself in the surging global artificial intelligence industry. 

France also has a problem: It’s right in the heart of the EU, currently known more for regulating AI than for encouraging it.

To carve out a spot in that tricky landscape, French leaders are now hoping to foster one particular segment of the industry, called open-source AI.

“Open-source” computer code — publicly posted to be used and repurposed by anyone — straddles the line between the public interest and a private-sector product. Sometimes developed by universities, sometimes by companies, open-source AI systems are now playing a growing role in the industry. For example, Meta’s powerful LLaMA2, an AI model released in July, is open-source.

In June, French President Emmanuel Macron announced new funding for an open “digital commons” for French-made generative AI projects, a €40 million investment intended to attract significantly more capital from private investors. “On croit dans l’open-source,” Macron stressed in his speech at VivaTech, France’s top tech conference: “We believe in open-source.”

A matter of national pride

There’s a bit of pride involved as well: Officials see it as a way to take on the overwhelming power of U.S.-based firms in the AI industry.

“We don’t want to live in a world with two or three or four monopolies and to have to negotiate the rights to innovate. So open-source can be a very important answer,” said Henri Verdier, the French ambassador for digital affairs and the country’s top tech diplomat.

France’s open-source focus comes as part of a hard push toward developing a domestic, francophone AI industry. At the same event, Macron said France would invest €500 million in creating AI “champions” — market and research leaders in the emerging technology.

One of its existing champions is an open-source AI firm. In June, Paris-based startup Mistral.ai — whose French founders hail from U.S. tech giants including Meta and Alphabet’s Deepmind — raised a whopping €105 million in funding by promising to create an open-source competitor to OpenAI’s ChatGPT. The firm’s backers include Cedric O, the former digital minister for Macron’s government.

Alexandre Zapolsky, a co-founder of French tech firm Linagora who, ahead of Macron’s set-piece announcement, had co-written a newspaper column calling on France to foster its open-source AI ecosystem, saw Macron’s speech as a major signal to investors, as well as his own administration.

“Our president endorsed open-source AI — and became a promoter of it — while speaking in front of over 2,000 of France’s top technology entrepreneurs and investors,” Zapolsky said. “And his message has been heard by all the layers of the French government.”

Following the speech, Zapolsky’s co-founder Michel-Marie Maudet launched OpenLLM France, a collective of developers and researchers collaborating via messaging platform Discord to build open-source Al. 

France has some academic strengths to build on. It has also been pitching itself as the best place in Europe to train power-intensive advanced AI models because its nuclear power plants offer cheap and abundant electricity. Irene Solaiman, policy director for leading open-source AI provider Hugging Face, said that France was “exceptional in the EU in having labs that develop high-quality language models.”

Some of the top minds in this field are already French nationals — including Meta’s Chief AI scientist, Yann LeCun — but that doesn’t mean it will be easy to attract talent in an extremely competitive industry. “The U.S. has a lot going for it. Like, it has really stellar academic institutions that work on a lot of the research that’s relevant to the field. It has a lot of cloud [computing] providers,” Solaiman said.

A Continent-wide opportunity

In embracing open-source, France is hoping to take advantage of an EU loophole that might offer a friendly regulatory lane for open-source systems. The bloc is currently finalizing its Artificial Intelligence Act, which would ban some AI uses and create obligations for those deemed risky.

The European Parliament, in its version of the AI Act, exempted open-source AI systems from following the strict compliance rules imposed by the law. Kai Zenner, chief policy assistant to Axel Voss, an influential German member of the European Parliament, says that EU governments support this approach, which suggests “chances are quite high” it will make it to the final version of the law. (The AI Act’s final text, expected to pass in late 2023, is currently being negotiated by representatives of European governments and the European Parliament.)

Europe’s Parliament sees open-source as an AI opportunity not just for France, but for the whole Continent. “We completely agree with the French assumption: We see open-source AI as a big chance,” Zenner said. “If Europe really wants to catch up with the United States and China in AI, then without drawing on models or data sets from the open-source community, we would never have a chance.”

Industry skepticism

Industry leaders, though, aren’t so sure the EU law will give them enough running room. The proposed exemption does not apply when open-source AI is used for commercial purposes, which would likely discourage investors and startups in the space. Members of the open-source AI ecosystem — including Github and Hugging Face — have asked European policymakers for more clarity on what constitutes commercial activity when it comes to making open-source AI components available to the public.  

They also worry that so-called foundation models — the big software engines powering generative AI tools such as ChatGPT — would separately have to abide by a set of obligations under the EU law whether they’re open-source or not. This worries tech giants as much as it does open-source startups.

“The latest amendments from the European Parliament — they seem to impose potentially some pretty complex and potentially somewhat unworkable conditions on open-sourcing large language models altogether,” said Nick Clegg, Meta’s president of global affairs. 

For France and other European Union economies, it feels like a big piece of the future is at stake. Despite being home to world-class universities and talent, European leaders have spent decades watching their countries fail to capitalize on various waves of tech innovation, with the riches going to giants in the U.S. and China. The EU, meanwhile, has established itself more as a technology rulemaker than as a creator and exporter. Policymakers now are determined not to let the same thing happen with AI.

Cedric O, the former French digital minister turned Mistral.ai’s shareholder and adviser, says that Europe has one other advantage when it comes to developing open-source AI. Unlike the U.S., it lacks powerful corporate actors lobbying against the open-source model on security grounds. 

“Europe has the ability to be part of the AI race,” O said. “I would say that — regardless of the fact that its AI is open-source or not — Europe has to do whatever it can to be part of the game.”

Laura Kayali contributed reporting.