France warns against killing a European ChatGPT

PARIS — France has a message: Don’t kneecap a potential future European ChatGPT with too much regulation. 

The European Parliament’s position on the EU’s Artificial Intelligence Act — a piece of legislation currently under negotiation — is too stringent and risks doing just that, France’s Digital Minister Jean-Noël Barrot told POLITICO in an interview Monday. 

As an example, he pointed to Google’s decision not to launch its chatbot Bard in the EU: “The objective is not to see non-European dialogue systems set up in Europe, but rather to see European ones develop. However, we must take these signs into account and avoid taking Europe out of the technology history.”

When it comes to Big Tech, France previously has been on the heavy-handed side of regulation, seeking to impose new rules on U.S. cloud giants before the EU, for example, and coming up with a flurry of content-regulation laws.

On so-called generative AI — referring to tools that can create new content, like text or images, using prompts — the tune is quite different. France hopes to foster homegrown companies to compete with companies like Google and OpenAI, the firm behind ChatGPT. A French startup called Mistral AI, founded by French engineers coming from Google and Facebook, is reportedly raising €100 million to develop ChatGPT-like models.

French politicians, including President Emmanuel Macron, have called for a middle ground between regulation and innovation in recent weeks. Earlier this month, former Digital Minister Cédric O wrote an op-ed arguing European Parliament’s position on the AI Act “de facto prohibits the emergence of European [large language models].”

In May, European lawmakers voted to hold “foundation models” that create content from limited human input, such as ChatGPT, to higher standards of transparency and human rights compliance. They also want developers to partly disclose what copyright-protected material was used to train them.

Both Parliament and Council have agreed on their respective versions of the AI Act, and they now need to sit down together to negotiate a common text. The Council’s position doesn’t include specifics on generative AI, as representatives of EU countries voted before the ChatGPT craze.

“The European Parliament’s position seems excessive at a time when we have a pressing obligation to develop generative AI models in Europe over the coming months, in order to be autonomous and not have to depend on non-European models in the years and decades to come,” Barrot said, arguing that chatbots such as ChatGPT would have to abide by some of the same rules as high-risk systems in areas such as health and transportation.

Overall, Barrot welcomed the AI Act but added it couldn’t solve the issues AI raises, such as privacy, disinformation and intellectual property rights. “But that doesn’t mean they shouldn’t be addressed,” the French minister said. He argued that multilateral organizations such as the G7 were relevant places to have such conversations “beyond the EU.”

In May, G7 countries decided to launch a working group on generative AI that would focus on governance, copyright, transparency and responses to foreign information manipulation, including disinformation.