Meta is opening a new front in the fight against the regulation of artificial intelligence (AI). In an open letter published this Thursday with around thirty companies including Spotify, Ericsson and EssilorLuxottica, Facebook’s parent company is calling on the European Union to “clarify” its AI regulatory strategy and is asking for “harmonised, coherent, rapid and clear decisions” on data regulation. With a threat: that the continent, “having become less competitive and less innovative than other regions”, will lose ground and see innovation take place elsewhere.
Read alsoVincent Luciani: “Europe will be a leader in regulating AI”
A text supported by Canadian Joëlle Pineau, vice-president of AI research at Meta and general director of the FAIR laboratory (Fundamental AI Research), presented this Thursday in Paris. “The main problem at the moment is the fragmentation of regulation in Europe and around the world. This brings complexity and uncertainty: when there is no clear regulatory framework, we don’t know which way to turn!”
Joëlle Pineau, vice-president of AI research at Meta, during a visit to Paris in June 2023. Credit: AFP / ALAIN JOCARD
Data, a crucial but sensitive commodity
Voted in the spring, the AI Act, the main regulatory text for AI in Europe, came into force on August 1 and will be gradually applied by 2026. But, according to Joëlle Pineau, “there are still many uncertainties, particularly in its concrete translation and its precise specifications.” “Uncertainties” combined with the weight of other regulations, including that on data protection (GDPR), which led Meta to be postponed without the the launch of its new generative AI interface in the EU – following Apple’s lead in doing the same for its AI features, dubbed “ Apple Intelligence ».
This standoff between tech giants and local authorities is not limited to EU borders. In early July, Meta was ordered by the Brazilian authorities to stop using the personal data of its platform users to train its AI models: after two months of fighting, the company finally change your privacy policy and now leaves the choice to users. A strategy designed to gain the regulator’s favor. In the meantime, Meta had suspended its generative AI tools in the country – wasting valuable time on research and adapting models to the local context, for which data is crucial.
Read alsoAI regulation: are innovation and competitiveness in danger?
“The models take months to train, and sometimes you have to start from scratch,” regrets Joëlle Pinaud, who is in favor of “complete transparency” in terms of data use and research in open source – two ideals today challenged by the growing competition between tech giants.
“I am in favor of international alignment”
The fact remains that regulation of artificial intelligence and the use of personal data seems necessary. In parallel with the open letter published this Thursday, UN experts have called for not abandoning the development of artificial intelligence and the management of risks associated with the “whims” of the market. And they are calling for international cooperation tools, such as a group of scientific experts operating on the IPCC model.
A vision that Joëlle Pineau does not oppose. “I am in favor of international alignment, as long as the risks are defined precisely,” she says, emphasizing the collaboration between Meta and American institutions such as NIST (National Institute of Standards and Technology) to define appropriate standards and criteria for evaluating and regulating AI models.
Faced with technological developments, regulation is in any case called upon to evolve. Among Meta’s research projects are “agents”, AI programs capable of performing actions by interacting with other programs, but also predictive models, baptized JEPA (« Joint Embedding Predictive Architecture »). Models that would no longer seek to generate a text word by word, as current generative AI models do, but through reasoning and understanding. Like a human brain. Research carried out in part in the Parisian premises of Meta, where a branch of FAIR, the laboratory created by the Frenchman Yann Le Cun, is located.
