MEPs adopted rules on Wednesday to regulate artificial intelligence (AI) systems like ChatGPT, a unique piece of legislation in the world.
The European commissioner in charge of the file, Thierry Breton, welcomed on X the “massive support” of Parliament (523 votes for, 46 votes against) for the text.
“This will benefit Europe’s formidable talent pool. And will establish a model for trustworthy AI throughout the world,” underlined the President of the European Commission Ursula von der Leyen, referring to “pioneering” legislation.
This bill was presented by the European Commission in April 2021. The appearance at the end of 2022 of ChatGPT from the Californian start-up OpenAI, capable of writing essays, poems or translations in a few seconds, gave it a new dimension.
This system revealed the enormous potential of AI, but also its risks. The dissemination of false photos or videos, larger than life, has thus alerted us to the danger of manipulation of opinion.
“Let the beginning”
French President Emmanuel Macron welcomed the vote, calling it “a first in the world, essential to protect everyone’s rights and data security while supporting innovation.” “It’s Europe that’s doing it!” », underlined on country.
With this text, “we have managed to find a very fine balance between the interest in innovating and the interest in protecting,” also estimated co-rapporteur Dragos Tudorache (Renew, centrists and liberals).
However, this legislation “is only the beginning,” he noted, recalling that artificial intelligence continues to evolve rapidly.
The regulation provides for a two-tiered approach. “General purpose” AI models will have to comply with transparency obligations as well as European rules on copyright.
As for systems considered “high risk” – used for example in critical infrastructure, education, human resources, law enforcement – they will be subject to stricter requirements.
For example, they will have to provide for the establishment of a mandatory impact analysis on fundamental rights.
Artificially generated images, texts or videos (“deep fakes”) must be clearly identified as such.
The text prohibits citizen rating or mass surveillance systems used in China, or even remote biometric identification of people in public places.
On this last point, States have however obtained exemptions for certain law enforcement missions such as the prevention of a terrorist threat or the targeted search for victims.
European legislation will be equipped with means of surveillance and sanctions with the creation of a European AI office, within the European Commission. It can impose fines ranging from 7.5 to 35 million euros, depending on the offense and the size of the company.
“We regulate as little as possible, but as much as necessary,” summarized Mr. Breton.
“Gaps, restrictions and exceptions”
The European Consumer Organization (BEUC), however, considered that “the legislation should have gone further to protect consumers”.
“The final text is full of loopholes, restrictions and exceptions, meaning it will not protect people, or their human rights, from some of the most dangerous uses of AI,” echoed digital rights group Access Now.
For his part, Markus J. Beyrer, director general of BusinessEurope, the voice of European employers, considered that this was a “pivotal moment for the development of AI in Europe”.
The Multinational Observatory (France), Corporate Europe Observatory (Belgium) and LobbyControl (Germany) fear that lobbies will weaken this implementation.
Many details “need to be clarified […], for example with regard to standards, thresholds or transparency obligations. The composition of the advisory board of the new European AI agency also remains unclear,” they warned.
The 27 EU states are expected to approve the text in April before the law is published in Official Journal of the EU in May or June.