The European Parliament has just adopted the AI Act, a text intended to regulate the activity of generative AIs such as ChatGPT, Bard or Midjourney but also, and above all, to combat all the excesses linked to artificial intelligence.
Prevention is better than cure. To identify, contain or even ban abuses linked to the deployment of artificial intelligence in our daily lives, Europe wants to adopt a specific legal framework. THE European Parliament endorsed Thursday, May 11 a 144-page text proposed in April 2021 by the European Commission. After the GDPR, and the recent DSA and DMA, AI Act (Artificial Intelligence Act) is part of the world-leading policy of European legal regulation. It aims to protect us, French and European citizens, by preserving our security and our rights from the dangers of AI, while not curbing technological innovation either. “Europe wants an ethical, human-based approach”sums up Brando Benifei, rapporteur for the text in the European Parliament.
This proposal for a regulation presents “four levels of protection: what is prohibited, what is high risk, what is medium risk and what is authorized”, adds Thierry Breton (European Commissioner for the Internal Market) quoted by France Inter. What should be prohibited first of all… We are talking here about population control systems like the scoring system or “social credit” massively deployed in China today and contrary in all respects to the fundamental values of the European Union. Also in the sights of Europe, which could pose some problems of adaptation to certain member countries (including France), systems for the remote biometric identification in real time of people in public places by the forces of order, for repressive purposes.
Another subject that could be banned is the massive collection of photos via social networks or video surveillance (without the consent of the persons concerned) to train AIs and algorithms and ultimately create databases of facial recognition. AIs that may use subliminal techniques to manipulate or harm the physical or mental health of individuals should also fall under the law. So much for the bans.
AI Act: impose human control of AI
But the AI Act has many other ambitions. It is above all, by highlighting the principles of transparency, traceability, security, non-discrimination and respect for the environment, to impose on AI systems, control, human validation, provide some kind of human control over the machine and inform the public, the general public, and therefore first and foremost the Internet users that we are. Regarding, for example, generative AIs such as ChatGPT and MidJourney, MEPs want all content created by an AI to be mentioned as such. A provision which aims to fight against the proliferation of illegal content, against the manipulation of information but also to protect copyright.
Another important point, the recommendation algorithms used by social networks and likely to be instrumentalized during an election period to influence the vote, could be classified in the category of high-risk AIs, and therefore placed under close surveillance. Companies (such as Google or OpenAI) that provide reference AI models will have to commit to respecting fundamental rights but will also have to register in the EU (European Union) database. In the event of non-compliance with these commitments, the financial penalties could go up to 30 million euros or 6% of the worldwide profits of the company concerned. A sanction regime reminiscent of that put in place by the GDPR.
Moreover, since AI is an evolving matter, if not living, the European Commission intends “Organize a kind of review every six months so that the regulations are not obsolete”, adds Nathalie Devillier, doctor of law, quoted by France Inter. National agencies, such as the CNIL in France, will work on the adjustments to be made. Another point not to be overlooked, European citizens will be able to file a complaint against AI systems if they believe that their rights are not respected.
When could this new European legislative framework come into effect? In practice, for the moment, the AI Act has been adopted in committee by MEPs; the text still has to be voted on in plenary session in June and then negotiations will continue at European Council level. A final adoption by the end of 2023 can therefore no longer be ruled out. However, as with the GDPR, a two-year grace period should be granted to the economic players concerned so that everyone can adapt to this new legal situation with complete peace of mind.