Europe takes a new step in the regulation of artificial intelligence
Europe is continuing the long and laborious process which should lead it to become the…
Europe is continuing the long and laborious process which should lead it to become the first jurisdiction in the world to legislate on artificial intelligence (IA).
On Thursday, MEPs adopted, by a very large majority, in committee, the future regulation proposed by the European Commission in 2021, which has been the subject of more than two thousand amendments.
As unprecedented as it is ambitious, this text, which first seeks to regulate AI systems with a clear risk, by clifying them according to their level of risk, from low to unacceptable, has caused a stir.
It is defined in a stormy context surrounding AI and its abuses: discrimination, misinformation, fraud… And a few weeks after more than a thousand personalities demanded a six-month break in AI research the sharpest.
Diagnosis of diseases, autonomous driving of vehicles, sorting of CVs, obtaining credit, evaluation of evidence in court… If the text has sparked so many quarrels for two years, it is because AI is interfering in all sectors. and touches on the sensitive question of the fundamental rights of the citizen.
“The path has not been easy, but we have managed to have a balanced text,” explains Dragos Tudorache, co-rapporteur (Renew), for whom “the people were at the center of all the discussions”.
Helped by recent current events, MEPs have above all added provisions on the very controversial “generative” AIs, such as GPT Chat which they want to submit to specific rules before they are placed on the market.
They are calling for protections against illegal content or the publication of copyrighted data used to develop the algorithms, which would allow creators to demand payment.
MEPs also extended the clification of certain AI systems in the “high risk” category, subject to stricter transparency and governance obligations.
They included those representing a significant risk to health, safety, fundamental rights and the environment. But also the systems used to influence voters during political campaigns and the recommendation systems of the social media platforms of more than 45 million users.
The practices to be banned, including citizen rating systems or m surveillance, have fueled the debate.
MEPs added to the list the use of emotional recognition systems, but only in the workplace, in educational establishments and for the management of migration flows.
They also banned the use of remote biometric identification systems, such as facial recognition in public spaces, except, a posteriori, in the case of “serious crimes” and only after judicial authorization.
They have also banned predictive police systems or the m harvesting of photos on the internet to create facial recognition databases.
MEPs also offer consumers the right to complain to an authority about an AI system or seek collective redress in the event of harm.
“We are showing the type of society we want, comments Brando Benifei, the other co-rapporteur (S&D). A society where social scoring, predictive policing, biometric categorization, emotional recognition, are considered unacceptable practices”.
This vote must be confirmed during a plenary session of Parliament in June. Then will open the famous trilogues – negotiations between the Parliament, the Council (therefore the Member States) and the European Commission – which promise to be fed.
“The high-risk clification of online platforms’ recommendation systems and the unnecessary inclusion of copyright requirements in AI law will increase compliance costs for successful businesses and have a detrimental effect on innovation”, regretted the European ociation of the computer and communications industry (CCIA Europe).
Welcoming a text that “better protects” citizens, the European Consumer Organization (Beuc) nevertheless says it is “very concerned that AI operators may circumvent the rules of the law” when it is up to them to declare that their services are not high risk.
“Furthermore, the basic principles proposed for ‘trustworthy AI’, which would apply above all to all AI, are only voluntary for companies or public entities, whereas they should be mandatory”, points out Ursula Pachl, Deputy Managing Director.
Legislators hope to definitively adopt the law before the European elections in 2024. The stakes are high: after missing the first train of artificial intelligence, the EU hopes to influence other regions of the world with its future law.