On Wednesday, June 14, MEPs gave a first agreement to the European project for the regulation of artificial intelligence (AI), a key step in the race launched by the legislator in the face of the frantic pace of innovations in the sector.
The European Union wants to be the first in the world to adopt a comprehensive legal framework to limit the excesses of AI, while securing innovation. Brussels proposed two years ago an ambitious draft regulation, the examination of which has dragged on and which has been further delayed in recent months by controversies relating to the dangers of generative AIs, capable of creating texts or pictures.
The European Parliament adopted its position at midday, during a vote in plenary session in Strasbourg. The approval obtained (499 for, 28 against, 93 abstentions) should make it possible to start negotiations with the Member States on Wednesday evening to finalize the legislation, if possible, by the end of the year. Commissioner Thierry Breton, who brought the text together with his colleague Margrethe Vestager, called for the process to be concluded “over the next few months”. The regulation will not come into force before 2026, in the best of cases.
Of great technical complexity, artificial intelligence systems fascinate as much as they worry. While they can save lives by enabling a leap forward in medical diagnostics, they are also exploited by authoritarian regimes to exercise mass surveillance of citizens.
Provide human control over the machine
Parliament’s position broadly confirms the Commission’s approach. The text is inspired by existing regulations on product safety and will impose checks based first on companies. The heart of the project consists of a list of rules imposed only on applications deemed “high risk”. These would be systems used in sensitive areas, such as critical infrastructure, education, human resources, policing, migration management, etc.
Among the obligations: provide for human control over the machine, the establishment of technical documentation, or the implementation of a risk management system. Compliance with them will be monitored by supervisory authorities in each member country.
The European Parliament intends to better take into account generative AIs of the ChatGPT type by calling for a specific regime of obligations, which essentially include those provided for high-risk systems.
The Commission’s proposal, unveiled in April 2021, already provides a framework for AI systems that interact with humans. It wants to require the user to be informed that he is in contact with a machine and to force applications generating images to specify that they have been artificially created. An obligation that will probably be extended to texts.
Bans will be rare. They will concern applications that run counter to European values, such as the citizen rating systems or mass surveillance used in China. MEPs also want to remove the exemptions provided for by the Commission to authorize remote facial recognition of people in public places by law enforcement. This subject should feed the debates with the Member States which invoke the fight against crime and terrorism to refuse the prohibition of this controversial technology.