The twenty-seven member countries of the European Union (EU) definitively approved, on Tuesday May 21, unprecedented legislation at the global level, to regulate artificial intelligence (AI) systems in order to promote innovation in Europe, while by limiting possible deviations.
EU legislators reached an agreement at the beginning of December, after difficult negotiations, with certain countries such as France fearing excessive regulation which would threaten the development of this promising sector. The text was adopted on March 13 by a very large majority by MEPs.
With “this historic legislation, the first of its kind in the world”Europe “highlights the importance of trust, transparency and accountability (…)while ensuring that this rapidly evolving technology can thrive and drive European innovation”declared in a press release Mathieu Michel, Belgian Secretary of State for Digital, whose country holds the presidency of the EU Council until the end of June.
This legislation will apply mainly from 2026. It adopts an approach “risk-based” and imposes constraints on the various artificial intelligence systems proportionate to the dangers they represent for society.
Bans will be rare
AI systems presenting only limited risk will be subject to very light transparency obligations, while high-risk systems, used for example in critical infrastructure, education, human resources or maintenance of order, will be subject to reinforced requirements before being authorized in the EU.
These requirements will include, for example, human control of the machine, the establishment of technical documentation, or the implementation of a risk management system. Bans will be rare. They will concern applications contrary to European values, such as citizen rating or mass surveillance systems used in China.
Specific rules will apply to generative AI, such as ChatGPT from Open AI, to ensure the quality of the data used in the development of algorithms and respect for copyright.
Artificially generated sounds, images and texts must be clearly identified as such to avoid manipulation of opinion.