© Pixabay/Dušan Cvetanović
The decision is unprecedented in the world: Europe has just given the green light to the AI Act, which aims to regulate the development of AI ;AI on the continent. The subject has been running since April 2021, but has been disrupted for months by the arrival of ChatGPT. At issue: certain countries like France or Germany who fear that this new framework will prevent the bloc from measuring itself against more advanced countries like the United States.< /p>
These countries are in fact redoubling their efforts to keep the subject out of sight. In France as in Germany, an increasingly dense network of specialized startups is developing. Enough to lead Emmanuel Macron to request a regular evaluation of the Law to prevent us from losing “leaders or pioneers because of that”. The final text is the result of final debates, ratifying a “political agreement” on this text whose mission will therefore be to limit possible deviations, but without “restricting’innovation”.
Europe gives the green light to the AI Act: here's what it contains
The approach adopted is two-speed. On the one hand, protection of learning sources – a subject that is still largely neglected across the Atlantic: it involves ensuring standards for the quality of source data (and therefore that the AI is not trained with data poorly verified), but also so that they do not violate copyright. A watermark system must make it possible to clearly identify that the texts, images and sounds are produced by generative AI.
But also more transparency: developers will have to be very clear about how their algorithms work and impose risk audits on the potential impact of their models on the fundamental rights of Europeans.
There is a huge nuance: these rules will only apply to the most strategic actors such as law enforcement or education… ; or leading companies. For young shoots, the frame is much lighter; a way to respond to fears about innovation. In itself, the final text summarizes existing legislation in various European states, while drawing inspiration from approaches already put in place to regulate tech in texts such as the Digital Markets Act.
Controls are primarily carried out by companies, but the Commission can intervene at any time if breaches are noted, with fines in sight which can go up to 7% of turnover with a limit of 35 million euros. There is, however, little real prohibition – essentially reserved for any development which could be contrary to the values of the Union, such as the development of new mass surveillance systems.
The moment is in any case judged “historic” for the different actors who participated in these negotiations. Thierry Breton at the initiative of this text presents the ultimate compromise as a real step forward. While emphasizing the pioneering role of the bloc on these issues. However, the legislative journey of the AI Act is still far from over. After this green light, all the technical conditions still need to be specified.
Once these conditions have been specified, the final text must be formally adopted by the Parliament and the Council before becoming a European law. For “the last mile of the Law” in national Parliaments (which receive proposals at the same time as the other parties and can contribute) the subsidiarity control mechanism applies: in certain cases transposition into national law may be necessary, while in others, the text applies via treaties.
- The course of the AI Act is in its last sprint after the adoption by the 27 of a compromise last Friday.
- The text thus especially creates obligations for the most strategic actors.
- A way, hope many parties, of curbing innovation.
📍 So you don't miss any news from Presse-citron, follow us on Google News and WhatsApp.
[ ]