European AI Act seems to result in compromise

Banana peels, water bottles and mountains of papers everywhere. The council chamber of the European Parliament, also plagued by a broken coffee machine, was a big mess on Thursday. “Everyone is exhausted,” said MP Kim van Sparrentak (GroenLinks). “No more color in the cheeks.”

A crucial round of negotiations between the European Parliament, the Council and the Commission started on Wednesday on the AI ​​Act, the European law to regulate artificial intelligence. After 24 hours of non-stop negotiations, the meeting was adjourned. The talks will resume Friday morning at 9 a.m.

It illustrates the pressure to find an agreement. With the AI ​​Act, Europe would be a global leader in regulating AI, something that countless experts have been calling for for some time. A European bill has been on the table since the spring of 2021, but due to differing views and rapidly advancing technical developments, it has not yet been possible to finalize the law.

Milestone

In Brussels it is still hoped that it will succeed this week. It would mark a milestone in the regulation of AI and set a precedent for legislation and regulations elsewhere in the world. But it is still uncertain whether the negotiations between the European Commission, Parliament and Council can be brought to a successful conclusion.

Also read
‘What will the world look like in 2040 if humanity allows AI without regulation?’

MEPs want to go further than the Member States in various aspects. It has been particularly striking in recent weeks that Germany, France and Italy, under pressure from a strong lobby, suddenly developed the most powerful AI models (foundational models) wanted to remove from the law. Companies such as Google and OpenAI, producer of chatbot ChatGPT, among others, would only have to comply with a ‘code of conduct’ when developing these types of models. The European Parliament was strongly against this.

A provisional compromise now appears to have been found in the negotiations on this point, several news agencies reported on Thursday. It would mean that powerful AI models would be subject to the law, although only the largest systems would have to comply with the strictest rules. However, less strict rules would apply to so-called ‘open source’ models, which means that anyone can view and use the technology to train their own model.

Lobby

This would mean that the European AI companies that lobbied the hardest against the regulations in recent weeks, Germany’s Aleph Alpha and France’s Mistral AI, would be left out as open source companies. Unless they are classified as ‘high risk’. There is still a lot of discussion about that definition, confirms a member of the European Parliament involved in the negotiations.

Member states want to make the rules on facial recognition much broader

And there are more obstacles. There is still great disagreement about whether and in which cases security and intelligence services may use real-time facial recognition, for example to detect criminals. If it were up to the European Parliament, such facial recognition would be banned in all cases. But member states want to make the rules a lot looser to continue using this technology.

A representative of digital civil rights organization Access Now mentioned the proposal on facial recognition that is now on the table on Thursday on X “an absolute disgrace.”



Reading list





ttn-32