The first law regulating artificial intelligence is getting closer. The EU states want to approve the draft today. But the big question remains: Can Europe catch up with the US tech giants?
Many called the events on December 9th last year “historic”: two years of tug-of-war over the world’s first law on artificial intelligence (AI), at the end a few marathon sessions – and then it was done.
EU Commission chief Ursula von der Leyen said: “From the outset, we considered it necessary to set clear guidelines at the European level to guide the development and use of artificial intelligence. And that is now the idea behind the European law on artificial intelligence Intelligence.”
No more German blockade
Great vision went in – and, quite soberly, an 892-page law book came out. It took more than six weeks until the small print was available in the capitals. And suddenly there was resistance again, especially in Germany.
Would Berlin once again block a law that was actually completed and in which it was continually involved in the Council? The scandal now seems to have been averted since Digital Minister Volker Wissing from the FDP also gave up his resistance a few days ago. Gritting my teeth – that’s how I can understand his statement on X, formerly Twitter:
“The compromise found on the AI Act now lays the foundation for the development of trustworthy AI. We will use the maximum scope and avoid over-regulation so that our location can be competitive and the value creation takes place with us.”
AI basic models should be given obligations
This contains his criticism, above all, of the planned handling of the so-called AI basic models. These are technology foundations on which a wide variety of applications can be built. GPT 4, for example, is the basic model – which powers probably the most famous chatbot in the world – ChatGPT.
In Europe, these basic models should have certain duties depending on the risk class, for example when passing on information, analyzing risks and documenting the data with which the AI is trained.
To ensure that this approach does not dampen the innovative power in Europe, said Wissing’s party colleague Svenja Hahn, who helped negotiate the AI Act in the EU Parliament for years, it is now crucial: “…that open questions are clarified, for example how does it work Classification of basic models into the risk system in practice – according to which criteria should this be done?” The Commission must now also ensure clarity when it comes to AI in criminal prosecution, so that no gaps in the law arise due to different national interpretations.
Keeping up with the USA and China
Franziska Teubert, managing director of the German Start-up Association, will breathe a sigh of relief if the EU succeeds in finally passing the AI Act. Because more than any line, speed is now important: “The regulation is not perfect, but you can work with it. Regulation also creates reliability for companies and investors. A lack of EU-wide regulation would have only provided us with patchwork quilts. 27 individual solutions in 27 Member States.”
Now their appeal should finally be implemented: “In order to keep up with the USA and China, we need European AI champions, and in the long term we need them to think less in terms of individual countries and act more in a European way.”
Train and test AI models
A line that Commission President von der Leyen also sets: AI startups and medium-sized companies should now have access to EU supercomputers, i.e. the necessary capacities to train, develop and test their AI models:
“This is similar to what Microsoft is doing for Chat-GPT by letting them run on its own supercomputers. We will also put common European data spaces at the service of start-ups and make huge amounts of data available in all EU languages so that the AI also works for non-English speakers.”
All of these are the next concrete steps towards the further development of a European-style AI. And the ambition to grasp things that you can’t quite imagine yet.
Kathrin Schmid, ARD Brussels, tagesschau, February 2nd, 2024 8:04 a.m