"The first legal framework for artificial intelligence is currently being created in Europe, the AI Act. It's about civil rights and competitiveness. Many member states and parliamentarians want to regulate ChatGPT - but regulation must not be an end in itself.
The USA invents, China creates and the EU regulates.
That's pretty much the first comment I hear when I talk about the planned European AI law, the so-called AI Act. We have been working on the law in the European Parliament since 2021, starting with the Commission's legislative proposal, the development of the parliamentary position and now in the ongoing negotiations with the member states. In more than 200 hours of negotiations, we as the EU are trying to do pioneering work: to create the first legal framework for artificial intelligence worldwide.
The goal is a framework for how society can take advantage of the opportunities offered by artificial intelligence and at the same time ward off dangers. Since regulation of technology would inevitably always lag behind it, the focus is not on the technology, but on the way in which AI is applied. Biometric facial recognition is practical and harmless for unlocking a cell phone, but we are familiar with its use for mass surveillance in public spaces in autocracies like China. Something like that has no place in the EU.
With the classification into prohibited applications and high-risk applications, where certain requirements regarding the quality of the training data and transparency and documentation requirements must be met, the AI Act focuses on a few areas. The majority of AI applications will not be affected by the AI Act. In any case, it would be a misconception that one law could preventively regulate all challenges that may arise from AI. In terms of the expected upheavals, the progress through AI is comparable to the Industrial Revolution. In the next few years we will have to adapt many of our laws, such as copyright law, to technological developments.
The fact that the AI Act is intended to become a principle for dealing with AI is also reflected in the two areas of content: On the one hand, there are questions about civil rights and the state use of AI technologies. This is where the biggest difference lies between the EU Parliament and the member states. As parliamentarians, we loudly demand a ban on biometric surveillance, social scoring or the use of emotion recognition by the state. EU countries, on the other hand, prefer to exclude the area of law enforcement, as the majority of countries want to use AI technology for surveillance purposes. So far only the federal government of Germany is on Parliament's side. This debate is therefore particularly heated.
On the other hand, the innovation aspect is at the center of the negotiations. Europe's competitiveness in the digital age also depends on the AI Act. There is more at stake here than many might imagine. In recent years, the EU Commission has launched a wave of digital policy legislative proposals that our companies are already having to deal with. But we shouldn't over-regulate every technology out of diffuse fear and we shouldn't burden small and medium-sized companies with the same requirements as the requirements placed on Big Tech companies. The goal must be to strengthen innovation instead of slowing it down and to create optimal conditions for new ideas for our companies and scientists. For example, through real-world AI laboratories, of which there are currently far too few in Europe. Simple risk classification procedures such as clear checklists can also help to give innovative companies the necessary clarity and legal certainty.
A relatively new, major pivot point in the debate is general-purpose AI systems, also known as General Purpose AI (GPAI). These are systems that actually do not fit into the risk logic of the AI Act. Rather, a GPAI system can be used for numerous purposes, all of which are impossible for the manufacturer to foresee. We are currently still in the process of negotiating the right path. Especially with regard to the underlying models such as GPT3/4, on which Chat GPT is based. The Commission and member states have no position on this in their texts, as the debate about effective generative AI systems only arose after their adoption.
Parliament wants to impose certain quality criteria for such powerful models, which are trained on an enormous amount of data. For example, a risk management system and technical documentation. There are different assessments of this among the member states. The German federal government proposes not to regulate these models in the AI Act, instead to regulate them within the framework of self-regulation at international level. The G7 countries are currently working intensively on such a system.
Many member states and parliamentarians want to take credit for having regulated ChatGPT. However, regulation must never be an end in itself; the goal must be that general-purpose AI operates on democratic principles, responsibilities are clearly regulated and the EU does not go it alone.
The negotiations at the political level will enter the next round at the end of October and with numerous other outstanding issues it will be extremely exciting to see what the final result of the negotiations will look like. We still have a long way to go here." [1]
1. Europas Wettbewerbsfähigkeit steht und fällt mit dem AI Act. Frankfurter Allgemeine Zeitung (online)Frankfurter Allgemeine Zeitung GmbH. Oct 10, 2023. Von Svenja Hahn
Komentarų nėra:
Rašyti komentarą