MEPs are still trying to reach a common position for future interinstitutional negotiations on the Artificial Intelligence Act (‘AI Act’) (see EUROPE 13167/6). German MEP Axel Voss (EPP), rapporteur on the dossier for the European Parliament’s Committee on Legal Affairs, gave EUROPE an update on developments ahead of the vote in the European Parliament committee on 11 May. (Interview by Thomas Mangin)
Agence Europe - With lengthy legislative processes and rapidly evolving technology, do you think artificial intelligence can really be regulated?
Axel Voss - I don’t think so. Of course, there will be new developments, including generative artificial intelligence, such as ChatGPT, which is already on the table. I think that the AI Act is not the right tool to solve everything. I think we either need to go for a specific approach to generative AI - which would be necessary - or we need to integrate it into the AI Act. For the time being, we have mostly worked in terms of risk ranking, with specific obligations. These are steps, but in the end we cannot really control the outcome of an AI system. They try to say that the outcome should be accurate, but with systems like ChatGPT you need to know what the basis of the data is and answer some questions before you try to regulate.
So you would lean towards new legislation specifically for AI systems such as ChatGPT, rather than adding them to the AI Act?
For reasons of content, I would say that a specific approach is needed. For reasons of timing, I think we should include it in the AI Act. For generative AI, we will need specific rules in one way or another. This can be done under the AI Act, but more needs to be done. If it comes to the point where we say we need additional legislation, we have to be aware that it will take at least three years, if the Commission has to propose a new text... but we cannot know who the players will be at that time n or even if ChatGPT will still be around. In any case, we must hurry.
As MEPs approach a common position on the AI Act, what do you see as the weaknesses of the current text?
There are still a few points to be discussed, the others are settled. Further progress is needed on issues such as ChatGPT, generative AI systems, the issue of intellectual property, especially for the creative sector, and most importantly, Article 5 (which deals with prohibited practices in AI: editor’s note). My concern with this approach is that it is no longer really about product regulation, but more about consumer law. In my opinion, we are going too far, with a multitude of areas in which everything should be regulated. This makes things complicated and not necessarily easy to adapt in the end.
Personally, I would like to see a starting point that says ‘now the EU has understood: we need to survive in the development of the digital environment and that’s why you come up with a strategy with money, with priorities’. I think this is a gap in the Commission’s approach.
With very different positions and the fact that the EU Council adopted its position on 6 December 2022 (see EUROPE 13078/12) while new developments in AI are frequent, do you think future trilogues will be particularly complicated?
I would say that the EU Council, from a practical point of view, probably has a better approach than Parliament. The EU Council is getting to the bottom of what is acceptable - or not - for competitiveness. As you know, the mind is dominated by the search for compromise. This must be achieved by the end of the year. The Commission is ambitious and would like this to happen sooner. I have my doubts. We are also open to making rapid progress, but we have to bear in mind that the positions are very different. Everyone will have to give up some things.
You have been advocating for the creation of a European AI Authority (EUROPE) from the beginning. Do you think the future ‘Office of Artificial Intelligence’ will be sufficient?
On digital issues, there are lots of industries. I would say that we could start with this regulatory tool. But we should be aware and come to the conclusion in the next legislature that it is not very healthy to have such separate interests with the Data Act (see EUROPE 13150/7), the DSA (see EUROPE 13163/26), the DMA (see EUROPE 13163/15) or the AI Act. An EU agency that balances the interests of the different legislative tools should be achieved. We need to have a better balance to support competitiveness within the EU, industry, while protecting the privacy of citizens. We need a better approach, in my opinion.
Some MEPs, including yourself, called, on 17 April (see EUROPE 13164/2,) for a high-level summit on AI. What do you expect from an event like this?
Achieving a comprehensive approach would be the most logical path, but I think it would be very difficult to achieve. If it’s not possible because one country, for example China, handles data and what it can get out of it differently, I think that Western countries should at least come up with some kind of standard where everybody has a common direction to build a space for data policies.