Parliament's Special Committee on Artificial Intelligence in a Digital Age winds up 18 months of work and vote on its final recommendations, establishing a 2030 AI Roadmap.

Brussels

press conference

use of robotics and artificial intelligence (T)

BENIFEI, Brando (S&D, IT)

TUDORACHE, Dragoş (Renew, RO)

Press conference by Brando BENIFEI (S&D, IT) and Dragoș TUDORACHE (Renew, RO), co-rapporteurs on the Artificial Intelligence Act

SOUNDBITE (English), Brando BENIFEI (S&amp;D, IT), Rapporteur, Good afternoon, thank you for being here. Welcome to this press conference. The first, the joint one, by me and Dragos Tudorache as leader co-rapporteurs of the AI Act. We take the occasion of today's hearing on the proposal for regulation to share a few messages on the work we are doing. Of course, the drafting work of our report is in full swing and we have established a very good cooperation not only between us, but also between our respective teams, as well as the IMCO and LIBE Secretariat.

SOUNDBITE (English), Brando BENIFEI (S&amp;D, IT), Rapporteur, We believe that this is a good starting point for a great teamwork on this very important fight. Me and Dragos also, we are very happy to share that we found a significant amount of points of agreement on some key areas of the text, which will make our draft report already strong and ambitious. Although it is important to clarify that the report that we are presenting will only consist of what we can agree on while the rest will be left for the amendments stage. So that we can show also clearly to all our colleagues, to the process that we are experiencing, all our positions so that we can find the best compromises when it will be needed, where it will be needed in the most complicated areas, where probably we would also have to work for some time with the council.

SOUNDBITE (English), Brando BENIFEI (S&amp;D, IT), Rapporteur, Now I will give the floor to colleague Tudorache for the issue of the calendar, which is obviously very important. We hope that the council will be able to follow the calendar that we are presenting today, to have a similar time frame so that we are going to have in due time these very much expected regulation. Because if we take too much time, if it will take too much time for the negotiations to end with the Council, we will come too late with the regulation that is, in the ambition, to be the first one in the world to be this kind of horizontal, encompassing AI regulation in a way that tries to put us as to some extent as some standard setter. But this means that we need to do this on time and avoid any fragmentation between us and the member states, and that we can really have a a proper regulation in place in due time. I wanted to just stress this point. Now I give the floor to my my colleague to that. I wish to thank you very much.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, Good afternoon to everyone. To complement what Brando was saying in terms of methodology of work and also calendar, in terms of what he expects us in the months to come. So you know that we have quite the complex governance here in Parliament working on this text. There are the two committees which are jointly responsible, IMCO and LIBE, as is my colleague, Mr Benifei said. We also expect opinions from other committees which are involved ETRE, TRAN, ENVI, CULT, we also have JURI with with exclusive competences on some provisions in terms of calendar. We have 11th of April as the timeline for the presentation of our initial draught report.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, Initial draft so a report report, that will come out of the joint work that we are doing together. Then we have 18th of May as deadline for amendments, opinions to be received from the other committees by 11th of July, consideration of the compromise amendments, of course, the work that we will do through the summer and early autumn and then a vote that we expect in our respective committees in October, with a plenary vote in the first plenary of November.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, As also Brando was saying, what will follow afterwards? Of course, we are hoping that the Council will be also progressing in parallel with our work in parliament, so that they will be there at the rendez vous in November or December at the latest, so that as soon as practicable, after the plenary vote, we can actually start the trialogues.

SOUNDBITE (English), Brando BENIFEI (S&D, IT), Rapporteur, We are now, before we open to questions, together with my colleague, we want to give a few examples of topics on which we found some degree of agreement, so that we can present proposals on these together for our draft report. On the definition, for example, we can share that we are not going to restrict it, although we cannot give too much details. But we agree that we will not narrow it down.

SOUNDBITE (English), Brando BENIFEI (S&D, IT), Rapporteur, I want to say this because this has been an important topic to discuss. This can be an important message from ourselves, as the definition is one of the most debated points of the AI Act. So we will not give all the details now, but we can give you this clear message that we don't want to narrow it down, not restricted.

SOUNDBITE (English), Brando BENIFEI (S&D, IT), Rapporteur, Another element we want to share is that we believe that a rebalancing of the responsibilities across the value chain is necessary, in particular increasing those of the user. Let us not forget that, for example, in many use cases involving fundamental rights, public authorities are the user. So it is natural that they should not be limited to just follow instructions and a few other requirements, but rather add an extra layer of care, in particular as the user is the one best placed to know the exact context and target population it intends to use the AI system on. And therefore should also be best placed to appreciate the specific risks.

SOUNDBITE (English), Brando BENIFEI (S&D, IT), Rapporteur, Another example is a user that is in possession of the datasets used to train the IRIS AI system, as in many cases the provider might only supply the overall architecture of the system without necessarily controlling the data, which are in the hands of the user commissioning the AI system. You see from these examples that this chain of responsibility must be enacted well and we need to strengthen this user responsibility to avoid a lack of accountability.

SOUNDBITE (English), Brando BENIFEI (S&D, IT), Rapporteur, I mentioned also the use in public context for public authorities. That would bring to a lack of trust and lack of real operability of the regulation. Obviously, we will share more when tabling the report, but I pass the floor to my colleague for other examples on this. Thank you very much.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, We have on these two points, definitions and the role of the user and the division of responsibilities, these are two areas in which our teams have already found common ground in preparing a draft There are report. two more areas which I will mention myself.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, One is governance. We both agree that in the current setup, as the one in the proposal, it does not do justice to the objectives that we have and that we share, which is to avoid fragmentation of what should be a digital single market and what we do not want to see is 27 different jurisdictions applying differently and avoiding or blocking what we need to achieve, which is consistency in the application of this law. Very much so also for the benefit of those that are meant to work with artificial intelligence, to grow and to innovate with artificial intelligence, which is our businesses.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, Another point of concern for us when we look at governances, is to make sure that in the set up that we will eventually put forward, we also take good account of the need to have the right competence at the level of the national authorities when it comes to artificial intelligence, an area which we know requires a high degree of specialisation and we need to consider very carefully how our public authorities, how our national regulators, will actually be able to find the right talent to be able to apply and enforce the law.

SOUNDBITE (English), Dragoş TUDORACHE (Renew, RO), Rapporteur, And a fourth area of of common ground based on our initial work is the alignment between this text and GDPR, which GDPR is a key which is so fundamentally important for us, for how we regard privacy here in the European Union. And clearly, there are aspects of this text, of the AI Act, where the interaction with GDPR is very important. I will of course, it's enough to mention remote biometrics to name just one example where this alignment will be essential.