Lucilla SIOLI

What Lucilla SIOLI has said (5)

• “Okay. Thank you. Thank you very much for these questions. So first, a reminder on the interplay with sectoral legislation. There is a lot of work ongoing with the also in collaboration with the AI board, which is made of member states, but in particular with other groups, for example, with the Medical Device Regulation group, we already have an initial draft of clarifying the role of the medical device regulation in the India Act, for example, and much more work is foreseen now in the coming months, in particular in relation to the machinery regulation, but also with GDPR, um, and with the um Data Protection Board. Um, on um, I take note that some find the code of practice complex. Ah defined it too narrow. Um, but we are, um, frankly, um, in our follow up of the process, trying to make sure indeed that is in line with the with the AI act and that it does not go beyond the act. Um, I also want to remind, in relation to disinformation and other kind of systemic systemic risk that, uh, um, the idea, uh, also for the chairs is not that this kind of risk are not addressed at all, uh, but simply that, uh, even if some risks are not fully mentioned, that there is a responsibility on providers to identify and make clear efforts and explain these efforts to identify other systemic risk, and that this is not going to be voluntary. Um, and, uh, for what concerns the diversity in the advisory bodies, uh, where the scientific panel, as illustrated in implementing Act, is already insisting on geographical balance and gender balance.”
  Medical devices
• “Thank you very much, for inviting us here. And good afternoon. So first of all, I would like to say that we also value very much the cooperation with the Joint Working Group on the AI act, and we think it's a very important channel for a very important and deep exchange on the on implementation of the act in general. Um, and I will start by simply recalling the state of play, the fact that that we are now finalising the code of practice, and this is the code of practice, should be, um, a voluntary instrument for uh, providers of general purpose models to, to demonstrate compliance with the, with the AI act. And this, uh, as Brando underlined, this, uh, code of practice is actually drafted by independent experts, which include some important names. For example, Joshua Bengio. Um, and this code of practice is following an iterative process. There are different drafts followed by input by stakeholders. And now we are at the fourth draft. So the final draft, which should be published in the course of the month of May. Um, and the third draft was published in March. So as a as an AI officer, we facilitated the process. Uh, of course, we speak to the chairs and the vice chairs on this aspect of fundamental rights. I think that Brando has been highlighting, um, we have also, uh, you know, flagged the, um, the idea that, um, the risk should not only be the four that we saw in the third draft, and the rest are optional. Um, and, uh, but I also want to remind here that, um, already in that section, in our opinion, the fundamental rights were not completely excluded from the draft.”
  Artificial Intelligence · Transparency and oversight of AI-generated content
• “In the meanwhile, we are continuing to work on the template for the summary of the training data. I heard the comments that that were made. Now the template was shared for consultation in the code of practice. We have received a lot of input and we are now reworking it. We are also looking into the threshold that we have set together. Um, the computer threshold to see whether because of the technological change, there is a need to to review it. And then finally we are also working on the governance. You may remember we have to set up all the advisory bodies by the 2nd of August, and we will soon publish a call for a scientific panel. We already had an implementing act which has already been finalized. Now we publish the call as well as the call for the advisory forum. So, um, in the meanwhile, we push on the work on the standards. Um, we are pushing hard on this because, uh, the the work is complex, but, um, it's an important priority for us to be able also to have the standardization being done on time, because we realize this is important for the companies, for the compliance with the AI act. So, um, Just to conclude, in parallel, you may have seen that we have published an AI continent action plan. And uh, and there so we always want to support also and remind the importance role of innovation in this market and importance of being extremely clear in the provisions that we put ahead given the innovative nature of the market. Thank you very much.”
  Transparency and oversight of AI-generated content · EU competences on consumer protection and product standards
• “Uh, there were measures that would address, uh, certain certain systemic risks. Uh, for example, cyber offense risks or chemical, biological and so on, which are. Which are risks that can impact on health, that can impact on life, that can impact on autonomy and so on. And so, um, but we agree that there is a need to extend more the risk taxonomy, um, also to, to other kinds of risks. And, and we hope that this is what is being now modified. Um, we will discuss it further when we have, uh, a fourth draft. And, uh, I also want to highlight that to clarify better the code of practice, we have published a set of guidelines to clarify certain concepts which are in the act. For example, what is a general purpose AI model? Um, when does the responsibility of a downstream provider start and so on. And these guidelines are published for public consultation until the 22nd of May. So also there we are looking forward to receiving, uh, your input, um, other implementation priorities beyond the code of practice relate, uh, there are many other steps to take. As you know, the act enters into application in a progressive way. In the meanwhile, in the AI office, we are continuing to build the technical expertise. For example, today we are having a workshop with a network of evaluators experts. We think it's important to be able to also facilitate the development of a bigger expert community in this particular area, because the area is relatively new.”
  Artificial Intelligence · Transparency and oversight of AI-generated content
• “Um, and for what concerns the advisory forum there we will look for having industry, but also academics and also civil society organisations. And we will also make sure that, uh, you know, there are representatives from all the member states, um, the diversification of sources. Maybe it's something, um, you can explain a little bit more of because normally for training these models, you need huge amounts of data. So these companies, they scrape the internet. They they use a lot of data. It's um, it seems to be diversified by definition. Exactly. Because the volumes of data needed are very large. Um, where this problem may arise in the future is more in relation to the fact that there will be increasing reliance on synthetic data. And then I agree with you that there may be a reduction in the diversification. Um, and last, on the last question, I would like to remind the rule of the grandfathering rule, the rule of application of the, uh, to, to to the different models, in the sense that the obligations that on the on the general purpose models that apply in August apply to the models that will be put on the market after that date. For the models that are already present on the market, they will have time until 2027. Uh, to, to to to comply with the rules. Therefore, I don't see um for the moment and I'm not aware of, uh, um, model providers waiting to, uh, put their models on the European market because of uncertainty about the rules. Thank you.”
  Transparency and oversight of AI-generated content · Transparency requirements for interest groups