Europe’s AI Factories - Rafal Duczmal

AI Factory concept is only partially about buying and deploying big machines. That's the obvious part. What we try to attach to it is the ecosystem services, where entrepreneurs can come and have their problems solved in a complex manner. Welcome to Supercomputing in Europe. The podcast where we discuss various interesting tech topics with supercomputing at their core. My name is Apostolos Vasliades from E&CCS, the EUCC National Components Centre in Sweden, and I will be the host for this episode. Today's guest is Raffał Duszmal, who is the chair of the governing board of the EUHPC joint undertaking. We will talk about the new AI Factors Initiative by the European Commission, Europe's position during this AI proliferation, and the future of AI development. Raffał, thanks for joining us. Thanks for inviting me. I'm glad to talk about the EUHPC and what your EUHPC is doing. Before we start with our main topic, give us a short introduction about yourself, your background and your current interests, maybe. I studied physics, so I generally love technology. I had this feeling that being a scientist or an engineer is not my treat. I have different, I could do something good, doing something different, but I still love it. I found the work when I could be close to cutting-edge technology, to researchers and do something good, not being one. I can understand them, but I work with governmental people, and my role is to be the bridge. One of the areas on my way was HPC, and when I got involved in that, I started loving it. Since the beginning of your HPC, I was representing Poland in the governing board. Since year and a half, I'm working as a chair of the board. I try now to do this job as good as possible. What exactly is your tasks as chair of the board? It's described generally by the regulation, but it's two brief sentences, like the chair is organizing the work of the board, and the chair is representing the board outside. But when you go into details, it's usually a lot of work. I try to talk as often as possible with the commission. I work with the J.U. and its employees, and with different countries trying to find what is the best way forward for us. When we end up meeting as a governing board, I'll try to prepare it as good as possible so we can agree on things, and that sometimes needs a lot of work in the background. In the end, the chairing of the meeting itself is just a small piece of my work as a chair. What can you tell us about the AI Factorist Initiative? AI Factorist Initiative is our answer, the European answer to the needs of today. We see huge developments in AI codes and AI applications to push forward these applications and what we can use them, what people can benefit from them. We need a lot of computing power. There are some things that you can do with your computer behind your desk or on a small cluster, but in the end, the most important solutions will come from the supercomputers. AI Factorist is our answer to that. We discuss how to deploy in European infrastructure that would be serving the best of the community. We are already doing that with the computers we already have in our portfolio. Many of them have big GPU clusters, which is the most used part of the computers for the AI, but we wanted something more, something also more than infrastructure. AI Factorist concept is only partially about buying and deploying big machines. That's the obvious part. What we try to attach to it is the ecosystem services. We would like these supercomputers to be kind of a central point of a bigger thing, where entrepreneurs can come and have their problems solved in a complex manner. They don't have to, in the end, understand that well how to run their code or solve such a big machine. We want people to be there to help them. We want people to be there to help them with data processing. We want people to help them find data that will actually help develop their models. Many things, ending with, if you like, we have a co-working space here for you. Just come. It's sometimes good to sit at one desk with some specialist, and we would like to provide this opportunity to people. EuroHPCJU has nine operational supercomputers around Europe at the moment. That among traditional HPC also serve AI users and applications. What will the AI Factorist place be in this landscape? AI Factorist should be the most emblematic, the first point in our portfolio that cut an attention of people who wanted to use AI. Even though our machines that we have right now, most of them can be used very effectively for AI codes. When we were building them, it wasn't that purpose. AI factories are built with that purpose in mind. They will be the best prepared for that, and they will have this ecosystem of services around them. When today, when you apply for computing power in our machines, we also help you. We have the national competence centers around Europe in 33 countries. We provide support for users that need the help to solve the problems, but this will be focused on AI and also focused on entrepreneurs, because we know that the scientific community in general is much better prepared to use such a big machine. This is something that the scientific community has been doing for years, also machine learning. It's not a new concept. We're doing that in the previous century, but now we have the capability. But if you think about SMEs, they don't have this capacity, but they could have great ideas and we don't want them to have this high bar to jump over to start. On the way, they will develop their own capabilities, but we are there to provide them this first few steps of system. So it's good to say that Europe is merging the computational power for AI work. Yeah, of course you can say that. This is more or less what we are doing since the beginning of your HPC. Two years back, when we decided to start this endeavor, the joint undertaking, we were thinking about the legal framework and how to operate this stuff. The purpose of starting this work was that right now, if you want to make a step forward, we need to combine our resources. That even the biggest countries in Europe, it would be kind of a struggle for them to provide all these computing power facilities just on their own. And secondly, when you are, for example, representative of Poland like I was for many years, you want to have your users do great things with machine that you provide, but it won't be constantly. So when we have this, also this European access to different machines, because right now when we buy a machine, for example, in Poland or in Germany, it's a coin investment. So part of the computing time of this machine is the national entities are in charge. And the second part, usually it's 50/50 deal, is in charge of the GU, so for the European purposes. So we try to use the infrastructure we have in the best possible way and keep it all the time running also on the national side, but also the European side. So we decided that we need to pull our resources to become a bigger player and also to deploy more of this infrastructure. I remember a report in 2017 saying that European researchers are using around 25 to 30% of the computing power in the world, but we in Europe, we have only 5% of this. So it meant that most of our work we had to do somewhere else, send our data, and even though we were doing that mostly with our like-minded countries and our allies, we see today that the world is getting more complex, let's phrase it like that. So we would like to have, and we wanted that this 8 years ago, that we have our own capacity to fulfill the needs of the research in Europe and for our entrepreneurs. That was always in our mind, this is the hard task, as I said before, scientific communities well prepared to use the machines, but we really, really want to create this space where our SMEs and our industry can also make this step further. They need to want that, but if they want, we are there. We want to be there open for their needs. Talking about the industry, how do AI factories aim to bridge the gap between academic AI research and real-world applications, real-world meaning, industrial? We have different instruments in Europe that the purpose or one of the purposes is to bring together industry and academia, because we always want that whatever we are doing, for example, in the framework programs, now in the Horizon Euro program, that it doesn't end up as a publication only or just a project website that is communicating that we and achieve something great, but there is no uptake of this. So bringing research, academia, and industry closer together is always our goal in the union. AI factories could also bring something here, some additional value. There will be situations where we have people from the industry coming, okay, we have this problem and it's not only a computing problem, there are some things that we don't know yet or we would like to know better, and because our computing centers, they are working in academia for years, okay, this is already a network well-established when we communicate all over Europe, and so it is also a good entry point to provide these people coming from industry or SMEs, okay, we know people who have knowledge that you are seeking. So we will bring you together, you will talk and you can come back to us together and we will provide you with the computing power you need, but they will help you set a problem. Sometimes they will already have these people on board, sometimes they wouldn't have to go outside because they may have already this knowledge there, and it's also kind of bringing academia with the industry. Most of our supercomputing centers in Europe, they are attached to the research organizations or universities in a way when industries coming to these AI factories, they are already coming to the academy, okay, but it's a good entry point and we really hope that when the AI factory will be rolling already, it will be kind of a magnet, people will be interested, okay, what can I gain from that, and we hope that the services we will be providing will be really interesting for them and they will find the use of that. But we must remember that it's a process, we will start in some point and we will develop it further, every encounter, every communication with industry will let us do our job better. With AI users, there is a consensus that Europe is lagging behind in AI, as you said, first with computing power and now within AI development, are AI factories going to change that? That's the goal, when we are thinking about lagging behind, I don't want right now to engage in all this discussion, I would like to say one thing, how we are doing things in Europe in comparison to different parts of the world, it's true that in most cases we cannot compete with them in terms of the muscle, okay, there are places in the world that they can throw money at the problem to bigger extent that we can and they will do that, we observe it, if you have enough money, you can throw it at the problem and sometimes you will solve it that way, but we don't want to compete in those terms, this is not a good competition, to those who are a little bit concerned that we may be lagging behind, I just invite you when the AI factories will be open and it will be quite soon, just come to us, let's talk, if there is something that you identify, okay, but you are missing this point, great, give us this information and we will try to use it in the best possible way. Are there specific areas of AI research and development that the AI factories will focus on? AI factories, as I said before, AI factories have their own specific areas that they want to specialize in and it's already quite a portfolio of different topics, from engineering aspects, medicine up to supporting public sector in developing new services for citizens or culture heritage, preserving cultural heritage, it's also a theme here. There is already plenty of information that was shared on the JUE website about the projects that we are finding here, so both the first seven and now the six that joined in March, so there is already hint what areas they would like to target, quite soon there will be more information about that. Let's talk users, who will be eligible to access the AI factories? Again, we will have this European part and we will have the national part, so for the national part it's quite easy, the rules will be defined nationally, so for the European part soon we will publish this access policy and then there will be, I would say the bridging protocol from now until the machines will be deployed because we want our AI factories to start with the resources we already have. Most of them are already somewhere in our network, some of them were already hosting entities or we are awarding the AI factory grant to hosting entities that they already have a URHBC machine and they don't want, they don't need to build a new one because the one they have or are building right now, like Jupiter in Germany or Alistair Koch in France, they were already by design prepared to serve also AI purposes. The start of the services, this ecosystem services as I call them, will be sooner than the many of the machines will be deployed, we don't want to lose the time. The access policy we are raising right now will be for this bridging period and there will be special access routes defined for the users from scientific community that try to use AI in research and more for the entrepreneurs prepared with tailored for their needs that they would like to use our machines, that's for the computing power. As for the services, we need to wait a little bit longer and this is still work in progress, we are in tight collaboration with our hosting entities establishing that, but there are a few principles, first of all we want the access to the AI factories to be fast, some of the procedures that were known so far like classical peer review process for the scientists, that is well known to them for the industry, it's usually we don't understand why it takes so long when I go to Google, it's like and it's done. Now we are working to be as fast as possible and it will be significantly faster than before you will see that in our access policy. So that's the first principle, second principle is we want to bring entrepreneurs, we want to bring SMEs inside, the services will be designed to be interesting for them and the third principle is like it's not like we want now to use our computers only for the private entities. The researchers are also welcome here, but you know the process, okay. And many of the services that we'll be offering as AI factories tailored for the entrepreneurs, it's also because scientific, this has different needs and sometimes they don't need this particular support, but we are already providing different kinds of support for them and I hope they like it. And if you see something that we are missing, tell us, we really love to have this feedback. Is there a approximate timeline when the AI factories will become operational? The first machines that we now need to procure, as you all know, the situation with procuring GPUs became a little bit more complicated lately and we still need to understand all the parts of the process. They may be some additional time that we will need to deploy this infrastructure, then we would need in different circumstances because we already bought some of them, we more or less know the drill. So that is still months ahead, deploying new infrastructure, it takes time. As for the services, we would love to have them sooner. I expect that in the beginning of April, we will be able to jointly with the commission as you probably saw in the communication from President von der Leyen, also from Executive Vice President Wirkunen, that they are really checking our progress. We should expect some announcements here about the first services opening already in April. What are your biggest hopes for the AI factories and its contribution to Europe's AI leadership? First of all, I hope that we will get the private side ready to our machines. It was always our goal and we are not satisfied yet. We have a lot of good stories already to say, but we are pretty sure that in Europe the backbone of our economy are our SMEs and there is a lot of potential still not exploited. So we would really love to have them working with our centers and on our computers. My first hope would be that the AI factories will be the moment that more SMEs, more entrepreneurs will think, okay, let's try with this Euro-HPC, what they can offer me. Second is that we don't stop here. AI factories is the next step in our development. President von der Leyen, some time ago and now that we are thinking about gigafactories, this is also something that we still need a little bit patience to be unrevealed. This is also a political action that is leading by the Commission, but as I said in the opening of our conference, we are ready to be the vehicle to implement that, but it's not our role to draw the regulations. For us it's the implementing side. Finally, do you see any challenges that the development of AI is facing in Europe and maybe globally and how do the AI factories aim to address them? I think one of the biggest challenges here is data. One thing is to have the computing power, but it's hard to imagine doing AI with our data. Nowadays there is a notion like data is everywhere, we produce and there are reports, we produce this and that number of amount of data every day, but when you'd like to use this data, there are many obstacles to do it in a manner that will really provide you an answer that you're looking for. First of all, you need to be able to process this amount of data to be prepared then to run your algorithms. This is actually a place where we hope that AI factories will be able to help people. We are in the conversation with some industrial partners, we're in conversation with public administrations or public administration to have access to different kinds of data sets that are well curated, prepared to be processed and I think this is something that will be expanding here. Also what is not exactly the AI factories, but what we do to help with that, because we have a lot of big data providers in Europe, data that could be useful to many commercial also purposes from our research infrastructures or technological infrastructures in Europe. So what we are doing right now, we have an open procurement for a service we call hyperconnectivity. When we award this contract, URAHPC will be provided with the ultrafast connection between our sites, but also with different data providers or data centers in Europe that we envisage that these are important points to be connected by this network. So we want to also use these services to make the data more available for people using clever machines. Rafael, thank you so much for being with us. Thank you for inviting me, it was a pleasure. This discussion took place during the URAHPC Summit in Krakow, Poland. The episode was brought to you by ENCCS, the EuroCC National Competence Centre in Sweden. ENCCS gives supercomputing training and support to businesses, academia and public administration. The Supercomputing in Europe podcast is made possible with funding from the European Commission and specifically from URAHPCJU, the joint initiative between the EU, European countries and private partners to develop a world-class supercomputing ecosystem in Europe.