Good morning, August Fala. Please welcome your hosts for today, Ana Care and Een Nihulawan.
Good morning and welcome to this international AI summit as part of Ireland's hosting of the EU presidency. We're delighted to be with you today as we explore the opportunities to innovate and collaborate through AI while building trust and safety. To begin our discussions, we're delighted and honored to welcome on Teshock Mart to officially open the conference. Please welcome on Teshock.
Welcome to the international AI summit and welcome to Dublin. And in particular, I want to pay tribute to our ministers, Peter Burke and Ne Smith in particular, from our enterprise department, for the work they have put into the organization of this international AI summit. And when we say, Céad míle fáilte, one hundred thousand welcomes to everybody who has traveled here, and I trust that you will find your days here productive and useful. And today marks the launch of European AI Innovation Month, and I'm delighted to be joined by Executive Vice President Henna Virkkunen, Commissioner Zaryva and Commissioner McGrab, and by colleagues from member states across the European Union, because with them we share a clear focus, ambition and determination, summed up by the theme of today's summit: to harness AI to revolutionize Europe's competitiveness.
For Ireland, this summit is also a flagship event in our six-month presidency of the Council of the European Union. Our decision to prioritize this event reflects the importance we attach to this moment in the development and deployment of AI. And the impressive assembly we've gathered here today — thought leaders, industry leaders, sectoral experts, and exhibitors — demonstrates the seriousness with which we are approaching what is happening and our determination to ensure that Europe and Ireland continue to be a leading force in the AI-driven era. In her State of the Union address last month, President Ursula von der Leyen spoke about AI as a tipping point of our times. I strongly agree with that view. And here in Ireland, we are determined to ensure that we are positioned to capture the dividend while mitigating the risks. In Europe, we have charted our own path through what is a fast-evolving situation, balancing innovation and opportunity with a trustworthy, human-centric approach to AI. We were the first to introduce a groundbreaking legal framework for AI with the European Union AI Act, recognizing the transformative potential of AI but also the risks. We have long been supporters of research, excellence, innovation, capacity building, and adoption of AI, including through our AI Continent Action Plan, Apply AI Strategy, and more recently the new action plan on cybersecurity and AI. And through it all, we've been guided by our European values. We want responsible, safe, and secure AI. AI benefits guided and guarded by human agency. This, as President von der Leyen has said, is the essence of the European way. Increasingly, we are seeing how prescient and how important such an approach is. I believe this strong foundation positions Europe well to really accelerate our AI transformation, to cut through the noise, the hype and the rhetoric, and to really zero in on the opportunities that answer many of the challenges we face in Europe today.
We know that to secure an economic and indeed a cohesive societal future which meets the needs and expectations of our people, Europe must urgently enhance our competitiveness and productivity. And that is why competitiveness is one of Ireland's three core European Union presidency priorities, along with values and security. And it is to the front of our minds in all we do, and all three priorities are interconnected. We are focused on the One Europe, One Market roadmap, including driving AI transformation. We believe that AI has the potential to transform enterprise performance in key sectors, to improve public services, to support education and learning, and to reduce digital divides across the world. But to realize these benefits, we need to deliver on the practical value of AI applications. And this is an issue I've discussed at length with my fellow leaders across the European Union, and it is a strong point of commonality. Our citizens and wider European society must see tangible benefits from AI, not as a secondary consideration, but as the primary objective. And this point has very much informed the themes of today's summit, with a focus on sectoral value creation and the European AI opportunity.
As we pursue this course, I believe that every member state must have agency and must act. In that spirit, today in Ireland, we are launching a new program, Intelligence Applied. It builds on our recently launched national digital and AI strategy, reflecting our determination to stay coordinated, agile and responsive in this fast-evolving space. We will leverage our key strengths — our people, economy and ecosystem — reflecting our strong track record as a tech hub and our decades of effort ensuring we've been at the forefront of every significant tech wave. We will build on this legacy to position Ireland as a leading applied AI economy across both private and public sectors. We will maintain our strong focus on skills and education through our new Department of Further and Higher Education, Research, Innovation and Science, relatively new, with new skills and literacy initiatives to build an AI-ready nation, including a new AI Future Skills program and a new national AI literacy task force. We will be supported by our innovation ecosystem, which positions Ireland well as a leading destination for applied AI research, innovation and commercialization. And this includes investment in our new national AI research center, Ring AI. And we'll continue to focus on and nurture one of the central elements of our approach to industry and innovation over generations, and that is genuine partnership with industry. We will work with industry and we will continue to do so. As a government, we have made long-term commitments to energy and infrastructure investment and to an enterprise-focused policy that will ensure competitiveness, adaptability, and productivity are enhanced. As I've said, we will continue to work closely with our industry partners, and we will support smaller businesses and startups to adopt AI through targeting key business sectors. We will establish a new national startup hub and AI startup accelerator program. Our public sector too is witnessing and seeing the value of AI. We are actively exploring how we use AI technologies to assist health screening and imaging services, improve customer service experience in our tax system, help to predict fire and flood forecasting, and prevent disease in agriculture, to name just a few areas.
We're also acutely conscious that as we navigate the changes and challenges AI will bring for the world of work, continued stakeholder engagement will be essential. And that is why I am establishing a new forum under the government's Labor Employer Economic Forum to bring together government, trade unions and employers to explore the impacts of AI and wider digital change on the workplace. While I believe our strengths in terms of our people, economy and ecosystem give us a competitive advantage, what underpins all of our work is trust. In an environment where businesses need certainty, citizens need confidence and investors need stability, Ireland does not see innovation, regulation or safety as competing goals, but rather as mutually reinforcing. We know it is critical that we get the balance right between innovation and necessary, well-resourced regulation. It is only by doing so that we can avail of the transformative potential of AI without compromising on our core values. The need for such a balance becomes clearer every day. To conclude, I'm genuinely delighted to welcome you all to Dublin and to officially mark the launch of European AI Innovation Month. We have an exciting day planned for you with some truly exceptional speakers. And I also encourage you to visit the exhibition space with over 50 inspiring examples of the remarkable potential of AI. In simple terms, AI will be one of the defining opportunities of our lifetimes, and Ireland will continue to work hand in hand with our European partners to ensure we seize it. In Ireland, we have a track record and experience in this space which equips us well for the challenge. We have been to the forefront of every tech wave over recent decades, and we will be to the forefront of this one in Europe too. There's a clear and growing sense of urgency in this pivotal moment. I am confident, I am determined, that Europe, as a great partnership and enabler of peace, progress and prosperity, will not be left behind. And I trust that you will all enjoy the day.
Taoiseach Micheál Martin. It's now my pleasure to introduce Commissioner Henna Virkkunen, Executive Vice President for Tech Sovereignty, Security and Democracy at the European Commission. She's going to talk now about the dual challenge of building safe and trustworthy technologies while strengthening competitiveness. Please welcome Commissioner Henna Virkkunen to the stage.
Good morning, Honorable Taoiseach, Prime Minister Martin, Minister Smith, Minister Burke. Thank you for inviting us and thank you for hosting this important AI summit, and thank you for the Irish Presidency for all the work you have been doing and you are doing to boost investments and innovations in the field of technologies and boost the competitiveness of the European Union. I would like to begin with a remarkable story from Ireland's past. In 1858, on Valencia Island off the coast of Kerry, a cable was brought ashore, opening one of the first telegraphic links between Europe and North America. Until then, a message crossing the Atlantic had to travel at the speed of a ship. Suddenly, information could cross an ocean in minutes rather than weeks. In many ways, it was the internet of its time. And more than 160 years later, Ireland continues to play a very crucial role in connecting Europe with the wider world. And this summit is itself a powerful illustration, bringing together today policymakers, technology companies and industry leaders to discuss one of the most consequential technologies of our time, artificial intelligence.
We have seen that AI has been developing very fast during the last months. And over the summer, we have seen AI agents escaping their environment, inserting malicious code, using social deception on humans. Such incidents undermine trust in AI and threaten our public safety. The good news is that in the EU we already have a legal framework to tackle that kind of challenges. The AI Act, which regulates high-risk AI systems and requires that providers in the EU assess and mitigate systemic risks from the most advanced AI, and this applies to all the advanced models placed on the European markets, no matter whether the provider is based in Munich, Shanghai or San Francisco. And mind you, there were no modern AI agents when the AI Act was adopted. Yet, our regulation is well equipped to deal with these developments. This includes risks from losing control of models, biological weapons, mass manipulation that could affect elections, and cyber attacks. The AI Act prevents overlapping regulation across member states, keeping the European space predictable and simple for our startups and industry.
And since the 2nd of August, enforcement powers have applied, and the Commission has already taken our first enforcement actions under the AI Act. In particular, the EU Code of Practice on general-purpose AI, which most AI providers have already signed, sets out measures for providers to have in place to comply with the AI Act obligations, and these measures include, for example, the use of external evaluators. Our AI Office has started enforcing these rules and sent requests for information addressing issues such as security, external evaluation and monitoring how models are used. And in addition, last Friday I called for a meeting with the scientific panel which is created under the AI Act to seek its advice and help on how to investigate recent incidents. The Commission has been loud and clear: AI must remain under human direction, oversight and control. And we stand ready to take all necessary steps to ensure that companies follow the AI Act rules so that European people and companies can trust AI models and systems, because we think that trust is the core basis for innovations.
Yet we know that no country or continent can do this alone. Europe will keep working with like-minded partners on model evaluation, early warning and AI security, and we will accelerate our work towards international rules for responsible development. The EU Code of Practice already exists and it's a great basis for this discussion. Every day we see how the most capable models develop further. Last week, Mistral launched a public preview of Mistral Large 4, and with this model, Mistral has moved close to the frontier, and this is important progress for Europe, but more needs to be done to stay in the race. As President von der Leyen underlined in her State of the Union speech, Europe needs its own AI capabilities, particularly to protect our independence and national security, because national security, defense, and sovereignty are increasingly shaped based on the control of these crucial technologies.
And this starts with owning the computing infrastructure needed to develop frontier AI models. And this is why we have launched 19 AI factories and 13 national antennas. And we are mobilizing more than 30 billion euros now to set up seven AI gigafactories. The Cloud and AI Development Act, once adopted, will complement this, tripling Europe's computing capacity over the next five to seven years and streamlining the conditions for the deployment of sustainable data centers. In addition, we are working with France, Germany, and other member states to build capabilities to develop advanced models and frontier AI approaches. And we are making sure more equity for our excellent startups is available. The 5 billion euro Scale Up Europe fund, combined with other existing tools such as the Strategic Technologies for Europe platform, is a great start in this regard. But we need to go a step further. This is precisely what was announced by the Commission President in the State of the Union: new ways to use public and private funds to invest in the most promising European companies and own capabilities. We are working closely with the European Investment Bank Group to set up a new public equity capacity with the member states as soon as possible, so that it's ready at the latest by the start of the next MFF.
Dear audience, while we manage the risks and when we boost our sovereign capacities, we need to make AI real for our competitiveness and for our industries. I think it's important to be clear here. These two aspects are equally important and they support each other. We are not doing either this or the other. We need to pursue both priorities at the same time. Today, together with the Irish Presidency, we launched the European AI Innovation Month. Four weeks dedicated to showcasing and discussing AI innovation in Europe. Nearly 100 events in over 15 European countries are part of this initiative, and I will have the honor to conclude the month on the 17th of November at the Apply AI Summit in Brussels. Europe's goal is very clear here: to use AI to boost our competitiveness, our productivity and our innovations, and to build a true European AI continent. One year ago, with the Apply AI strategy, Europe took an important step in this direction. The strategy starts from a very simple conviction: Europe will lead if it applies technology at scale in the sectors that matter the most to our economy and to our society. And this is why Apply AI is built on an AI-first principle. Companies and public administration should rethink their processes, products and services around what AI can deliver.
And one year on, we see first results in healthcare. AI can help flagging medical scans containing early signs of cancer, and this allows doctors to focus where it's most needed. Our network of 350 screening centers across the EU are helping validate and deploy such AI tools faster. And in energy, Europe struggles to keep energy affordable while expanding renewable generation. AI can help predict electricity demand and optimize power distribution. And this is why in June we officially launched the AI Grids initiative framework to develop, train and deploy a pan-European AI model for power grids. And in manufacturing, AI models can analyze images of products in their making, identify deficits in production lines and cutting waste and downtime. With Apply AI, we are supporting the development of such reliable and game-changing models for industrial applications.
And in robotics, where Europe can become a global leader, world models will enable machines to understand and act in the physical world. And we want to ensure that the technology and the production capacity for this huge market are built in Europe. And therefore, we started investing in large projects applying advanced AI models to robotics. And there is more that our Apply AI strategy is enabling. From European multilingual AI-driven news platforms to open-source models for floods and wildfires, helping cities and emergency services with climate emergencies. And we also have support on the ground with our network of experienced centers for AI that helps businesses and the public sector to transition into AI-ready powerhouses. And perhaps you will see some of these examples in practice at the AI Office exhibition here on their stand today. So in total, across different sectors, we have already mobilized nearly 2 billion euros aimed at concrete industrial challenges and genuine needs of our society.
But for Europe to become a global leader in industrial AI, we must continue our Apply AI journey across three axes. First, we must sharpen our sectoral focus across key industrial sectors from health, transport, and agri-food to advanced manufacturing, defense, and space, where we combine industrial strengths and public interest. Second, we must strengthen collaboration across the AI ecosystem: trusted data sharing, access to infrastructure, shared testing environments, and closer links between large industrial players, innovative startups, researchers, and public authorities. And third, we must build stronger implementation vehicles that pull resources, remove common obstacles, accelerate deployment, and create European scale. In this context, I'm also very much looking forward to the contribution of the special envoy for industrial AI to be published in November. Building on this report and the work of the Apply AI strategy, we will soon launch dedicated industrial AI alliances in strategic areas to steer deployment and address persistent bottlenecks.
Dear audience, ladies and gentlemen, allow me to conclude with a thought inspired by one of Ireland's greatest poets, Seamus Heaney. In his poem, Digging, he looks at the pen in his hand and decides it will be his tool, as the spade was his father's. Every generation finds its own instrument. AI may be ours. His words were, 'I will dig with it.' And so the time for Europe to dig has arrived. Thank you very much. And I wish everybody an inspiring summit.
Now, it's not often you get to meet somebody who is a Nobel laureate, a fellow of the Royal Society and the Royal Academy of Engineering, and whose research has made the cover of Nature nine times. But these are only some of the accolades of our next speaker, who is chief scientist at Alphabet and co-founder and chair of Google DeepMind. Ladies and gentlemen, please welcome Sir Demis Hassabis.
So Demis, this is so intimidating for me as somebody who has a relatively small scientific background meeting somebody who has such an extensive scientific career behind them. And I want to start there with asking you where you began in your academic career, because you did go and study computer science. You then went to do a PhD in cognitive neuroscience that came from your background, you know, creating games. Tell me what piqued your interest first of all in computer programming and then maybe trying to mimic the human mind while you were doing that.
Well, I guess I've been very lucky in my life that I've managed to sort of find many interests that I got very passionate about early on. And I guess the number one thing, even though I've done all these different things as you say, that sort of through thread is really a fascination, as far as I can remember, about how the brain works and what intelligence is. And then that led to, you know, when I sort of discovered programming and computers, how could one implement some of these systems on a computer and have the computer do these things automatically. So really that whole thing, you know, everything I tried to do was with that goal in mind and using AI as maybe the ultimate tool for science.
And you really have utilized it as an ultimate tool. You won the 2024 Nobel Prize in Chemistry with John Jumper for your work predicting protein complex structure, developing AlphaFold, which now scientists across the world use. I guess this demonstrates the potential of power, the potential of AI to be used in science, but there are some ingredients that make a good problem for AI to tackle.
Yeah. So I mean, if you zoom out, AlphaFold was, I guess, our first big project applying AI to a big problem in science, the 50-year, you know, grand challenge of what is the structure, the 3D structure of a protein. But it really developed from some of the earlier projects that we did like AlphaGo, where we built a system that could beat the world champion at the game of Go. And that was a very important system because it was really the beginning of the modern AI era, 10 years ago now, because it learned for itself how to play the game and the strategies, so rather than being programmed with the rules and the heuristics. So it's a kind of new type of system, and the great thing about it was that it was able to invent new strategies that we'd never seen before, and even the best human players hadn't seen before. And I was waiting to see a capability like that emerge from an AI system in order to then apply it to science. And I think it's a general way of seeing how to go towards solutions. What we look for is problems that can be described as massive combinatorial spaces, huge search spaces that you couldn't possibly search by brute force. And so what you need to do is you need to learn using these AI systems a model of how that domain works, whether it's Go, mathematics, or protein structures, and you learn that from as much data as you can get your hands on, and then that model can guide a very efficient search process to find the needle in the haystack, the answer. And that's what most of our programs have done, and it's actually a very general-purpose approach that can apply to many things.
Can I come back to that a little bit? So, when I found out that I would have the privilege of interviewing you, I didn't have enough time to go and learn how to play Go. But we are a games family and I've put it on our to-do list. We're going to get this now. Is that important that we have an idea of play, of creativity when we're thinking about coming up with innovation and when we're thinking about what AI should be able to do? Like this is something that I get concerned about sometimes because I worry that creativity will get dampened almost if we use AI too much. And I question to myself like what is the friction that we should preserve. But maybe what you just described there is part of it, that we have to be able to be creative in play and have that joy of failing and trying again and maneuvering our way around it. Is that something that's important to you when you're working with AI?
I think it's massively important. I mean, it was certainly important for me in my early career playing games like, you know, chess for the England national team and so on. These great games have always been a great training ground for training the mind, our own minds, in a kind of safe domain where you can try out a lot of things and learn from your mistakes. So that's also why we started with games domains to train our early AI systems like Go and Atari and these types of games. So, you know, I think the idea of play and learning has always gone together. And I think for these AI systems and for us to get the best out of them as scientists and artists and creators, I think the creative input from the human designer or human scientist is essential. And these systems are great tools for, you know, analyzing data or maybe surfacing some insight or finding some structure in that data. But what they can't do is ask the right question. And I think that's the hardest thing in science, is actually asking and articulating the right question. That's even harder than finding the answer. And I think coming up with the right hypothesis, the right conjecture, and the right question to ask requires, you know, creativity and judgment. And so far, you know, our AI systems are good, maybe good at solving things. But it's the human scientist and artist that brings the creative process and question.
So how did you respond when you saw that OpenAI had proposed a solution for the Navier-Stokes problem then? And I know there's lots of, you know, conversation going on whether that did occur or not. To me, I wondered about it because, like, well, was the purpose just to get the answer? Have we lost out on that collaboration, creativity, the avenues that other researchers will go down by doing that? Is that something there that you think the scientific community across the world needs to be keeping in mind when we're talking about harnessing AI?
Yes, I think so. Look, the maths, you know, the recent math results are very impressive, and we were working on similar types of problems and making progress too. But so there's several different things to unpack there. So on the one hand, it's great that we have systems that can potentially get answers to or help us get answers to these complex questions. But in mathematics specifically, I think a lot of it isn't just about the answer, it's also about the journey to get to the answer. And then often what you find is you have these very famous conjectures with math, like Fermat's Last Theorem or the Millennium Prize problem like Navier-Stokes, and the idea is not really just to solve the problem, but it's also to develop new types of mathematics on the way to solving the problem. Sometimes that new mathematics is more valuable than the solution. So I think there's a bit of an issue in that. I think we would have done it in a different way, of kind of working with the math community to kind of understand maybe how the answer was arrived at, not just sort of come up with the answer out of the blue. So I think there's some kind of complex issues there to think through, and I think the mathematics community is now kind of going to propose things. But on the other hand, it's different from other types of domains. So one reason maths and coding and games are amenable to these types of AI approaches that we have today is they're what's called verifiable domains. So what we mean by that is, you know, you can verify the answer whether it's correct or not. So in a game you can see who wins the game. So then you can go back and then judge how good the move was or how good the strategy was. And obviously with a math conjecture you can come up with a proposed answer or an equation, you can check whether that equation is valid. And with coding you can come up with some new code and then see if it runs right. So those domains specifically, you know, you can check the answers, they're verifiable. So the question is, that's why so much progress has been made in those particular areas. Now the question is, what about the natural sciences, physics, chemistry, biology, and engineering, increasingly complex domains, real world, that will have more real-world impact, say, than some of the math solutions? But there I think you'll still, although AI will be very helpful and helpful tools, and obviously we're using it a lot in biology to help understand disease and hopefully one day cure many diseases, you still need real-world loops. So you need to come up with a hypothesis. Maybe the AI system can help you do that, but then you need to check it experimentally. Maybe you need to gather new data and then check whether your hypothesis was correct and then advance and iterate to the next idea. And so I think for the moment it looks like there will always be a kind of real-world loop in order to verify and check the answers for the more natural sciences. So I think that will always kind of limit the speed at which that can go at. So I think these are kind of very interesting live debates right now about the differences between, I guess, the more theoretical sciences and the more practical physical sciences.
Having said that, I think as well, I've been talking to a lot of mathematics friends of mine, Fields Prize winners and others, about what should happen next in mathematics. And what I've been suggesting to them is their skills are more needed than ever, but that maybe they need to pivot slightly to slightly more applied areas of mathematics. For example, AI safety and security. We need many more mathematicians with the kinds of skills that they have, PhD-level skills, to kind of prove and verify things, ideally formally prove guarantees and bounds on what these new systems can do as they get increasingly more powerful. And I think we actually need, you know, 100 times more resources and work and talent going into that area so that the safety and security part of it keeps up with the capability.
I think it's really important what you just said because I am hearing from young people. I teach undergraduates, you know, well, should we bother going into computer science now? Should we bother going into mathematics if the AI is going to solve it for us? But I'm heartened to hear that you're like, it's a tool we need to expand. This will mean that humans need to go into those other fields. But you mentioned there cybersecurity, Demis, and that brings us into the space of thinking about other applications of AI. I would happily stay talking about AI in science, but we can't ignore the fact that, you know, this is a very commercial space. There's a lot of geopolitical tensions going on, and if we do want to make sure that AI is used for the benefit of humanity, there are things that we should perhaps be doing to make sure it gets built in those ways. You've written extensively about that and you've discussed it. Can you share with us here what you feel needs to be done now in order to make sure that we are developing AI in a way that is safe, that is responsible, and that is actually to benefit humanity?
Well, we've been thinking about that since the very beginning. So when we started DeepMind back in 2010, which is sort of the dark ages of AI and nobody was working on AI at that point, we foresaw, you know, we thought it was a 20-year mission, and kind of amazingly we're sort of on track for that, but we thought it would have these incredible benefits like, you know, potentially solving diseases or helping create new energy systems, you know, actually dealing with a lot of the other challenges facing society today, pressing challenges. I think AI is going to be an amazing tool for all of those things, but also it's a dual-purpose technology, general-purpose technology, so it can also be repurposed by bad actors for harmful ends. So there's two things that I worry about. Sometimes alignment problems. So that means, you know, can we make sure that the system follows the goals that we give and stick to within the guardrails that we set for those systems? And that gets increasingly difficult the more automation there is in systems. So now obviously we've been entering in the last year the agentic era where we're, you know, actually having these autonomous agentic systems that go out and do things for us, and of course we want that because they're way more useful if they can act in the world, but also that increases the risks involved with what those systems might do. We've seen that recently, earlier this year, with some of these hacking incidents. I think these are early warning shots for us that, you know, the systems aren't that powerful yet, so the incidents aren't that major, but they could be looking into next year and the next couple of years if we don't do more work as an industry and individually as companies and governments on securing these systems and building, you know, secure sandboxes to test them in that are probably offline, you know, and not on the internet yet when you first test them. So there's a lot of work that we can bring best practices from other sectors that have dealt with, you know, safety-critical systems, whether it's biology or nuclear. I think there's a lot of lessons that can be learned about containment systems, and I think we need to start taking that seriously and bring experts in from those areas, but also just from cybersecurity in general. And that's something we're kind of world-class at at Google, is cybersecurity. We've had a lot of experience of that, and we're bringing to bear and upgrading that onto the agentic era that we see today. And I think one of the big things we need to do is asymmetrically power up the defenders, trusted defenders, whether that's banks or governments, to defend from cybersecurity vulnerabilities and use these systems in a defensive way, maybe ahead of the time before their general release, to kind of fix the vulnerabilities in our software systems before they're out, you know, these systems are out in general release and then bad attackers can use it. So we've got to kind of asymmetrically power up our defensive side to keep ahead of the attackers.
And is that something that you think should be done across countries in terms of governance and policies? Do you think the EU is perhaps taking the correct approach in making sure that there are these guardrails on the technology as it's being developed?
I think you've got to get the balance right of seizing the opportunity and the incredible opportunities that are out there, both economic and scientific. I think it's, you know, that's why I spent my whole career working on AI and getting to this point, helping us get to this point, because I think it will be the most unbelievable and beneficial tool ever made for us to advance science and medicine and these types of areas. But yes, it's an international, you know, it's a cross-border technology. It's going to affect everyone in the world. So to me it seems obvious we kind of need eventually a sort of international standards body of some sort, and I wrote an essay over the summer that was quite, you know, well shared and read about how we might proceed towards something like that. But we've got to do it in a very kind of nuanced and nimble and dynamic way, partly because we don't want to stifle the innovation, all the amazing things that come out of it. We want those things. But also there's no complete consensus even in the people working on this, the scientists and the researchers working on this, on what exactly the risks are and how big they are. We know that they're out there and we need to quantify them more. But there's a lot of uncertainty around it. So what you need to do is basically build up dynamically these kind of certification processes that are proportionate with the risk as we see it and can quickly change depending on where the technology goes, because, you know, one of the problems with things like regulation is that it's very slow-moving by its very nature, and governments have to catch up with how fast the technology is moving. So, you know, if you legislate for something three years ago, it will be out of date today and actually probably won't even address what the real problems are. And so we need a framework, and that's what I proposed, a sort of standards body that can quickly adapt and test the things that are most critical with the systems as we see them today. And then what we've got to also do, I think, is put in 10 times or more resources, and we're doing a lot of this at Google and at DeepMind, research and computational resources into the safety and security side of these AI systems.
And then would Google DeepMind be classified as a frontier lab then in what you had proposed, which sounds like a great idea? If you were a frontier lab, you're at the cutting edge of this, that comes with a lot of prestige but also some responsibility too.
That's right. And so that's what we were proposing, is anyone can qualify to
Be a frontier lab and have a frontier model based on the benchmarks that we're all trying to hill climb and optimize towards. But then indeed, if you are at the frontier, you know, we just released Gemini 4 yesterday, which is, you know, a frontier model, then it should come with extra responsibilities because you're breaching into the unknown, and you know, which is great, but it needs to be approached scientifically and thoughtfully and carefully when you're right at the frontier. And those companies, you know, half a dozen or so, and if you include Chinese companies, you know, maybe eight or nine that are at the frontier or near the frontier, I think that comes with extra responsibility. But it's not just the frontier AI companies, it's also frontier AI labs, it's also enterprises in general, all enterprises, probably many of you in this room, if you know, you should for your own businesses look to see, you know, which suppliers are acting responsibly. That's what you want in your businesses. And when you start thinking about having agents handling customer data or clients, you know, you're going to want guarantees and monitoring systems and other things for your internal infrastructure and IT framework. And I think that's going to be a differentiator, that kind of trust and safety and security as well as performance of the systems going forward in the next couple of years.
A last question, Demis, and I think that's such a great point about the supply chain. We already trust in that. Europe is leading in terms of climate modeling and weather prediction because we have our national meteorological services collaborating in an open way with their models, which use physics and machine learning at the same time. So that's something where Europe is already excelling. But that spins me to a point where we have so many scientists looking at the climate crisis, the collapse of nature across the world. And do you think it's reconcilable that if AI is to grow, as we will be talking a lot about today, can we reconcile that with the energy crisis, with the climate crisis?
Yeah, I actually think I'm very optimistic about that. I actually think I'd go the other way on that, is I think AI is going to be part of the answer to it. So in the near term there are these questions about obviously the energy it uses. Having said that, you know, I think we at Google, for example, and many of the hyperscalers are committed to green energy. So actually there's a lot of investment and money available for green technologies, be that small nuclear reactors all the way to, you know, solar and better batteries and so on. So actually a lot of my friends who are, you know, in like running green technology companies or investing in them, a lot of the problem is there are viable technologies out there, but they're mostly still kind of at the lab phase, and what they need is investment to be able to scale them at scale and then become, you know, cost-effective at scale. And now there's a demand for that. So I think that's one thing, but the most important thing is the AI systems we're building, if we build the right types of AI systems, should be able to help with things like climate modeling, weather modeling. We have the best weather modeling system in the world, WeatherNext, you know, it's modeling hurricanes and other disasters like fires and floods much better than traditional systems are. And I think going further forwards, I think AI systems will help with new sources of energy, fusion. We work on plasma containment for fusion reactors, to new materials like superconductors, better batteries, better solar materials. So I actually think with the help of AI, that's what's going to end up giving us the technology to tackle the climate question.
Great. So a note of hope to end. It's been such a joy to speak with you today. I'm so sorry that we don't have the whole day to have these discussions, but ladies and gentlemen, please give a round of applause for Sir Demis.
Next, Europe is home to some of the world's most successful industrial, telecommunications, and digital commerce companies. As AI reshapes business models and competitive dynamics, our first panel of the morning will explore how European firms are transforming technological capability into global leadership. For this discussion on the European AI advantage, leadership, innovation, and global competitiveness, please welcome our moderator, Murray McInnes, former European Commissioner and former first vice president of the European Parliament. Jenny Lingfist, senior vice president and head of business area cloud software and services at Ericsson. Dr. Peter Corte, member of the managing board, chief technology officer and chief strategy officer at Siemens, and Robert Gates, co-founder and co-CEO of Zalando SE. Please welcome them to the stage.
Okay, welcome everyone. You've heard the high-level speeches and fantastic interview there with Sir Demis. But AI isn't new to Europe or to industries. So I don't even know where to start, but maybe with Siemens. So Peter, we were chatting backstage, it isn't new and you're using the technology. Talk to me about that.
Yeah, thank you. And indeed, I mean, AI actually, Alan Turing, when he really created that, how it feels, you already had that idea about AI at some point coming through. That's why we have the Turing test. We've been building at Siemens since 1975 on, of course, expert systems that would do that. We, for the longest time, we had a technology called CNNs where we were building neural networks for specific industrial applications. And of course, now with transformers, we have this enormous amount of data coming our way because the devices become more intelligent, we need connectivity, we need to put this all together, and then we need to make sense out of this. And so the traditional technologies didn't work. That's why we have the transformers, and that's why we're excited. And so now, finally, it's becoming accessible to everyone. But indeed, it's not new. But of course, the amount and the speed is really new.
Okay. And what's interesting is that you're very different companies. I mean, when I look at Zalando, when I look at Siemens and Ericsson, and you mentioned connectivity, you've been making recent statements about how AI and connectivity are really a great marriage. But what do we need to do more of?
Great question. And also AI being not new to Ericsson, having worked with AI for 20 years. So lots of AI in the mobile networks today where we have actually 50% of the world's global traffic flowing through Ericsson networks, and we have lots of capabilities in the networks in terms of how we can steer the traffic to secure we can all stream and call, but also securing mission-critical communication at the same time. But I think the important part for us is, and thanks Peter for that bridge, is not only what we do in AI in the networks, but the role that the networks will play for enabling the value of AI. So AI not only being about data center investment, but the real value, of course, will be when we get AI out into enterprises, into businesses, having physical AI in the world, and then we need the connectivity layer to secure mobility, and we need trusted networks for that. So I think one part to answer your question is just to make sure we have that notion that we need trusted advanced connectivity in order to really realize the value of AI. And where are we on that scale of we need it, so we're not there yet. Where are we? It's very uneven globally. So we have a few front-running markets: US, we have India, we have Japan. Europe is accelerating. And of course, we have a lot of really strong assets in Europe, both in terms of global technology leaders, also in terms of companies developing and using AI and data, but Europe needs to accelerate in order to be out there with the front-runners. And I think there's a lot we can build on, but there's also some prerequisites that need to come in place for the market for that to happen.
For your company, which is customer-facing, what has AI meant for the development of the company and for the customer experience?
Yeah, I mean, for us, AI has been, or machine learning has been really a core since the foundation of the company. So we use AI or machine learning actually for like inventory allocation, like in our warehouse systems, we use for recommend systems, or as well for predicting like delivery times for customers. I think like Zalando in its current form would not exist without kind of machine learning or AI. But obviously, I think like recently we've seen as well a lot of like faster developments with AI. So we see it now in the efficiencies of the way of how we work. We see it in software engineering, like pull requests, like the software changes, they happen much more rapidly now. We see it in all our command systems actually, they get better and better on a daily basis. And as well like predicting like delivery times actually gets much more accurate. But we see it as well how it actually really affects our customer experience in a good way. So for example, like size-related returns. So when you actually help customers to detect the right kind of merchandise that really fits them. Just, I mean, last year actually AI has helped us to get the size-related returns down by 8%. And we now actually use it, which I'm very excited about, is to use AI for like something we call like lifestyle AI. So like an AI that really is knowledgeable about like the entire world of fashion, that knows everything about products, about the brands, and then really connects with the consumer, with individual consumer on an individual level, and based on taste and based on needs and based on size profile, helps to find the right answers for consumer in a very personal and natural language way. And I think that's quite exciting. So it really affects as well the customer experience in a good way.
I want to talk to you about data because it is the key, I think, to unlocking the potential for you. You have your own data source. I mean, how valuable is that, and how do you use it to serve your markets?
Yeah, I mean, I think data for data is the key ingredient, is like what actually makes AI then kind of work and powerful. And obviously, like we harness a lot of data. We have more than 60 million people that order with us on a yearly basis. So only like in Europe, and interact with the merchandise, give feedback on products, and this creates an enormous amount of like very valuable data that we didn't actually use for. On the one hand, actually we use it for the brands that actually work with us to actually get better in terms of predicting, in terms of manufacturing, but we use it as a kind of empowers really all our systems, the recommender systems and everything. So it's really at the core, like at the core of our kind of flywheel that we actually have. And I think for that, it is incredibly important for companies such as us that create this data are as well able to mine the data, provide research, and as well drive innovation, because that's fundamentally, I think, what the value creation and the innovation and progress actually will mean.
And data similarly for your company where you're working with industry. I mean, you have some incredible statistics around the efficiencies gained by using AI and harnessing the value of data.
Yeah. So for us, what we want to do and what we're doing is we talk a lot about, of course, AI infrastructure which we need, and we talk about the frontier labs and the frontier models which are important, but then there's a layer which is actually not enough talked about, and that is the application layer. The application layer is really the layer where the value is being generated for the customer. And at Zalando or all of our other customers, at Siemens we serve 30 different industries, and they all have very different data and they're very different logics behind that. And what we realized is if you take the frontier models and compare them, they're all fairly similar in the end, but because the speed is really fast, but everybody's catching up more or less, the big differentiation becomes when you put context, when you give it actually prior cases, when you give it really the understanding of the context of what the AI is supposed to solve. And so we realized this in the end, we need to get as much sufficient data to our customers to make that work. So I give you an example. We are having a lot of controls out there. Controls control any kind of machine. So every third machine that is leaving a factory today is controlled by a Siemens controller that you have to program. And that used to be very, very cumbersome and takes long. Highly skilled automation engineers takes a couple of weeks. We built an agent that is very contextual and that is actually doing this in a couple of minutes and is really checking and validating and doing the testing and commissioning and all of this, and it can be done within minutes by almost a layman user. And so these are the examples we see. However, this necessitates that we have access to that data. And industry, where I think Europe has a unique chance. It's not about replicating again technology that is out there. Frontier models, there's many frontier models. We don't have a scarcity of frontier models. What we have is a scarcity — we don't have enough applications that are truly adding value to our customers. So we have to have the application layer, and for that we need to have specific data that is in industry. In Europe we have a lot of industry, and so we will have to come together and work together as Europe to make industrial data available to our customers, which is important because you cannot go on the internet and download that data for production and design, but you actually have to share it across all European constituents.
So that's an interesting point. We actually have a huge resource in Europe called data, but we need to collaborate better. Jenny, you're nodding vigorously. Why do you think that is the key to get innovation to work better for Europe?
Yeah, great. And maybe broadening beyond data also, as we said, and I think the heading says as well, the European AI advantage. We have a lot of strong assets. We have a lot of R&D. We have heavy tech companies with R&D in Europe, both European companies such as us on stage. But we also, it's a very global business, and we need to identify where in Europe do we have a chance to play, which are the critical technologies for the future, and which of these do we have a chance to play in, and which of these will we need to build partnerships and trusted tech alliances or trusted alliances with other global companies. And this will be critical, in my view, for building this European tech advantage. We can't do everything, then we will spread too thin, but really identifying where do we have the strength, which are the critical technologies, and how do we build partnerships. And I think Ireland is a fantastic example. And today as well, it's not only European strong companies, but it's really global, and we have a lot of global companies investing in Ireland as well. We have one of our most advanced AI software development centers also here in Ireland, and that's really strategic for us because of this global ecosystem that exists here.
One of the things you said recently was about a need for a shift in mindset around collaboration. What do you mean by that?
I believe it's tied back to again sort of the strategic, maybe taking even broader perspective then. You asked me about the global sort of where are we in the global scale, and then there is an ongoing tech war between US and China, and of course Europe plays a role in here, and that has ramped up the geopolitical situation. It's quite different now to how it was before. So I think in that context, it's a different perspective for us on how we build partnerships and how do we build trusted partnerships. There's been a lot of discussions about trust today, also yesterday, and of course one part is how we build trust in the application layer of getting the adoption up, of trusting AI, but another key part is having the trusted full stack of what's beneath the AI model and the AI application, making sure we have trusted advanced connectivity, but also the other layers in there. And for that we need to work differently with partners globally.
It has absolutely started. We actually had a gathering this morning with the tech alliance, so it started earlier for sure, but it's ramping up and it will play an even bigger role, I believe, going forward to secure the European competitiveness.
Robert, I want to talk about this thing about trust because in a way there's two layers to this. You've used this technology almost forever in your companies to build up, so you trust it. It's the customer, the person, it's the media is talking about this of trust in one sense your company, you know, strategies both of these things. So how important is this for you, and how do you reassure the customers that deal with you and your suppliers in all of this that you have trust at its core?
Yeah, I mean, trust is probably, I think, the essence of our business model. Because obviously customers need to trust us that the merchandise is what it is, that it actually arrives as well on the day that we predict it to be. So trust is very much at the core of our business model, and we as well know that trust is very hard to gain and very easy to lose. So we don't really take it lightly. And I think my view or our view on AI is very much it needs to help to create more trust and not needs to erode trust. So that is about how we actually think about AI when we drive like responsible personalization, or where we as well drive, we use AI to enhance product descriptions, for example. There's always the human in the loop that in the end actually as well checks like is it actually accurate, so that the product description is more accurate. And I think that's ultimately very, very important for the experience that we create. But this is why we are a bit cautionary against proposals such as the Digital Fairness Act, because it really as well limits the way of how we design experiences and as well limits, I think, the potential of innovation for companies such as ours, and that's what limits as well the customer experience.
I might as well pick up on your hook there about legislation or regulation and move to Peter on this because I was a legislator and now I can see the impact from your point of view in terms of trying to innovate. Where's the place for what I would describe as effective regulation, and are we there yet?
I think it is a constant evolution. The big question is if you put this on a global scale, and part of my job, I have the benefit of, I was just coming from Dubai where they're building 5 GW of data centers as part of the Stargate project. They have an AI minister too, and they have a very different kind of conversation than right now that we have. I think yes, we can probably say we were the first, if we're the first in innovation of regulation, but I'd rather have innovation first and then the regulation that comes with it. So I don't want to have innovation in regulation. For example, to have the innovation per se that happens with AI and then really start to regret. If you think about the speed, the speed of AI and the developments, all the frontier models that constantly improve so quickly, it's very, very hard to keep up from a regulator's perspective in the time and the sequence of how this works. So in a way we need to have a new innovation of how we do good governance in the age of AI because it's too slow. It's not happening in the way that we need it on a global scale.
Could AI help legislators?
