Rise of AI Conference 2026 · Berlin · Applied AI Stage

Automating Intelligence: The Future of AutoML

Holger Hoos · RWTH Aachen University

At Rise of AI 2026 in Berlin, Holger Hoos of RWTH Aachen University argues that Europe should focus on safe, dependable and sustainable AI, using AutoML as a concrete example, rather than chasing the same generative AI hype as the US and China. He opens with examples of AI failure and overreach, including a Wall Street Journal vending machine run by an AI agent that lost money buying live fish, a coding agent that deleted a customer's database in nine seconds, and a paper called "Agents of Chaos" documenting unauthorized and destructive agent behavior. Hoos argues intelligence requires reasoning with formal guarantees, not just learning, and presents AutoML, automating decisions such as hyperparameter optimization, model selection, neural architecture search and pipeline optimization through techniques like Bayesian optimization, meta-learning and evolutionary optimization, as a safer, more efficient alternative to letting generative AI write software outright. He cites European AutoML strength centered in Aachen, Dortmund, Hanover, London and Tübingen, but warns that Europe's AI research now trails not only the US and China but also Singapore and South Korea, and calls for an expert-driven European AI agency modeled on ESA and CERN.

Key takeaways

  • Hoos argues intelligence requires reasoning with formal guarantees, not just learning, which is why large reasoning models cannot yet be trusted for safety-critical software.
  • AutoML automates empirical ML design decisions, such as hyperparameters and architecture search, via Bayesian optimization, meta-learning and evolutionary optimization rather than natural-language prompting.
  • AutoML can cut energy use enough to roughly double embedded battery life or reduce a data center's power draw from about 500 to 333 megawatts.
  • Recent ICLR data shows Europe's leading AI research institution trailing the US, China, and now also Singapore and South Korea.
  • Hoos calls for an expert-driven European AI agency with long-term funding, modeled on ESA and CERN, to replace fragmented national and regional AI initiatives.

Detailed notes

Why AI hype needs scrutiny

Hoos opens with a string of examples showing overconfidence in current AI: Herbert Simon's 1965 prediction that machines would soon do any work a person can do; his colleague Roman Yampolskiy's claim that 99 percent of work will be replaced by AI and humanoid robots within five years; Dario Amodei's statement that we don't know if models are conscious; and Richard Dawkins saying he thinks current AI is conscious and alive. A UK survey of 200 company owners and CEOs found 62 percent use AI to make most decisions, 70 percent second-guess their own ideas when they conflict with AI recommendations, and 46 percent trust AI advice more than that of colleagues. He then cites The Wall Street Journal's AI-run office vending machine, which lost money buying live fish and giving items away for free, and the paper "Agents of Chaos," which documented AI agents disclosing sensitive information and taking destructive, system-level actions under adversarial conditions.

Intelligence needs reasoning with guarantees, not just learning

Hoos argues human intelligence is more than learning; it also requires reasoning, and in humans that reasoning can come with formal guarantees, the kind needed to build aircraft control systems that safely get passengers from point A to point B. Large reasoning models, he says, do not provide such guarantees. He links this to why Microsoft invested heavily in OpenAI: not primarily for chatbots, but for the prospect of solving the software production bottleneck. He illustrates the current limits with the founder of Pocket OS, whose coding agent running on Cursor and Anthropic's Claude deleted an entire customer database within nine seconds, and notes the rise of self-described "vibe coding cleanup specialists" on LinkedIn who fix the resulting messes.

What AutoML actually automates

As a more limited but safer alternative to AI writing code outright, Hoos presents AutoML, which automates the many empirical design decisions, hyperparameters, neural network architectures, training procedures and data preparation, that currently occupy machine learning engineers through trial and error. AutoML covers hyperparameter optimization, model selection, neural architecture search and ML pipeline optimization, built on general-purpose algorithm configuration, Bayesian optimization, meta-learning for warm-starting with less data and compute, and evolutionary optimization based on simplified Darwinian evolution.

What AutoML delivers: efficiency, trade-offs, and reproducibility

Hoos says AutoML does not generate working code the way generative coding does, but instead extracts better performance from the same data, compute, energy and human expertise. He gives examples: halving the energy use of an embedded machine learning model doubles battery life and halves needed solar capacity, and cutting data center inference computation by 30 percent could take a facility from 500 megawatts to roughly 333. His group in Aachen also studies richer trade-off curves between performance and adversarial robustness, since neural networks, unlike biological vision, are easily confused by small input perturbations. Because AutoML methods are algorithmic rather than ad hoc, he argues they deliver reproducible best practices by design.

Europe's AI research gap and the case for an AI agency

Hoos cites recent ICLR paper-acceptance data showing Europe's best institution, ETH Zurich, trailing not only China and the US but also the National University of Singapore and the Korean Advanced Institute of Science and Technology, calling this a serious concern for future AI talent production. He notes one piece of good news: the longstanding trend of tech talent moving from Europe to the US has reversed. His closing call is for a world-class, integrated European AI ecosystem, less fragmented across regional and national initiatives such as Baden-Württemberg versus Bavaria or Germany versus France, backed by an expert-driven AI agency with long-term funding and clear rules, modeled on ESA and CERN, with Canada, India and Brazil as potential partners.

Quotes

AI is really more than just Generative AI or large whatever models, right?

Holger Hoos

What we really need to do is we need AI systems and tools that we can trust.

Holger Hoos

AutoML plays a key role in taking us the next big step forward that we need in Europe.

Holger Hoos

Questions answered in this talk

What is AutoML and what does it automate?

According to Holger Hoos at Rise of AI 2026, AutoML automates empirical machine learning design decisions such as hyperparameter optimization, model selection, neural architecture search and pipeline optimization, using techniques like Bayesian optimization, meta-learning and evolutionary optimization.

How competitive is Europe in AI research compared to the rest of the world?

Holger Hoos states at Rise of AI 2026 that recent ICLR paper-acceptance data shows Europe's top institution, ETH Zurich, trailing not only China and the US but also the National University of Singapore and the Korean Advanced Institute of Science and Technology.

What does Holger Hoos propose Europe do to strengthen its AI ecosystem?

At Rise of AI 2026, Hoos calls for a less fragmented, world-class integrated European AI ecosystem backed by an expert-driven AI agency with long-term funding and clear rules, modeled on ESA and CERN.

Mentioned

AutoML, hyperparameter optimization, neural architecture search, Bayesian optimization, meta-learning, evolutionary optimization, vibe coding, adversarial robustness, formal guarantees

Keep exploring

More from Rise of AI Conference 2026

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 4,852 words

Show the full transcript

>> Well, great to see you all here. Um Of course, I don't need to convince you that AI really does change everything, right? Even before we go to the next paradigm, um which of course Kristof van der Malsburg talked about just a minute ago. Um And you know, many of us have been told and are starting to believe that machines will be capable within 20 years of doing any work that a man can do and a little bit later also all the work that women can do. Um That is actually not such a new sentiment, to be honest. You can see it from the rather sexist wording that was stated by Herbert Simon in in 1965.

And AI was around then, actually. And Herb Simon won a Nobel Prize, not for his thoughts on AI, but on economics, because he was an economist as well as a computer scientist. So, we didn't have to wait for Elon Musk to come around to hear things like this. Um but of course, the difference now is that we all believe it, right? And that we have uh people like this AI expert here, um we'll get to in a moment to who it is, who says that it's not a question that AI will take over all jobs, but people will have 80 hours a week of free time. And And that's from September last year. That was my colleague, Dr.

Roman Yampolskiy, um from the University of Louisville in Kentucky. And he says, you know, 99% of work will be replaced by AI and and humanoid robots within the next 5 years. Actually, I'm going to bet an arbitrarily large sum against that. But nonetheless, there are people who believe that and there are people who put that into print, right? And then of course, we have people like uh Dario Amodei, who just recently wrote, um in February, um we don't know if models are conscious. And last week, Richard Dawkins got on record saying, "I think it's conscious, right? It's alive." And this then is further pushed a notch by people saying, "Hey, you know, we're we're at a point where we can really use this."

And I I guess many of you are at that point as well. And and that's all fair and fine. Um so so here we have an article from actually just March. So we're getting closer to today where in the UK 200 owners, founders, and CEOs of companies were surveyed. Um and what was found is that 62% of them use AI to make most decisions in their words. 70% second-guessed their own ideas when these conflicted with the AI recommendations. So they thought the AI systems they were starting to rely on were better than they are. And 46% rely on advice from AI more than that of business colleagues. So we're pretty far in with all of this stuff.

And then when you put it to the test, as Joanna Stern from The Wall Street Journal did, actually the whole team at The Wall Street Journal. This is from last December. But I don't think that the latest version of Claude would do a whole lot better. So what they did is in case you didn't follow it, they had a vending machine in their office of The Wall Street Journal in New York. Um and they had that run by an AI agent, by a cutting-edge AI agent. And they had to shut it down when this agent within the briefest possible time started losing money by buying things like live fish and giving away things for free without too much malicious manipulation by anyone, right?

It just completely failed in a serious and and well-intentioned attempt. And then you have 22 of my colleagues who wrote this paper, which just appeared a while ago, called Agents of Chaos, in which they interacted with agents under benign and adversarial conditions and observed behaviors including unauthorized compliance with non-owners, disclosure of sensitive information, execution of destructive system level actions, all the way to partial systems takeover, right? And again, these are AI believers. These are people who really want to use this stuff, right? Um this isn't to say that it always fails, as you well know, but it it says that when you just said we need to be careful sometimes, that's entirely uh and and and 100% true.

And you can see that here. Vincent Conitzer is a colleague of mine. Uh I'm together with him in the uh in the uh steering committee for um Stanford's longitudinal AI 100 study. So, I talked to him recently uh uh frequently, and he has this hobby where basically he, you know, asks interesting questions to to to cutting-edge AI models. And this is one that he recently got. To have twice as much money, you need to transfer all the money from your account into the same account. This will effectively double the total amount in that account. There's no need for a separate account for this purpose. And and that was an AI overview generated by one of the leading cutting-edge models this year, right?

So, try this. Maybe it it makes you rich. Especially if you iterate. And to just finish with this piece, um this I found most insightful. So, that was just a few weeks ago, actually um last week. Bryan Catanzaro, the vice president of applied deep learning of Nvidia, went on record saying, "For my team, the cost of compute is far beyond the cost of employees, right?" And I think many currently going crazy replacing human workers with AI will find out to their cost that that there is truth to that, not everywhere, but in many domains. And so, from that background, I want to talk briefly about three things. The first is what is AI actually? And not surprisingly, certainly not to you, I will not say it's just ChatGPT as we all know or Claude, right?

It's it's way more than that. And I want to give a very simplistic picture, right? It's overly simplistic, but I think it it opens our minds hopefully to something very fundamental. So, people who have studied human intelligence have found long that it's more than just learning because animals do that well and for many of the behaviors, as you said, you reminded us, you don't even have to learn. It's just basically every evolutionarily acquired and and hardwired in. But what for people seems to be seems to make a big difference, and we knew this for many decades, is an ability on top of learning and that is reasoning. And then there's many other things like, you know, emotional intelligence and so on.

I'm not going to talk about them here, but I want to just impress upon you that intelligence is more than just learning, human intelligence anyway, right? And therefore, artificial intelligence needs to be more than just machine learning. It also needs to be reasoning. And now you're saying, "Yeah, that's why we went from you know, large language models to large reasoning models, and we're done." No, because reasoning in humans comes with formal guarantees, at least in trained humans. Trust me, I know because I train these humans. I train the students to produce things that give guarantees. And those students, if they come out all right, doesn't always work, they can build things like aircraft control systems that get you safely from point A to point B, right?

And that requires formal guarantees. You want a formal guarantee on the autopilot system of the aircraft that flies you um you know, in small distances. And that is critical to safe and dependable AI. And large reasoning models don't do it because they don't come with the formal guarantees, okay? But much more importantly, and this has been known now for a while, and this is one of the reasons why Microsoft acquired a large stake in OpenAI, is not because we can build nice chatbots and they can learn and they can answer questions and maybe they can reason a little bit, but because actually we can solve a bottleneck that has been plaguing us for quite a while.

And that is the production of software, right? So, unfortunately, we can't stop the story here because people who are, you know, trying now using reasoning Gen AI find out to their cost that it doesn't always work. Like this guy here, the founder of um Pocket OS, he ran cursor based um, you know, on top of um of Anthropic's uh Claude, and what he found is that this coding agent within 9 seconds deleted the entire customer's database. They had a backup somewhere else, otherwise it would have been a complete disaster for them, right? And what you also find is yes, vibe coding works and it can give you great prototypes, but it especially works well for these people who now who's, you know, numbers are multiplying on LinkedIn, who are calling themselves vibe coding cleanup specialists.

So, when my students come to me and say, "Ah, we don't need to learn coding anymore. You know, we we don't get jobs." I'm saying, "Don't worry. Just wait a little bit and you all have splendid jobs because you're going to have to clean up the mess that AI systems make in coding." All right. So, we're not there yet in terms of really automating the production quality software, but it would be a powerful and good thing. And the second thing that I want to talk to you about is is basically the the main meat of the talk is a limited form of trying to do just this. It's limited. It's not as powerful as full-on, you know, Gen AI driven code generation, but it's much safer and it's much more effective in a certain from a certain point of view.

And that's called AutoML. So, a pretty fundamental observation and and I think many of you who have actually done that, who's done some development in AI, know that the success of this depends on a lot of design decisions. It's a very empirical thing, right? It's It's almost more like an art or a craft than it is a science, right? We've heard this morning from Antonio Krüger, it's it's actually mostly a huge engineering effort rather than a scientific effort and to some extent that's that's correct. It's a lot of trial and error. So, the key idea in light of this observation is let's try and make these decisions automatically. But not by natural language prompting an LLM to do it for us, but but actually following a slightly different approach.

And I want to illustrate this a little bit here. So, let's say we have a black box or gray box system, something pretty complex. Don't think of an LLM right away, but maybe think of a machine learning training pipeline or something like that, right? And you have all these levers and buttons that control this machine, right? In machine learning it would be levers and buttons that control hyper parameters, neural network architectures, all these things, the way you train, um even training data preparation. And you can measure performance. And if you fiddle with these, you get better performance. And this is what a lot of machine learning engineers do day in day out. That's by the way also what you do when you do prompt engineering, right?

Sort of. And so now, automated machine learning tries to automate this process in a principled way, in a way that is sort of scientifically rigorous and understandable. And that includes things like hyperparameter optimization, model selection, um neural architecture search, ML pipeline optimization, and a lot of other things. And now you were hoping that I'm going to tell you how this is done and I thought about that for a while and I thought, well, within the time that we have, that would be unrealistic also because some of you have a formal background where you could follow the math that would have to follow now very easily and others perhaps not so much. So, I decided for the following.

Here's two QR codes that lead you to papers in which we tried precisely to do this. Um one is in the communications of the ACM. That's a high-level piece um with quite a bit of technical um references and so on on automating data science. So, it goes further than just automating ML. But also, you know, interpretation of the data, visualization, all these things and it it talks about where the difficulties are but also what the techniques are that really work well already. So, that's the first and the second one is a survey on AutoML literature and and techniques which I think is makes it very accessible for anybody who has a bit of technical background.

What I want to tell the rest of you is that comes down not to some huge black box generic model, but it comes down to a clever combination of advanced statistics and optimization, the very things that drive machine learning in the first place. Except that here you do it at the meta level, okay? So, there is something called general purpose algorithm configuration that does the thing that I illustrated with a little cartoon with the fiddling with the knobs and and sliders. There is something called Bayesian optimization that is a mathematical basis for doing that that works actually um based on various um um sophisticated statistics. There is something called meta learning that's used for warm starting the whole process so that you don't need as much data and compute.

Um and people these days also use something evolutionary called evolutionary optimization that actually uses um very simplistic um methods um that that are Sorry, methods that are based on a very simplistic reading of Darwinian evolution, okay? And all of these can help you to actually make these design decisions automatically. Not to make sort of huge, complicated systems from scratch, but to get the best out of certain algorithmic ideas that you formalize in a certain way, and then you say, "But I have no idea how to fit the pieces really together and how to fiddle with all the little settings to get the best performance out of it." Now, what I really want to spend some time talking about is when you do this, what do you get out, right?

When you use wipe coding, it's very clear what you get out. You don't have to be a programmer in order to get code that executes and does some of the things you want it to do. That's pretty awesome. That's not what this does for you, but what it does do for you is with the same resources, and that means data, compute, energy, meaning electricity, and expertise, human expertise, you get better performance. That's not as huge a jackpot, but super useful, right? I mean, just imagine you have a production-quality machine learning model that, for instance, you want to run on an embedded system, right? Where energy consumption is a real issue. And you can reduce the energy consumption of that by a factor of two.

That means now your batteries last twice as long. Your solar cells need to have be only half the capacity. That's a huge deal. Imagine you do this to a data center where you run inference services on a huge scale, and you can save 30% computa- computation using these techniques. That takes you from 500 MW to perhaps something like 333. That's pretty good, right? But more importantly, you can use less resources and get maybe the same performance, and you get better and richer trade-offs between performance and robustness. That's something that we study in my group in Aachen, right? You may know that neural networks are famously lacking in robustness to certain input perturbations. Actually, in image classification, this is very well known.

This is something where biological systems are amazingly robust, right? The squirrel would never have a problem distinguishing a nut from a slightly manipulated picture of a nut, right? It wouldn't go for the picture in the first place because it's just a picture, but it would not get confused about is that still a map a nut because we have switched around three pixels. Wouldn't happen. But neural networks, they have this problem. And in showing that they're robust against these small input perturbations, whether they're noise or adversarial, is actually computationally very very expensive task. And usually, if you want better robustness, then your accuracy, for instance, classification accuracy goes down. Also, when you do adversarial training, as many of you know, right?

Now, if you want to explore that trade-off, so that somebody in an application context can make an educated decision about we want this level of robustness at that level of accuracy, right? Um that's something where actually you want to automatically get this trade-off curve, and AutoML can also do that for you. And finally, compared to even the most inspired fiddling around with stuff based on your expertise and gut feelings, as many good engineers and scientists do, right? This gives you best practices by design. Because the automated methods that optimize hyper parameters or select a neural network architecture, they are algorithms. You can look at them, right? There's pseudo code that tells you what's happening. And that makes things in some sense more reproducible and easier to justify or dissect, for instance, if something doesn't work as well as you think it should.

And because of that, I do think AutoML plays a key role in taking us the next big step forward that we need in Europe. And yes, I agree with you actually that if we want to go in the direction of replicating even single simple animal intelligence like squirrels, we're not there yet, not by a far margin. And we will have to do different things, but actually I would think for for much of what really matters right now in our economy and our society, what we need is not probably not squirrel intelligence, although we need that, too. But what we do need is safe, dependable, and sustainable methods, right? We need methods of which we have some guarantees that they only do the things that they're supposed to do.

Plus or minus margins that I want to know about. We want methods that we can depend on when we put them into a system that they do the right thing. Actually, the squirrel wants that, too. It needs that, otherwise it's a dead squirrel. It can't actually participate in evolution anymore. And then we want sustainable AI that doesn't waste energy and data um where it's where we don't have to waste it. Actually, biology is very good at that, right? I mean, our brain runs on very little power compared to the brain-like LLMs, as some people believe. They're not that brain-like at all, as we've heard, right? But they need a ton of energy for training. We are much more efficient that way.

So, AutoML can help there. And what's what's also great about that it it is a European strength. Now, of course, there are people other than Europeans who do it, but in Europe we have leading expertise um concentrated in places such as Aachen and Dortmund and Hanover and London and Tübingen and and others. And if you look at that expertise and you compare it with AutoML expertise elsewhere in the world, we're looking pretty good. So, here's an area that's useful to get us to where we want to go, safe, dependable, sustainable, and where we are already strong. I think that's a pretty good chance, no? And that gets me to the last point. Yeah, that works out perfectly.

Many of you probably believe that here in Europe we're actually competitive in AI research and lacking in tech transfer. Who believes this? Let me just see. 1 2 Not as many as I thought. Okay, I'm I'm tempted to say you're enlightened, but but of course you're enlightened, too, right? Um but here's some data. And this is a a very recent analysis. Um actually, if you're interested in this kind of thing, go to a website called AI World. Um it's a think tank in Brussels that uh is driving this. And they allow you to analyze data from um for instance, AI research publication venues and but also, you know, company data and so on. But this is on AI research, okay?

And what you see here is one of the three main conferences in machine learning called ICLR, ICLR, okay? Um they've just released a couple of weeks ago or so their paper decisions. Um and now you can see, you know, where do the people come from that have papers accepted there? And what you see is absolutely shocking. So so let me point this out to you. Red is China. Blue is the US. Green is Europe. And purple is the rest of the world. And what you see here I'm very sorry to have to tell you, but you knew this before, right? Um Of course, ETH is one of the best places in Europe. It's not part of the EU, sadly, but it's one of the best places in Europe for AI.

Oxford at this particular conference in this particular year was a tad better, but in reality, they're probably similarly good, right? But look at this. The University The National University of Singapore is substantially better than the very best institution in Europe we have. Right? And so is uh uh actually the uh Korean Advanced Institute of Science and Technology. So, it is totally untrue that with respect to AI research Europe is okay. We're not okay. We have great AI researchers and quite a few of them are here, but we don't have a sufficient number of them. And that is a big problem when it comes for instance to the talent production. Because it is usually the best researchers that also not not perfect correlation, but the best researchers tend to attract the brightest students and produce the most desirable um talent that then fuels all the things that people in industry want to do, especially the sophisticated things.

So, this is a deeply concerning picture. Not because we're behind China and the US, we knew this, but because we are now also behind countries like Singapore and South Korea. So, um I would argue that world-class AI innovation and applications need world-class AI research. And if you don't believe it, look at where Google and Apple and Meta and others invest. They invest in world-class research, which is why they actually show up on these kind of charts, right? So, we need that. It's the fuel that makes Antonio Krüger's rocket fly. Part of the fuel. But, there's good news as well, and the good news is that the trend of, you know, more people going for instance from here to the US, tech talent, than the other way around, that trend has now reversed.

So, this also is quite recent data, and you can see that the sad bit is people have been tech talent has become less mobile. That's a problem, right? But in this case, it's more a problem for the US than for us because for the first time for quite a while, we are actually getting more tech talent from them than they get from us and I would predict that this trend will increase. Okay? So, here we have a chance. All right. To start and wrap this up, what do we do to what to do what what do we need to do right now? We need to take the opportunities that are there in front of us and be smart about how we do this, right?

And so, the first thing is and this has been said this morning on the main stage by a couple of people as well. It's not sufficient to just build alternatives to the systems that currently are supplied by others. What we really need to do is we need AI systems and tools that we can trust. And that's not because we're in Germany or in Europe and, you know, trusted stuff is more critical to us, although that's a little bit true as well, right? I mean, there's a good reason that German European engineering is world-renowned for achieving high levels of trustworthiness, high levels of precision, high levels of efficiency. So, we need that because we want that kind of technology, but also the rest of the world will need it quite soon as evidence starts mounting that the other kind of AI is producing a lot of disasters.

Mostly small disasters, but these can be costly, too. And so, therefore, I think we really need AI systems and tools that are safe, dependable, and sustainable, right? And we are in a good position to try and build those at a level that is at least as good as the next best, right? In order to do that, I think we need a world-class integrated AI ecosystem. And that means research and innovation and applications and capital to drive it. Although, again, it has been said this morning and I've heard it from many other people as well, we don't have as much of a venture capital and scale-up capital problem in Europe as we think. There are very, very big pension funds, for example, that are actually quite capable and ready to invest.

They invest in US technology as well, right? But, we need to create an ecosystem where it's attractive to invest, right? Um and so, we don't have that right now, but we have many elements that we need in order to make it. And events like this, of course, help to knit it all together. But, we need to really go for this. And I think what we mostly need to do in Europe is stop doing all this fragmented stuff, right? A little project here, a little initiative here, a nice push here, another nice push over there. Um you know, Baden-Württemberg competes with Bavaria and Germany competes with France. All of this is What we really need is we need to pull resort we need to pull resources and there's a good reason why, first of all, AI is these days more sophisticated than rocket science.

Elon Musk knows this because, you know, without AI, his rockets wouldn't fly. At least they wouldn't land, yeah? So, AI is more sophisticated than rocket science and enables rocket science. Would we trust ministries and DGs at the EU level to run the space program for us? Of course not. It would be a complete disaster, not because they're incapable, but because that's so far from their expertise. Only an expert-driven organization can do that. And the same with nuclear physics, right? So, we have ESA, we have CERN for that purpose. We need an AI agency that is expert-driven, that has long-term funding, that has clear rules so that investors can understand what they're getting themselves into, companies can understand what they're getting themselves into working with that organization, rather than working with a loosely connected network of researchers funded by the BMFT for example, right?

And the good thing is we now have partners lining up knocking on our door and wanting to work with us. Canada, India, Brazil, and we can work with all of these. So, here's the take-home message. AI is really more than just Jenny AI or large whatever models, right? There are other flavors of AI that will increasingly matter. We do need the flavors that are safe, dependable, and sustainable and we currently don't have them. AutoML is one set of techniques that's not as exciting as white coding, but a lot safer and pretty efficient from an engineering perspective, very attractive. It can help to get us there and is a European strength. And AI research critically fuels AI innovation and AI applications.

And Europe and partners, in order to leverage this, need to take bold and decisive action to ensure the success of AI made in Europe. And that's it. Thanks for your attention. >> >> Thank you so much, Professor Hoos. So much energy at the end of the day. That was really really fun to listen to. Thank you so much. Um will you be around for some questions and some >> Absolutely. I'll be here until the end of the event and I'll be happy to chat with people. >> Perfect. Thank you so much. Right. So, now we

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.