Rise of AI Conference 2026 · Berlin · Applied AI Stage

Brains, Symbols and Intelligence

Christoph von der Malsburg

At Rise of AI 2026 in Berlin, Christoph von der Malsburg argues that present AI, impressive as a passive "encyclopedia," remains a black box that learns inefficiently and cannot yet safely act on its own, citing Anthropic's refusal to let a government control weapons with its AI as a sign of how shaky current control is. He proposes animals as the model for future "agentic AI": many are born with most of their behavior already in place, and should be able to learn a new environment the way an employee learns a company, by observing rather than months of training data. He argues biology-based AI is regaining interest, citing ventures founded by Jeff Bezos, Yann LeCun and Jeff McCaleb and new German funding calls, but says none of them yet knows how a squirrel does what it does. Drawing on his own decades of research, he lays out a framework built on structured perception, a "jigsaw puzzle" model of object representation in the visual cortex, mapping between sensory input and abstract categories, and innate behavioral schemata, closing by arguing Europe could out-compete data-hungry American and Chinese AI with an approach that does not depend on mass data.

Key takeaways

  • Christoph von der Malsburg calls today's agentic AI "playing with fire," citing Anthropic's refusal to let a government control weapons with its AI as proof control remains shaky.
  • He argues agentic AI should learn a new environment the way a new employee learns a company, by observing and asking, rather than through months of pre-training.
  • He identifies the binding problem, representing which neurons belong together as one coherent object, as AI's central unsolved representation problem.
  • He built a fully neural face-recognition model, following work from two companies he co-founded around 20 years ago, that recognizes faces in about 20 milliseconds.
  • He argues Europe could compete in AI through biologically inspired systems that gather data through their own curiosity rather than mass data collection.

Detailed notes

What is wrong with present AI

Von der Malsburg opens with the practical problems of deploying today's AI in a company: it learns inefficiently, requiring laborious new data for every application, and is a black box even to its creators, making its behavioral control unreliable. He distinguishes two paradigms: AI as a passive "encyclopedia" that gives impressive answers but only reacts, and the emerging push toward agentic AI that acts directly on a computer or in the world, which he calls "playing with fire" at its current reliability, citing Anthropic's refusal to let the American government control weapons systems with its AI as evidence of how shaky control is. He also points to self-driving cars, still needing millions of miles of human-driving data, as proof direct AI-world interaction remains unsolved.

Animals as the model for agentic AI

Von der Malsburg argues that agentic AI should be able to "learn on the job" the way a new employee learns a company, by observing and asking questions, rather than through extensive pre-training. He points to animals as proof this is possible: many are born with most of their behavior already built in, such as a pigeon able to fly out of the nest or a squirrel able to judge branch strength, find nuts, avoid predators and build a nest without being taught. He has advocated biology as a model for AI for decades against mainstream resistance, but says the mood is shifting: he cites Jeff Bezos founding Prometheus, Yann LeCun founding AMI, and Jeff McCaleb founding two further companies on AGI and on biology and neuroscience, led by Dileep George and Doris Tsao, each funded at the billion-dollar level, alongside smaller German funding calls he has applied to. He cautions none of these ventures yet knows how a squirrel does what it does, and that the roughly 15 to 20 disciplines studying brain and mind largely fail to talk to each other.

Structured perception and the binding problem

Von der Malsburg argues that raw sensory input is an "ocean of complexity" that never repeats exactly, but that the physical world has deep structure, comparable to how computer graphics generate realistic scenes from underlying rules. Citing the no free lunch theorem, he argues a learning system must be tuned to its application domain rather than seeking one universal optimum. He identifies representing a coherent object as the field's central unsolved problem, contrasting symbolic AI, which uses compact symbols for whole objects, with neural approaches that represent detail at the level of individual neurons but cannot capture which neurons belong together as one object, the binding problem, which he says he first proposed 40 years ago and remains largely unaddressed. He offers a jigsaw-puzzle analogy: pieces of a scene, or "net fragments" of neurons in the visual cortex, only activate when they fit their neighbors, which is how newborns segment an object from its background using continuous motion alone.

Mapping and abstraction

Von der Malsburg describes co-founding two face-recognition companies, one in Bochum and one in Los Angeles, roughly 20 years ago, built on mapping an incoming image onto a stored template, work he says the field later abandoned, though he has since built a fully neural model that recognizes faces in about 20 milliseconds. He connects this to Moravec's paradox, roboticist Hans Moravec's observation that AI can play chess but not replicate what a mouse does escaping into a hole, and argues abstraction, relating the concrete world to categories such as language, is achieved through mapping. He traces "schemata" from Immanuel Kant and Edmund Husserl through to Marvin Minsky's and Roger Schank's frames and scripts, arguing those never solved how to connect abstract descriptions back to real situations, something children do effortlessly, as when a simple drawing is instantly recognized as "papa."

Instinct, imprinting and cooperative networks

Von der Malsburg argues that behavioral control, driven by instincts and drives in the midbrain, is a missing piece agentic AI will also need, since humans are guided by hunger, danger and similar drives far more than commonly assumed. He cites Konrad Lorenz, Nikolaas Tinbergen and E.O. Wilson's descriptions of innate perceptive and action schemata, including Lorenz's gosling imprinting on a wiggled boot mimicking its mother, still used to train geese to follow aircraft on new migration routes, and the innate infant schema for a simplified face, which later develops into recognition of seven distinct facial expressions. He summarizes his own framework, "cooperative net architecture": only a small fraction of the brain's neurons, an estimated one in 1,000, are active at any moment, each held active briefly by roughly two dozen connected neurons, forming a self-consistent pattern across the whole brain through self-organization, comparable to a physicist's energy function.

A path for Europe

Von der Malsburg says his personal strategy has been to wait for the limits of deep-learning and backpropagation-based AI to become visible, and that renewed company founding and government funding calls signal that shift is underway; an earlier attempt of his own to start a company eight years ago, backed by Berlin VC Blue Yard, failed because young engineers at the time wanted only deep learning, whereas he has now assembled a team for a German funding application and is preparing another application to a second German funding program. He argues American and Chinese AI is built on large-scale data collection Europe is reluctant to pursue, and frames biologically inspired AI, which gathers information through curiosity and play rather than mass data, as a technology Europe could use to start over and compete, despite established AI already attracting hundreds of billions in investment against almost none for this approach.

Quotes

The present AI is a completely black box, even to the people who have built it.

Christoph von der Malsburg

None of those companies I've spoken about have a clear idea how a squirrel does it.

Christoph von der Malsburg

As soon as you are hungry, or as soon as you get into danger, you know very well who is calling the shots in your brain.

Christoph von der Malsburg

Questions answered in this talk

Why does Christoph von der Malsburg think today's agentic AI is risky?

According to Christoph von der Malsburg at Rise of AI 2026, present AI is a black box even to its own creators, making its behavioral control unreliable, which he illustrates by citing Anthropic's refusal to let a government control weapons with its AI.

Why does he use animals as a model for AI?

According to Christoph von der Malsburg at Rise of AI 2026, many animals are born already able to perform complex behavior, such as a squirrel judging branch strength or finding nuts, which he argues shows a path to AI that learns on the job rather than needing extensive pre-training data.

What is the binding problem in brain and AI research?

According to Christoph von der Malsburg at Rise of AI 2026, the binding problem is the unsolved question of how the brain represents which individual neurons, or which parts of an image, belong together as a single coherent object, a challenge he says he first proposed around 40 years ago.

Mentioned

Agentic AI, Binding problem, No free lunch theorem, Moravec's paradox, Schemata, Cooperative net architecture, Imprinting

Keep exploring

More from Christoph von der Malsburg

More from Rise of AI Conference 2026

More on AGI & Future of Intelligence

Transcript

Auto-generated transcript, not edited. · 4,398 words

Show the full transcript

Thank you. I have been put into this session which is really about application of AI of the present AI. So, I am a little bit misplaced, but I can expect you all know how difficult it is to work with present AI if you want to introduce it into your company and into a factual work. Um the first sorrow is of course learning inefficiency. When you want to apply something to a specific field, you need new data. That is very laborious. That makes the whole thing very expensive thing. Then the present AI is a completely black box even to the people who have built it as you know. Nobody knows what what goes on in the in the machine.

What decides on on what it decides. And so, it is um a dangerous thing because the behavioral control is a sort of squeezy kind of problem. There are actually two paradigms you can think of of AI. One is the present one, AI as an encyclopedia. This is really awesome. When you open your chatbot and ask the most wildest most intricate question in the most intricate field, you get awesome answers. It's it's tremendous encyclopedia, but it is totally passive. It waits for your question. It's it's not really an active thing. There are these movements today a lot of spoken about here in this conference about agent AI giving the AI the possibility to act on your computer, but I think at the present time it's just playing with fire.

It's a very dangerous thing. Just think of Anthropic's decision not to allow the American government to control their weapons with AI for a very good reason because your control is very indirect and very shaky. And then there will be the hopeful thing in the future agent agentic AI that you can rely upon to do the things for you that you wanted to do and to to take action. Now, um a big burden a big uh a block on the on that front is the direct interaction of AI with the world. That is a tremendous limitation of of present uh the present technology. Just think of autonomic autonomous cars. They have to be had to be trained with millions of miles of humans driving cars and and the system collecting data.

Uh direct um interaction between the system and the environment is not a done deal. It has to be uh hand-held by by humans by giving pairs of realistic input and abstract interpretations. So, you would like an agentic AI to be able to learn on the job. If you want to onboard a person into your company or a team, you rely on the person or the team to know what in a general fashion what the business is about, what basic kind of applications about. You ask them into your company, you say, "Look around. You can interpret you can find out what is going on here. You can interview people. You can read documents." And off you go.

It's a very easy process, right? And compare that to um onboarding AI to your system. [clears throat] And then you have to rely on the behavior. I I just put the word character here, but uh uh you know, reliability, responsibility, knowing who uh who has to report to whom, um all the those things um you can you have to expect from uh identic AI. Now, the big uh example, the big model for identic AI is of course animals. And uh a very basic fact uh about biology is that many animals are born with a full complement of their behavior. They don't need to learn anything. You know, a a pigeon that jumps is is ready to jump out of the net uh nest or a squirrel, they know what's going on.

The pigeon can swoop through the branches of of trees in in a very delicate way, land on on a on a on a on a tree, or a a squirrel can find uh its nuts and and avoid predators and and knows how to do do build a nest, and just jumping from one branch to another, judging how strong that branch is, what the distance is, and so on, they bring that to the table. Now, there is um uh I have been advocating uh this model for through all of my life. For decades I have worked on uh I trying to understand how that works, but I was completely against the uh the mainstream. My wife always makes a joke about me.

You remind me of the of the guy who sits in his car in the morning driving to his job and on the on the radio they say be aware on your freeway there is a somebody coming the wrong way. You know what the one one one person all of them. That is my situation but I feel now the the leaves turning. It's a lot of talking about taking biology as a model. Jeff Bezos has >> [clears throat] >> founded this company Prometheus and founded with a billion. Yann LeCun has founded this AMI and he got founded with billion. And then there is Jeff McCaleb with his company with his Astro also. He just founded two companies.

One about AGI and one about biology and neuroscience led by Dileep George and by Doris Tsao and founded with a billion. And then there are two programs here in Germany. There is a BMFDR call for novel technology of of AI. I applied last week. And there is one by SPRIND that is looking for novel technology of AI. They they throw millions not billions but millions on this subject. So I feel the time of a change in perspective has come. That is all wishful thinking for the moment because none of those companies I've spoken about have a clear idea how a squirrel does it. Neuroscience or different sciences that are working on the problem of brain and mind there must be 15 or 20 of them, you know, psychology and neurophysiology and computer vision and philosophy and and all of them.

A psychiatry They speak different languages. They look at part of the problem and the whole problem of how the brain brain works is totally has been totally stalled for decades in a tower of Babel phenomenon. They don't talk to each other. They can't talk to each other. I know what I'm I'm talking about here. Okay, so what you get from me is is pretty unique. I give you some of the uh um main the main perspectives that are necessary here. The main problem as I said before is that the natural >> [clears throat] >> sensory input that you get from the outside is tremendously complex. The images alone that you get in are subject to your eye movements, are subject to perspective movement, are subject to all the movements that go on out there.

It's an ocean of complexity of variance. None of these images ever repeats. So, what do you do about it? >> [clears throat] >> One realization must be that that the world out there has a lot of structure. There is physics, there's chemistry, there is computer graphics. Think Think of computer graphics. They are able to create visual scenes in all complexity. They know how to create them, how to how to interpose objects and how to deal with texture and illumination and all of that. So, that should be taken as a model. If you say that to a computer vision person, he says, "Oh, yeah, that's very unfair. That's a forward problem. We have to invert that problem, go from the images to the to the scene description behind it.

Um but uh that is an example of how structured the world is. The um I will come to uh one or two of the important uh uh uh points here. So, the brain has to be has to be tuned to that world. Some of you may know the uh no free lunch theorem saying that an optimization learning system has to be tuned to the application domain. There is no optimal system if you [clears throat] believe you have the optimal system, you it's trivial to find examples that uh wreck your system. So, our brain is tuned to the world. What are the important points here? And then there is um in natural intelligence something that is uh all the people I I've talked to today and all the people you talk to about artificial intelligence, behavioral control, like uh for instance, empathy, feelings, uh internal drives, all of that is taken to be a human uh um phenomenon that will may never be um um realized in in AI.

But of course, if you want to have agentic AI, you need the equivalent of behavioral control like in a squirrel. The squirrel knows exactly what to do, how to find a a sexual partner and how to how to build a nest and so on. So, that is the uh that is the problem you have to uh to face. Now, um the first step I want to talk about is um you have to cut the uh tremendously complex input I spoke of into pieces. They are structured pieces. The image of a person is something that is structured to it and you have to cut the individual object out of the scene in order to to find regularity.

And if it works, ah yeah, here you have an example that is uh is it's an amazing thing that your brain is able to to deal with this this array of of pixels you have never seen before, but you know exactly what's going on. You you can distinguish two fields of pixels one moving the one way the other the other and you can put the whole thing to the together as a rotating translucent sphere, right? So the it is known from neurobiology that newborns, human newborns, animals they know to segment object out of the background on the basis of continuous motion. Like you have here, continuous motion. So the brain is prewired to do that, to find pieces of structures that are coherent in that sense that they in within the object you see motion and against the background there is no continuity.

In order to to capture this kind of structure you need to have a clear picture and now I come to the keystone that is holding up the whole field to the question how do you represent a coherent object? As you know, there are two camps in AI, the good old-fashioned AI, GoFAI they call it. Um the one camp is inspired by biology by by mathematics and thinks in terms of uh symbols. Symbols are compact entities that talk about the whole object. They have uh great advantages but they have also great disadvantages. The other camp, the neuro fuzzies they they were often called, they believe [snorts] that you need to represent detail in the form of individual neurons, and they have that disadvantage, too, because they don't know how to how to capture a whole set of neurons that belong together in the sense of an object, for instance, the binding problem, which I have theorized uh 40 years ago, which is still sort of um brushed under the under the carpet these days.

So, how do you represent a coherent object? What is the representation in our brain of mental objects in general? Now, here I come with a an analogy, a jigsaw puzzle. A jigsaw puzzle is put together out of pieces that you can see here, and the condition for a piece to belong to place, of course, that it fits its neighbors. It's a very strong, very type type um uh condition. And the representation of uh now, this is not not appropriate, because it represents a whole scene. Imagine a jigsaw puzzle uh uh uh jigsaw puzzle that represented just a dog, just a just an object. So, the object is a holistic thing. It takes its meaning the the meaning of the individual pieces are are taken from the whole thing, from the the dog as a whole, and the dog as a whole is of course composed of the pixels.

Now, how do you translate that into the the brain? The pieces in the brain are not moved around. So, a primary visual cortex, the place where the image come into your brain, >> [clears throat] >> a two-dimensional sheet of neurons, has in each position a large collection of jigsaw pieces, so to speak, local texture representing net fragments, as I call them, and these are activated by the incoming images in image, but they can only get active if they fit to their neighboring uh patches, neighboring um and uh network net fragment. And uh so the whole thing, the the incoming image and and that condition that it fits is, of course, the coherence inside an object, like coherent motion.

So, um the the uh the first and and and simplest version of this is, which is pre-wired at birth, that in each position there are different tiny networks that correspond to different directions and speeds of motion, uh to be selected by the actual image, and one of them can only be activated if the neighbor if the neighbors talk about almost the same motion, if you have motion continuity. So, you get the picture of the very simple model here, and then the whole thing learns very quickly not to to to pick up local motion, but also um local texture, and so on. So, if you have that, you can make use of a tremendous simi- um structural regularity in the world that the present technology is totally shunning.

It's amazing. The different images that I take in from an object, like your your your body, for instance, uh >> [snorts] >> the very different images that move around in your in your retina, that move around under perspective motion, under illumination, uh rotation in depth, and so on, they are all structurally related to each other. You can map them to each other. So, you have to have the ability to map. And in order to map, you need a representation. All the details in the right arrangement, right? I think this is is a is a no-brainer. So, let me go back to the previous slide here, mappings. Um I had actually uh co-founded two companies, one in Germany in Bochum, and the other in Los Angeles, and they were based on face recognition.

For quite a while, I was a uh these companies were uh world champion in face recognition. This is 20 years old. Uh 20 years ago, we made some money, but uh they the technology went over this idea for reasons I don't understand. Isn't it obvious that in order to recognize a face, you take a model template and compare it to what you see in the image? It's It's Isn't that a no-brainer? So, uh back to neurobiology. Here, I run into my colleagues who say, "Oh, Malsburg, you are crazy. There is no mapping in the brain." In order to realize this kind of thing, this is the primary visual cortex where the image moves as quickly as as as Obama moves or your eyes move here in in the fusiform complex of your brain in in temporal [clears throat] cortex where you have your template of the face of Obama, um you need to establish a mapping, activate nervous fibers between thousands of them, between points in the in primary visual cortex and point in the

template, activate as quickly as the as the image is moved. I got stalled on this problem for two decades. I had geniuses of students who who helped me uh solve that problem. Believe me, it's a done deal. Uh the I I have a a completely neural model which uh recognize faces in 20 milliseconds or so. Okay, mapping. Um So, let me go forward. Um obstruction. This is also this is a key issue which is totally ignored or so sort of uh well, one has to actively uh look away from it. As I said in the beginning, the the first problem of intelligence in the real world is you have to deal with the vastness of the incoming signals and relate that to an abstract level on which you can uh exert intelligence.

You all know about Moravec's uh paradox, don't you? Moravec, a roboticist, who said, "Oh, you AI people, you can uh you can play chess. Yeah, yeah, all these highbrow cognitive things, but what a little mouse does when it it disappears in a hole, you can't do." That is Moravec's paradox. Uh AI is acting on an abstract level. Uh LLMs work on the level of language. Language is a highly abstract thing. When you say bottle, that relates to a zillion of different images that could correspond to bottle. So, how do we establish the relationship between the variable, the variant of the world, and the this abstract level? Mapping, of course. There's a philosophy starting with Immanuel Kant in the 18th century speak of schemata.

Uh it's uh Husserl speaks of them. Um cognitive science uh speak of them. the gofi the um the other camp have talked of you know people like Marvin Minsky and Roger Shank have talked about frames or scripts or so abstract descriptions of what is what what a restaurant scene for instance but they didn't know how to relate these abstract scripts to real situations. It is a mapping problem of course and this mapping is done by our kids with no problem. I mean kids this is a typical children drawing and when you say what does it seem oh can't you see it it's papa here there are the two ears the two eyes there are the hairs everything's there what is missing?

They see the different the the relationship between these very abstract representations and the real world without any problem. That is a basic a capability of the brain. abstraction and I think I gave you the essential the essential pieces to go from the the present technology to a completely new kind of technology. One missing piece is behavior. You know we humans always speak as if we are not guided by instinct or or building drives we are all very rational beings no wrong. I mean as soon as you are hungry or as soon as you you get into danger you know very well who is calling the shots in your brain. When you go to a precipice which is really deep not not like here you physically pulled you feel physically pulled back.

We are all driven by these tiny structures in our midbrain. >> [clears throat] >> Intensive research now in neurophysiology is extremely complex complicated to find out how this works. The description of the machinery that installs drives in us has been given by people like Konrad Lorenz or Nikolaas Tinbergen or um E.O. Wilson a long time ago. They speak of schemata uh perceptive schemata that for instance which are triggered by the outside world and uh um action schemata. So the prime example is of course How are we in time? Okay. Uh I'm near to my end. The the prime example is of course Konrad Lorenz's um example of the little gosling hatching from an egg. And the first schema that is activated in the brain of the of the gosling is find your mother.

And the schematic description of mother he tells us, we have to believe him, is your mother is a moving object in your neighborhood that goes ga-ga or so. Uh and uh Konrad Lorenz took a Wellington boot and wiggled it in front of the goose and imitated that noise and the goose imprinted on that uh on that uh Wellington boot and followed it around wherever he went. It's a technique he used today to train geese to follow a lightweight planes for instance to teach them their new um new uh migration routes. So in us humans a very potent schema is the face. Um psychologists tell you that you can with a schematic description of a face like a like a uh table tennis bat with two dots and two dashes on it which you hold and and move in front of uh of of a baby, the baby gets fascinated, looks it.

So, this is a schema that is imprinted, that is innate. And then the I see here. There's the time. Then the baby homes in to the caretaker. Um that's a kind of slow imprinting. And as the kid knows to distinguish seven different facial expressions, emotional expressions. And that is an antenna into the cultural world. It's a very important schema. Anyway, the mystery behind how this is realized is that in the midbrain there are uh networks of nodes and links which describe the schemata and which can by a mapping process be connected with sensory schemata and motor schemata. And so, just imagine a new technology that is based on on on on this kind of self-organizing system and is can be endowed with feelings, with feelings of responsibility and and fear and so on.

So, this is my one but last slide. Um so, I have come up to formulate something in which I call cooperative net architecture. Uh the net I've talked enough about nets. There are nets that represent objects. There are nets two of them here. There are nets that represent uh mappings uh which together then form composite nets. So, what I claim is that the state of your brain is correct in this moment, in this fraction of a second, is represent [clears throat] is can be described as a subset of the billions of neurons in your brain that is active subset of 1 in 1,000 neurons they estimate is active and each of these neurons is held active for a short moment by something like two dozen of other neurons with which it has connections and these part descriptions have to fit together consistently into the net that that covers your whole brain.

That is what is going on in your brain. These networks are called cooperative because the local connections have to fit together with like the pieces of a jigsaw plane into the whole thing. This is a matter of mathematics. Mathematics has to be brought in here to this field. What what is possible in terms of cooperative nets is of course a mathematically pretty time thing by by self-consistency. And the whole thing is generated by self-organization by a goal function which the physicists would would would call about an energy function which gives the which measures the agreement of an individual neurons with the other neurons active neurons with which it is connected at the time. So, strategy.

My personal strategy has been just wait for this craze of deep learning of learning based on back propagation of error runs its course and the limits of that become visible. I think this conference has shown has spoken among others about the limits of this thing. I think the founding of of those companies I've spoken about our the new calls for proposals in Germany are a sign that that the feeling is growing. For a while I couldn't I tried to start a company 8 years ago. I got the money. Or I had I had the impression I would get it by by a marvelous little VC here in Berlin called Blue Yard, but I couldn't get a team together.

Young people knew better than I where the thing would would feel. And I talked to them and they rolled their eyes and said, "Leave me alone. Where is the gold fund? Where is the the the error function? Where is uh deep learning? Uh uh leave us alone." The thing has has has changed now. I applied to the BMFT uh program last week with a team of young people whom I could convince. And uh there's another one by SPRIND uh which I am in the middle of uh preparing for. My personal strategy. So, on a larger uh scale, I think uh that has been spoken about much here in this conference, what do we in Germany or in Europe do on this front?

AI at the present time is dominated by America and China, who are very cavalier in going on doing with data. Uh that is something uh Europe doesn't doesn't uh want to go along with. They their richness is the the humongous numbers of of users in the in the in the networks as something uh Europe abhors. So, here is a technology with which you can start totally anew. Forget about data. Those systems pick their data on their own on the basis of curiosity and play. And uh start anew and outrun the established AI. Uh it is uh at present a very unequal fight because the present AI has been has been and is being invested with hundreds of billions of dollars.

This hasn't been hasn't been financed at all in the last two two decades. So a lot of work has to be done, but I'm uh confident that this race can be won in Europe here. Thank you. >> >> Thank you so much on behalf of all of us. I think that was very inspiring. You took us down a rabbit hole that you have been in for a long time. >> That's right. >> So that was really fun. We don't have time for questions, but I think you might be around. >> Yes. >> And so take a chance to pick your brain. >> I'm I'm I'm threatened with an interview right now, but after that I will be available.

Thank you. >> Perfect. Thank you so much.

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.