Transcription
Artificial intelligence is developing exponentially, and it is our only chance, and we should really understand this now and act accordingly. From 2026, it is the year of great upheaval. How do you perceive the current state of human robotics? What is the current situation right now, at this very moment? With AI, we are now at a point where we can truly say, if not now, then when. And therefore, what is not possible today will not be possible in 10 or 20 years, but probably already in half a year. If there is no one left sitting in the office, except perhaps artificial intelligence, who will earn their pension then? Yes, hello Mr. Knoll, welcome to Everlass AI. I am pleased that we are speaking today, Mr. Knoll. You studied electrical engineering in Stuttgart in 1985 and in Berlin in 1988. You received your doctorate in computer science summa cum laude, habilitated in 1993, were a professor in Bielefeld, and have headed the Chair of Robotics, Artificial Intelligence, and Real-Time Systems here at the TU Munich since 2001. With over 1000 publications, more than 40,000 citations, you are an IEEE Fellow and much more, including a Fellow of the University of Tokyo. And recently, you founded your own AI startup, One Alpha. Please help us first to understand how you arrived at this point. What originally drew you into computer science and then into robotics? Well, as you already mentioned, I studied electrical engineering, communications engineering, and computer science. At that time, communications technology was experiencing a huge boom, and the telephone was considered the largest machine in the world. Yes, and switching technology was, of course, a very important aspect or industry in the course of digitalization. There are, so to speak, two areas that defined digitalization. On the one hand, professional communications technology, and on the other hand, consumer electronics. On the communications technology side, it was the digitalization of the telephone network and digital switching systems, and on the consumer electronics side, it was essentially the CD. Yes, that was precisely the time when the CD was being developed. This was also a European development, just as, by the way, the entire compression methods were essentially European developments, as was pulse code modulation and similar things, which were also done in America, but also in Europe and in Germany. We had the leading telephone industry with many companies at that time, and it was naturally quite logical to choose this field, especially since it also fascinated me personally. So, from switching technology, the path to computer science is not too far. Especially when you have switching systems, these are very large systems. They encompass all aspects that challenge you. Yes, security. You have real-time. You have to search and find, so to speak, pathfinding. Yes, if you want to call me, the computer has to decide where you are and so on. And then, of course, mobile communications also emerged at that time. So, there were, of course, analog mobile networks, of course. But digitalization, keyword GSM, which is also a European project, by the way. These were already emerging, and it was a huge field. Now, after that, that was the transition to computer science. Yes. And then you asked about robotics, and robotics was, of course, a field that was completely separate at the time. Industrial robotics on the one hand, but also AI, artificial intelligence, it was already foreseeable back then that these were essentially one and the same. Therefore, if you look at the, we can perhaps discuss this further, one is the automation of the hand, and the other is the automation of the brain. And in their convergence with the body, physical AI embodiment, that is what was clear even back then, and what, by the way, the founders of AI were also well aware of. So, if today people say that these are two different fields being merged, I don't see it that way. Okay, that is very interesting. I would like to delve into that directly and follow up, because I recently had a conversation with a roboticist who told me that artificial intelligence and robotics are now being merged. This is also the trigger why many people are now engaging with artificial intelligence and the subsequent physical AI at all. Driven by generative AI. That was the trigger for most of us to engage deeply with the entire field, and he told me that 5 years ago, he was sometimes chased away from events when he talked about how generative AI and robotics could be merged and how the disciplines were currently blending. How did you experience that? Well, I see it a bit differently. Yes, because if you look at the founding fathers, especially Marvin Minsky in an interview from the early 70s, yes, where he already said that in a very short time, AI will be able to repair cars and everything that, so to speak, operates a body. Yes, you can't, if there's just a box on the table, it can't repair a car. So implicitly, but also explicitly, it was always clear that robotics is a very essential sub-field, or conversely, AI is a sub-field of robotics. Yes. And that was already very recognizable on the research side. So there is the old Stanford Shaky project. Yes, Shaky was a mobile robot that is not so different from our current mobile robots, except that everything was much larger and, of course, not powerful, but things like all the planning procedures, Blocks World, also the question of how to describe geometry in order to navigate between objects, navigation, all of that comes from that era, the late 60s, AI and robotics. From my perspective, it was always one, and you can see that in the early conferences. So, in that respect, I don't see a contradiction, but rather it is actually, if you will, a very natural process that this is now converging. And the reason for this is, of course, that mechanics and mechatronics are now much more powerful, but the first attempts and the first systems clearly come from AI research into robotics. Yes, industrial robotics, as I said, has developed largely independently of this, but it was more about handling processes, a very specific area of human skill. Yes, very interesting. So, I have several questions for you right away about what is currently happening in the field, what your predictions for the future are. Let's first delve deeper into how what we understand today as physical AI, or perhaps a rather new term, physical AI, differs. Perhaps a few decades ago, when we talked about robotics, have you experienced upheavals? Is it driven, let's say, by, say, 2022? Has the entire field developed faster, is it currently developing faster? What are the biggest breakthroughs you have perhaps noticed in recent years? So, perhaps let's start at the beginning again, because you just asked how I see it, yes, and how I worked in computer science, yes. Back then in Berlin, the institute where I also did my doctorate, under the direction of Professor Hommelstand, was located in computer science and dealt with robotics and process data processing. At that time, there were two other universities that had robotics in computer science. So, explicitly. Possibly there were others, but these stood out. On the one hand, of course, Munich, TU Munich, yes, my predecessor here, and Karlsruhe, where robotics was located in computer science. So, there was no ambiguity at all. Yes, and that's how it developed. There were, let's say, various highs and lows of AI development, and in Germany, we always followed certain trends from America, which is not always smart. Yes, we see that again now. And we had quite capable computer science chairs, but they were actually phased out in the 90s. Yes, now it has been recognized that it has become an absolutely important basic technology, and these chairs are now being re-established, but it certainly would not have been bad if they had been continuously occupied, like this one. And we have also, let's say, grown from virtually zero to a very considerable size in the last 25 years. Yes, that also creates demand. Let us first go into your career in detail. You were head of the Neurorobotics Project sub-project in the Human Brain Project for 10 years, one of the EU's most ambitious research projects with a budget of over one billion euros. Your platform connected brain models with robot bodies in closed-loop systems. For everyone who has never heard of this, what exactly is Neurorobotics and why is it so important? Neurorobotics attempts to control robots based on biologically accurate models. Yes, that means looking at how nature does it, how we do it, and what we can adopt from it. Yes, and we are talking about brain-inspired and brain-derived. Mhm. It is an old dream of humanity to, so to speak, copy flight from birds, but as it always turns out, what is a proof of existence, phenomenologically it is clear, we can fly on our planet, but the engineers have developed different airplanes that, let's say, do not deal with the flapping of a bird's wings. However, of course, in terms of aerodynamics, they are strongly based on what can be learned from birds. But as far as I know, there is no airplane that seriously flaps its wings, but rather other principles are used. Yes, so, the dream is to learn from nature and then transfer it. Neurorobotics is a very explicit term to operationalize this. And quite a bit has been achieved in this direction. Yes. Not only in the Human Brain Project, but also before. Mm. So, there is quite a bit that can be learned from the human brain or living beings, especially in the area of image processing, from the organization of the brain. However, one must self-critically note that the question of whether engineers have developed a certain solution before bio-inspired solutions become prevalent, usually turns out in favor of the engineers. We will see how it develops. So, within the framework of the Human Brain Project and even before, we have also developed robots or were part of groups that developed robots that are very similar to the human skeleton, yes, that is, they emulate the dynamics of human movement. And this is based on the hypothesis that intelligence can only develop in a body. Yes, there are certainly bodies with little intelligence, but a brain that just lies around doesn't really exist. That means that a prerequisite for the development of intelligence is first of all physical development. Yes, we also see this in the history of evolution. Yes, simple organisms that were initially just there are at the beginning, and this has developed gradually. So walking, the ability to move, first presumably in the sea, then on land, and so on. Yes, you know this whole history. Then, one can assume that a body is indeed the prerequisite for a certain type of intelligence to develop. For example, to survive. Mhm. And we humans do not have that many abilities, apart from our intelligence, to assert ourselves on this planet, which we have also managed to do. But in a direct fight with a wolf, we have little chance. That means we have to behave more intelligently, and our brain helps us with that. Yes, okay. That is a highly interesting statement or insight. So, when you say that intelligence, human-like intelligence, can only function in the human body, does that also mean, by consequence, that what we are currently seeing with LLMs, for example, would not be sufficient to achieve such general artificial intelligence or even superintelligence? I would say it needs the body, so it also needs the robot in that regard. What we are currently seeing is, of course, the transfer of knowledge that we as humans have incorporated into the cyberspace we have developed. That is, everything we have stored as texts or images, i.e., multimodal information, in cyberspace on our computers, on our computing machines. This can, of course, now be processed much better. So, we are now moving from purely syntactic, i.e., searching for a word in a text, to more semantically oriented information processing. I can ask a question, yes, with a contextual connection, and it will be answered meaningfully. So, this is what we have now, but with the systems we currently have, we cannot get out of this cyberspace cage. That means, if we want to connect with reality, yes, if we want to learn something in our space, then we will have no choice but to have a body. And one can consider different structures. Yes, for example, humanoids. Yes, humanoids walk around and have experiences like we humans do. For example, if I say, yes, or colors, yes, explain to me what green is, yes, because you naturally see a different green than I do. That will be relatively difficult, yes, simply put. There are, of course, all sorts of theories about this, but if you say, for example, I will throw a stick between your legs. Yes, how is the system supposed to know that if it has never had a stick thrown between its legs? Yes, it can only learn that, so to speak, projectively from what might be written in cyberspace, through such metaphors, but not experience it itself. So. And if one wants to do that, then there is no way around it than to place bodies in our real environment and give the computer systems the opportunity to construct their own world beyond what has already been fed into these systems by us humans. What is your opinion in this context on world models, which are becoming increasingly popular? Jan LeC has now very prominently raised over a billion euros in Europe from Paris and wants to work precisely on this, i.e., developing world models. There are some who say, yes, perhaps physical robots are not needed at all, but we will do all of this through virtual worlds. Well, if I manage to model our environment completely with the highest precision, including all physical phenomena, then that would indeed be feasible. Yes. Mm. But I would assume that it is much more complex, and we are still not sure if all the phenomena that truly guide us, that guide us in our individual development, yes, and in tribal development anyway, if they are all captured. I would have great doubts about that, and I also think it is more economical if these world models, which can of course be generated from the data that can be obtained from these humanoids, are actually captured with real systems and then one also knows how to process it sensibly. So, I consider that to be much more practical and much more logical than trying to construct world models through this detour. Of course, but world models in themselves are of course crucially important, and this is nothing new, by the way. Yes, so, back in the 70s, there was a movement in AI called naive physics. Yes, and it was precisely about the question, if I let go of an object in the air, a ball, that it falls down. Yes, because these are the simplest things that a robot system or an AI system should perhaps know. Yes, so in that sense, it's nothing new. And in Bielefeld, during the special research area that caused a sensation at the time, "artificial communicators," we investigated how robots could assemble a construction kit airplane. And then the question is, how can you get the robot to assemble it? For that, of course, it needs to know everything about, for example, torques for screws. Yes, if you have a system that you work with practically, that's no problem. Yes, that is, you say, let's assemble the airplane from these wooden elements, and then you notice where the problem lies. If you say, I'll do this purely virtually, then you first have to consider whether this is even a task that will ever occur, and what is it like then? You get from one thing to a thousand. I don't think that's a very successful path. Also, by the way, not from the experience with the Human Brain Project, because we did exactly that in our Neurorobotics platform, yes, we actually took physics models and trained robots with them, and it works. Yes, we have shown all of that, yes, others have shown that too. Yes. But it is very, very tedious. If you are interested in artificial intelligence, humanoid robotics, and the technologies of the future, then make sure you have subscribed to this channel. Many viewers watch the videos, benefit from them, learn a lot, but forget that they have not subscribed to the channel and therefore do not get these videos played to them every time. Therefore, do yourself a favor, check it out, and now have fun with the rest of the video on the Neurorobotic Project. Now, specifically, what were the most important insights or consequences from the project that you were able to take away? So, let's just take, if one says one takes humans or biology as an example to draw conclusions. Neural networks are an example that comes to mind. One is clearly oriented towards the human brain. What were the most important insights from the project from your perspective? Well, it was a very long-lasting project, and there are, so to speak, different phases. Yes, we first learned how to virtualize an environment or, rather, a structure for training robots. Yes, that is, how to actually teach it to the computer, yes, and then perform certain learning processes, and what can be learned from that. This is, by the way, a point that is currently very topical again. Yes, if you look, for example, at how humanoids are now being taught to grasp, yes, you see a lot of this, especially in China, but we also have our AI factory data collection farms at the TU, where people present objects to robots all day long, and they then collect data with their hands and their tactile sensors in their fingertips, because there is so far little data on how a robot hand feels grasping. Mm. Yes. But in order to have large foundation models that are practically useful for grasping, for grasp planning, for executing grasps, it is a highly complex process, which we humans, of course, are not even aware of. Yes, but look how long a child takes to grasp properly. It is a very complex process. It is not even consciously perceived. If you want to teach this to a humanoid robot, it is very complex. Yes. And one has to collect this data. And we tried to virtualize this back then, in the sense you just mentioned. But it is very difficult. The second point, and it is not certain which is easier, to do it in reality or to virtualize it. Google had an experiment at the time where they let 14 or how many robots grasp day and night. It was observed by camera with simple two-finger grippers. Until these robots, which also informed each other, were actually able to grasp simple objects that were in a box in front of them. Yes, that was also the starting point for us to say, let's do this virtually. As I said, it worked, but it is very, very complex, and the result is probably no better than what you get in reality. It's clear, because reality is the patchmark. Yes. Yes. So, the further things are, of course, there has been a lot of architectural development in the Human Brain Project. Yes, we had, there were, so to speak, two essential pillars. One was the entire neurosciences, i.e., decoding the brain. What is actually happening there? Yes, in the classical way, and on the other hand, the question of transfer to technical systems. Neurorobotics was only one sub-area. There was also the sub-area of neuromorphic systems. Yes, and essentially two systems were driven forward. One system from Heidelberg from physics, and one system from Manchester, the Spinnaker system, which is still being developed today, which is essentially a purely digital system for the implementation of neuromorphic computing. Mm. Yes. That is, inspired by our brain, to have algorithms and hardware that come as close as possible to what we suspect happens in our brain, or let's say, transfer a small part of what we know from the biological brain to computer systems. When we talk about humanoid robots and physical AI in general today, most people quickly think of what we now see on television through Unitree, for example, or the statements of many tech CEOs who say that from 2026 onwards is the year of great upheaval, that humanoid robots will really enter industry, perhaps even the first households. How do you perceive the current state of humanoid robotics? Where are we right now, at this moment? So, perhaps I can say something about the development of AI first, and then we will move on to the development of humanoid robotics. Yes, so, AI is ultimately a result of the development of microelectronics. Yes, of course, theory, of course, one can think a lot about it. The founding fathers also did. Yes. But the implementation, the performance we are seeing now, can only be achieved since microelectronics has received such a performance boost. There are, so to speak, according to my assessment, four essential basic technologies that drive AI forward. First, of course, processor technology, i.e., microelectronics, memory technology. Yes, we can store gigantic amounts of data today. Yes, when I started, you mentioned it earlier, we had hard drives of about 10 megabytes. Yes, 10 megabytes. Then communication technology, we talked about that at the very beginning, yes, which has also become orders of magnitude more powerful, and above all, of course, computer science, i.e., software engineering. We can handle systems of such complexity today, which was completely unthinkable before. So, and these four are developing exponentially, yes, and are, so to speak, the pillars on which practical AI rests. And therefore, we can also assume that AI will also develop exponentially. Yes, my colleague Brauer once said that for him, AI stands for coming informatics, and everything that was still in the realm of fantasy 50 years ago, we have today. And AI is essentially all around us. Yes, even this camera we have here, of course, has a lot of algorithms in it that would have been AI 50 years ago. Yes, so that is, so to speak, AI is developing exponentially, and therefore, what is not possible today will not be possible in 10 or 20 years, but probably already in half a year. Yes, so we humans, perhaps also as a brief remark, underestimate ourselves because we cannot deal with exponential development, but always linearize everything, we overestimate ourselves in the short term, yes, but we underestimate ourselves in the long term. That is very important when we, so to speak, you are asking about development, when we focus on bodies, i.e., the actual robot, then one must see that mechanics, of course, develops much more slowly than microelectronics, where we still see Moore's Law, also exponential development, even at the chip level. Yes, we do not have that in the field of mechatronics, of course. Nevertheless, we see that mechanics is increasingly being replaced by fast control loops, by software. Above all, by very powerful sensor technology, which makes it possible today. Yes, for example, acceleration sensors, which used to be large gyroscopes and so on. One could never have built a humanoid robot with those, but what we have in acceleration sensors today is tiny, highly precise, and very cheap. So, sensor technology is a very essential driver, and there we also have the basis, the basis also in microelectronics. That means, we will also see an acceleration here, we have seen an acceleration, but it is progressing somewhat slower than software technology and thus also AI or computing technology. So. It is now indeed the case that we have, as you see in the press and on television, very powerful humanoids of different forms. Yes, this is also an interesting development, by the way. Yes, previously, people were happy if they could get such a thing to move on a gallows and walk a little. Yes, by the way, the development was significantly driven by Honda at the time. Yes, until this Honda Asimo in 2006, there had been attempts to make walking machines since 1986. MIT Lab was the other development from which Boston Dynamics emerged. Yes, there were these beings, and then there were other universities in Japan. At TU Munich, there were also such walking machines, walking robots, also very successful. That means, it is a convergence, because one first started with walking, analogously, if you will, like in nature. Yes, first walking and being able to move away, unlike a plant. Plants do not need a brain because they are stationary. Yes, that means, one has made progress in mechatronics, but it is progressing significantly slower. We are now at a point where it is definitely worthwhile, because these systems can run stably, to connect them with AI, i.e., with the actual brain. By the way, there was already the so-called Remote Brain Oper Project or something similar in 1987 from the Japanese, where they said, we have a large computer here, we will connect it wirelessly to the body. Just for historical accuracy, and as I said, Shaky was also a system that was controlled by a central computer via radio back in the 60s, late 60s. So, what we are seeing is a development towards different humanoids. Yes, the Chinese, one must admit, are currently leading in these robots. Yes. We have systems that are more or less built for show purposes, especially for playing football. Yes, we have others like the Unitree, which is very good at walking, running fast, and then also doing somersaults in boxing matches, etc. This is a development where people say, yes, what is special about it? You can say, look at the state of the art from 10 years ago, and then you will see what is special about it. So, it is an enormous engineering achievement. And we have systems that are then more intended for industrial use and for all these purposes, so we have this differentiation, yes, which already shows that the systems are relatively robust now, in contrast to 5 to 10 years ago. Yes, we are certainly seeing this development. And we are indeed seeing application areas that are now becoming interesting and are being served by new providers. In the factory, I would say we are slowly getting there. Yes, but it is missing, and the question is, what kind of tasks are suitable for classic humanoids? The development seems to me to be rather in the direction of saying, okay, we now have the possibility to walk, we have the possibility to operate manually. Yes. We will see where this is used. The question will also be, do we really need a full humanoid, or is a torso mounted on a mobile robot on wheels perhaps sufficient? Because typically, one of the big advantages of walking is that we can overcome height differences, climb stairs, and so on. We don't really need that in a standard factory, if we are mainly performing transport tasks, for example. Yes, then the robot naturally has to be quite strong in a humanoid. Yes, but for material transport tasks, we don't need high manual dexterity. Yes, and it is indeed manual dexterity that is still lacking. Yes. So, if we now
But conversely, to say, if we are in car manufacturing, yes, and in car manufacturing, we have different trades, yes, then we have seen for a long time that automation in body construction is practically 100%. You see Kuka arms everywhere or these six-axis classic robots from other manufacturers, which have reached enormous maturity. These have nothing to do with humanoids, of course. If you then go into assembly, that is, where the interior is equipped, so to speak, you see a still relatively low level of automation. One can imagine, of course, that humanoids, because they can go into the body, so to speak, and perhaps also, if they have the necessary dexterity, can perform certain tasks. There would, of course, still be a lot of potential. How far we can achieve this in a short time, let's leave that aside. Yes, there were many points in there that we can delve deeper into. I therefore have an overarching question that might put it in perspective again, because you also said that we are seeing an exponential acceleration, which is not decreasing for the time being. Most people still underestimate how quickly everything is developing now. And you have just mentioned the four major areas: microelectronics, memory technology, communication technology, and computer science. And the question that arises is, what is ultimately the bottleneck, right? Some say, ultimately everything converges on energy costs and ultimately fails only due to energy. Everything else is already almost solved. If we only look at humanoids, you have mentioned dexterity, for example, which is still a bottleneck in this area. What do you see as the biggest bottleneck that would need to be solved or addressed now to achieve the acceleration? Yes, well, yes, first of all, one must of course, perhaps briefly back to your initial question, whether this can really continue with energy, I dare to doubt it. We have often had the situation that, due to the switching technology alone, computer systems became more and more energy-hungry. But for thermal reasons alone, this could not be continued. Intelligent solutions were devised. Yes, if you compare your PC, you are too young, but if you compare your PC from 25 years ago, what performance it consumed and how much touch frequency it had, and with today, then today it is only a fraction. So today you have very clear systems. 32-bit, they last forever. So, that has been achieved. Yes. How it will continue with these large data centers, we will see. Yes, it is quite possible that people will have clever ideas that will make it superfluous again to a certain extent. That is not foreseeable, completely clear. Yes, I would also not advise anyone who wants to invest there not to do so now, but one simply doesn't know. There have been so many methodological breakthroughs, yes, that it is quite possible that we would see a change there too. So. If we now think about humanoids or robots in general, then we can say, either we take a remote brain approach, that is, we couple the system, the robot, to a large data center, or we say, it is perhaps not so bad if it is autonomous. That is, if the intelligence is in this body, so to speak, then we cannot supply it with gigawatts of energy, of course. We must then ensure that the intelligence it needs to perform its task and secure its existence in the household or in the open field, so to speak, remains manageable. And for that, new solutions are certainly needed. That is completely clear. But one also needs, I mean, one must not demand in the first step that the humanoid somehow performs an extensive literature search on the internet while it is running at the same time. That is also clear. That is, the requirements are completely different and accordingly the solutions will also be different. And one goal is certainly the reduction of energy consumption. Mm. But you spoke of the overarching question, and I tell you, it will still happen. Yes, my mantra is: all intellectual achievements of humans that can be transferred to computers will also be transferred to computers. Now, of course, the question is, which ones can be transferred? Yes, we will see a lot there. But that should not frighten us, because, of course, unless one were to build a system that corresponds to humans again, but that is not necessary, because humans are humans and a purely biological, a purely biological replica of humans, i.e., keyword neurobotics, would not only be for ethical reasons, it would simply be completely pointless. Yes, it would be completely pointless. So, long story short. It's simply that we will see specialization here, we will see breakthroughs here energetically as well. And it is also, incidentally, the question, perhaps even more broadly, where robots will actually find long-term use. We will see many new areas that we haven't thought of, but conversely, I dare to doubt that everyone will have a heavy humanoid or several heavy humanoids at home in the future, simply because it is very complex and it will probably, as I said, it could also turn out differently, but it will probably rather be the case that we will have specialized cognitive devices, i.e., devices that perform certain tasks very well, but others not. For example, intelligent vacuum cleaners, which many people already have at home. That is a very long development, but one can also imagine, what do I know, a washing machine with an arm or such things. Yes, that is probably more economical than a humanoid that can do everything. But technically it is quite conceivable and is, of course, then a question of price. If you are employed and watch these videos, you already know that by 2030 at least 57% of all working hours, i.e., over half of all jobs, will be fully automated. At least that's what the current McKinsey study says. And according to Enhropic, this affects not only routine jobs but also particularly well-paid knowledge workers from the middle and upper-class segments. You already know that at Everlast AI, we have successfully supported over 2100 companies in AI integration over the past few years, from medium-sized companies to true world market leaders. We repeatedly see that people in companies who take responsibility for AI projects, proactively educate themselves and build real skills, are making themselves indispensable, while positions like AI Manager or Chief AI Officer are still hardly filled properly in any company. But the problem is that there are no proper further training courses that teach you AI knowledge from real practice. And since we have received so many legitimate questions from the community in this area over the past few years, we have decided to build kidernen.de, the number 1 AI further training platform for the German-speaking market, and to pass on our knowledge to everyone who has understood that AI skills are at least as important as being able to use the internet. So, if you want to build the essential skills of the future now, not only to secure your future but to make a real career, then just take a look at kidernen.de. From now on, you will find various courses there that will provide you with a shortcut to the skills that really matter, in addition to all the YouTube videos here. You will also find the link in the video description, and now back to the video. You said at the beginning that the body is an important prerequisite for human-like intelligence and that ultimately every cognitive achievement, as you just said, of a human can also be transferred to a computer. And intellectual achievement, yes, we will see. Sensor. Mm. And regarding this, I have a question because many would argue that humans are characterized by their emotions, for example. And you spoke at the beginning of Marvin Minsky, who wrote a book, "The Emotion Machine," and in it he essentially argues that emotions are also essentially chemical processes that can be replicated. And the question is, first of all, how do you see this? So, emotions, how important are they, in doubt, for such an AGI or superintelligence? Yes, well, we still know far too little about the human brain and its function, even after the Union Brain Project. So one can argue about how many percent has been understood, but I think we have only understood a small part. That's the first thing. Above all, the processes that lie beneath it. The other question, of course, is why should we want to replicate it? Yes, emotions in us have various purposes. Fear, for example, should we build a humanoid that is fearful? To briefly interject here, there are many who say that these functional emotions are needed so that the robot, in doubt, notices when it bumps into something, ah, I should leave that alone in doubt. So, one might have to differentiate between functional emotions. That can be done. Yes, but as I said, the way to get there is still relatively long, and whether it is desirable, I put a question mark there. I can imagine that one could have different classes of emotions, so affective computing is also a relatively old field. Yes. I think it would be useful to program emotions into it, or at least emulate or simulate them, when it comes to increasing the dialogue efficiency between the robot and us humans. Yes, I could imagine that, for example, the robot would also say, man, I've explained this three times now. Why are you so slow to understand? Or vice versa. What is certainly more important is that the robot can interpret human emotions better. Yes. Yes. And now, of course, the argument always comes up, yes, the AI can drive young people to suicide, which is of course a big problem. Yes, that's why it's also a question of whether one should really prioritize that. Yes, I think it would be more useful to first focus on the hard problems that we have already discussed. Yes, so, how does a machine learn to grasp? How do I improve dexterity and so on so that it has practical use. The rest can certainly be theorized for a long time, and it can be replicated to a certain extent. But that is also nothing new, if I may remark here, because Josef Weizenbaum already showed in 1966 with his program ELIZA how, with little effort, one can drive people to an emotional state to such an extent that they believe the computer is their most intimate confidant. And he also describes in his book that his secretary no longer allowed him to watch when she was writing to the computer. It was via teletype. Yes. Because she was too intimate with the computer, and it was a super simple program, anyone can replicate it in the browser today. ELIZA, so it is really very interesting how easily we can be made to believe that we have an intelligent counterpart. What, in your opinion, is the right measure or the right benchmark from which we can say, now we are really dealing with human-like intelligence, so-called artificial general intelligence? The Turing Test has long been considered a benchmark. It was solved in 2024 by GPT models. Yes, you know, maybe you are a robot. Who knows? If you were a humanoid now, I would be convinced. Yes, but I don't think you see it, you know me, I'm not touching you, but we can find that out relatively easily. Yes, I don't consider this discussion to be very fruitful. Yes, because we have so many problems to solve first, and we have very useful applications. Yes, so that I believe we should continue on the path of constructing useful helpers, yes, and not always paint the devil on the wall that AI will wipe us out or that, well, I myself consider that, let's say, one should never say never, it can happen, yes, but at the moment it is not foreseeable. Yes, we also see so far, yes, a holy grail, because you are asking about it, is for example the car, that is, the self-induced occurrence of emergent behavior. That is, the machine does something that the programmer did not intend, and out of its own initiative. Yes, here too, one can say, it can also be a kind of, how should I say, simulated behavior, yes, that the machine now simulates creativity. Then people are always impressed. But it is not actually something new, but it has perhaps only looked in areas where the human who is judging it has not looked before. Yes. Yes, it is difficult, but, but as I said, I believe emergent behavior is indeed such a point where one could say, okay, now it's getting interesting. Yes, I absolutely agree with you there, we must of course remain pragmatic and, above all, look at how we can adapt correctly in Germany and what we should focus on specifically. Let's talk about your Providentia project again. You and your team have recreated a digital twin of the A9 motorway. What exactly was the goal of this project? A part of the A9 motorway, not the whole thing, it's relatively long, but the hypothesis is that if we want to digitize traffic and also digitize the motorway, then we must first be precisely informed about the current state. Yes, that's why we said we would build sensor infrastructure on the motorway, observe traffic and all objects in traffic, i.e., pedestrians, all types of vehicles, and then build a digital twin. That is, we have an image of the real thing. Yes, we talked about the importance of transferring the real world to the computer in the case of humanoids. Now, because of or the physical AI, because of the acquisition of environmental states. Here, it's more about having a basis for all sorts of things that can then be done on the computer. For example, to increase safety. If I have a large-scale capture of traffic events, down to the vehicles or pedestrians or road users, then I can also warn the driver, for example. Watch out, someone is coming, or the car warns, a significantly improved emergency braking function is implemented. Then I can ensure greater comfort. That is, I can, for example, look 10 km ahead to see what the traffic situation is like there and don't have to just look at Google Maps or my navigation system, is it yellow, orange or red, but I can inform myself. And in that sense, it is also a contribution to the autonomy of the citizen, if one has more information about the state of the motorway, and finally the question of optimization. Yes, one wants to reduce roads and increase throughput. So, can I then ensure that smaller distances become possible in cooperation between the car's sensors and the infrastructure's sensors? These things will all become possible. Of course, for that, you also need a proper mobile phone connection, which works with very low latency, but we are in the era of 5G, 6G is on the horizon, one can imagine all sorts of things, and then it is also about, for example, in the event of a landslide, automatically triggering alarm chains, because these systems are capable of assessing the situation on the road. So, do I need to call the fire department and such things? Yes, which normally takes a relatively long time. If I have systems that can accurately capture the state, then I can say within seconds, I need the fire department, ambulance, and so on. Yes, that means I have a whole range of functions, functionalities, that I can then implement if I have a proper state capture. And that is exactly what we have shown with the Providenzia project. That is, we have captured a short section of the A9. We have equipped a country road section with sensors, and also a large intersection in Gchim Hochbrück, so that we can capture all traffic situations or all, let's say, scenarios. And we make it available to everyone who wants it. The problem at this point, of course, is that there is currently no real buyer for all these technologies, because the automotive industry builds cars. It would actually be a state task, but that is always difficult. In this context, it is certainly still exciting to talk about your CAS project, so that we can cover everything comprehensively in this context. You are developing a central software architecture for software-defined vehicles designed for vehicle generations from 2033 onwards. You say that understanding cars as software platforms is simply necessary to remain competitive in the automotive market. What exactly does that mean in concrete terms? Understanding cars as software platforms. Yes, Elon Musk actually showed us that 25 years ago. That is, the computer on wheels. And it is indeed the case that we are facing a situation where people generally, in China perhaps the most, but certainly in the USA and here as well, increasingly expect the car itself to be their digital platform. Yes. And therefore, it must also be able to offer all the comfort we are used to at home, even while driving or when stuck in traffic. And it is actually very desirable, in our opinion, that the car can actually drive itself. Yes, because then one can actually do various things. Why should I, if I want to drive sometimes, the classic example is an old tour, sure, then it's nice to be behind the wheel yourself, but when I drive to the office in the morning, why should I always be there myself? It's a safety risk, it costs effort. In that sense, it would actually be better if the vehicle could at least cover certain routes by itself. Yes, that is, the software penetration is the decisive factor, and the mechanics, if you will, and the entire drive technology are actually only secondary. You see this development in China, you also see this development in the USA for a long time. Yes, I mean, we are very proud, and rightly so, of the high standard of vehicle technology and our chassis and so on, but one must unfortunately concede, people don't really find that relevant anymore today. Apart from some specialists, of course, the trend is simply towards saying, what can the vehicle do digitally, is there a reasonably sensible computer installed, can I see all my functionalities implemented well here, yes, the user interface and so on. These are the decisive points, and there we are no longer the ones, we must admit that self-critically. Although we now have the opportunity, surgically, with the help of software engineering and advanced computer science, to regain a leading role, simply because we might even become the fastest. And CAS, Central Car Server, is an approach to install a central computer in the vehicle with corresponding gateways to the peripheral units, i.e., sensors, actuators, which can, however, be very well expanded. So, it is indeed expandable, which has the most modern functions for processing streaming data, i.e., video, network, and so on. And then also has a way to integrate functionalities very quickly. On the one hand. But what was much more important to us was actually, because these are developments that are happening anyway. What was more important to us at our institute or as part of this project was that we say, we can also create the software in a very short time with the help of AI techniques. That is, my mantra is, we will actually no longer need humans for the creation of a complex software system, as is the case in a car, in the medium term. Except for the creation of the specification. We have to say what the thing should do. We also have to know the regulations. So, what does the road traffic regulations say, what does the UN prescribe, and so on and so forth, and we also need a description of the target platform. So, what does the vehicle actually look like? What technology is installed there, and so on. Yes, so apart from these three things, that is of course a human task. We want to say what the thing can do, but the rest can actually be done by the computer. What is the concrete task of the software in this case? So, is it about enabling autonomous driving as such, or is it more about? Well, first of all, first of all, we are rather neutral regarding autonomous driving. So, one can essentially implement all forms of automation on this platform. Yes, so level 1, 2, 3, 4, perhaps even 5, i.e., fully automated, fully autonomous, or fully automatic. And it doesn't matter here. It's more about how I get from my description of the desired behavior to executable code, and there we are of the opinion that this must be done without humans, not immediately, but we must move in this direction. Yes, and I mean, you know all the statements about programming ability or programming by large language models in this case. That is a rapid process, and it would certainly be nice if the automotive industry would also understand this and act accordingly. How do you assess the current state of autonomous vehicles in this context? So, if we now look at Waymo, or you mentioned Elon Musk and Tesla earlier, who are also pursuing various approaches, some with lidar, radar, Musk rather only with computer vision. What do you see as currently leading, or what do you think will ultimately prevail? Well, these are different approaches, aren't they? One can say, a vehicle must be completely autonomous. That is, it must be able to navigate in any environment like a human driver. Yes, for example, even in the desert. Then you really need very high-performance sensors and very high-performance computing power in it. One can also say, we'll do a specific city, and we'll map it, and then we'll have a high degree of safety there. It might require less sensor technology. You see that the successful companies are already operating with a lot of sensor technology. Yes, and Elon Musk's insistence that he doesn't want to use radar is not necessarily successful so far. Of course, that can change. Of course, one can always argue, humans don't have radar either, so why do I need it? But it turns out that a car, which is a body like a car, is still something different than when a human sits in this machine. Yes, that is still a big difference. So, I would say it's a continuing open race, which, however, requires large investments, which the Americans are making, i.e., they are raising them. Waymo is not profitable, it's a bet on the future. Yes, while we in Germany, keyword new S-Class and so on, are rather very reserved, to put it politely. Yes. It would certainly also be a viable path to say, we'll make things so cheap that it's like today's navigation system as standard in every car, that the self-driving car is actually the standard of the future, and everyone buys this accessory. But with that, we remain competitive. One could also hold that opinion. Yes. You have now spoken about the automotive industry in Germany. We have spoken about Elon Musk and Tesla, who is now also saying that Tesla will probably rather become a humanoid robot company in the future with Optimus. And I would be interested to know how you currently assess the state of humanoid robotics, especially in Germany, compared to international competition. We have spoken a lot about China, about the USA now. Where do we stand now? You are doing impressive research here at the Technical University of Munich, of course. What is often criticized is that people say, yes, we are doing great research in Germany, but we are not really bringing it to market. And the money is ultimately earned in America. How do you see the current position? The money is primarily made available in America. That is a very important point, and we are not moving in Europe, and I consider that a huge problem. What you can see again and again is that companies are simply bought up when they are successful, and therefore a growing potential is building up in the USA, while here, well, we invent at least as much as the USA, but the marketing then actually takes place there, you are completely right about that. That is something that should be urgently reversed. The question about the status is, I would say, we have been at the forefront of humanoid robotics for a long time. In terms of research, we are certainly still at the forefront today. This applies to AI as a whole, incidentally. It's not that we couldn't build such systems, only that the capital for implementation is missing, and we also lack courage. One must say that clearly. This is also an old story, incidentally. We built the first fully transistorized computers in Europe at Siemens back in the 60s and 50s. We even invented the computer as we use it today. But we didn't make anything of it. In AI, we now have a moment where we can actually say, if not now, when? Yes, because, as has often been said, but I can say it again, we have all the data, we have the industry, which the Americans only have very, how should I say, sporadically. We have the data, we have the industry, we want to preserve the industry as much as possible. Yes, so let's use everything that is available. Let's use, for example, American technicians, American providers, on an interim basis, but we should of course try to get this development back into our own hands. For that, we need capital. As I said, we still have excellent research. We are certainly at the forefront there, one can say that. Yes, we must of course also do the right thing to keep it that way. But we are not bringing it to market, keyword autonomous driving, in the sense of the word. But that is unfortunately a decade-old problem. I don't have the perfect solution for that either, except perhaps to raise people's awareness that they should not reject technology, but actively welcome it. Yes. Yes, if I may say so, regarding the shortage of skilled workers and young talent, if there is no one left to operate the machines, except perhaps robots, if there is no one left to sit in the office, except perhaps, let's say, artificial intelligence, then who will earn their pension? Yes. That is a very important point. I also spoke with Alexander König a few days ago, and he said that this is a central question that arises: how do we achieve this translation, i.e., that people are open to technology and welcome the development of human robotics, not least here in Germany. Unfortunately, I often observe that very negatively, if you look at the comment sections sometimes, it's immediately said, I would never let a robot care for me, stay away from me. With this whole. Yes, then the question is, who else will care for you? Yes. So. Then there will be no one left. Yes. So, there will be no alternative. That means that this technology should be massively promoted, quite apart from the fact that the entire planet has this problem. That means it is also potentially a huge export market for us, because if we no longer export as many cars, which is one of the essential pillars of our prosperity, then the humanoid robot would certainly be an adequate replacement. And that is also what Elon Musk dreams of. To what extent his solution is equivalent to the Chinese one, let's leave that aside. Yes. The speed of development there is enormous, but we have many advantages in Germany, incidentally, in principle or potentially even over China. Yes, simply because we have an excellent engineering environment. We still have excellent engineering. We have the entire mechanics, so to speak, in our portfolio, mechatronics, we are certainly, we are still world-class there, to transfer that into new products, for example, humanoid robots, should not be a question at all. Yes. Yes. And in that sense, it is indeed a shame when such doomsday fantasies go around. But here too, one can learn from history, because we had the same scenario when robotics was introduced in the automotive industry in the 70s. Yes. Namely, it was always said that there would be no jobs then and so on.
Continue. Yes, all people will become unemployed. That has by no means happened, but quite the opposite has happened. Yes, the countries that have automated the most, yes, that have the highest robot density, also have the lowest unemployment, or rather, have been able to preserve these sectors at all. Yes, so the one with the highest robot density has, I believe, between Singapore, I don't know the exact figures now, but Korea has always been very, so South Korea has always been very far ahead and, yes, well, the South Korean automotive industry is extremely successful. Yes, we also have a relatively high robot density, but we have also fallen behind, and we should urgently reverse that. Yes, what most people underestimate is what you can achieve yourself through your own entrepreneurial strength and get out of complaining and tackle things yourself, and you have founded a company yourself again, namely the company One Alpha. Tell us about it, what exactly does One Alpha do? Yes, we, well, we simply looked at what can be done with AI. Uh, let's say in the intellectual environment. Uh, and where are the biggest complaints? And we hear every day bureaucracy, bureaucracy, bureaucracy hinders us the most. So, it's relatively obvious to say, well, then let's dedicate ourselves to the topic. That means, with this company, we want to, on the one hand, provide a tool to make bureaucracy manageable, not to reduce it, because experience shows that never happens, but to make it manageable. Yes, so that is a concrete goal, so to speak. Yes, uh, we make this bureaucratic superstructure simply manageable by, so to speak, pushing everything onto the computer. And this is possible with this semantic information processing that generative AI and the next AI developments provide us with. Certainly, one must know how, and above all, one must also integrate it into the gears. That means, our second, uh, our second intention is, uh, especially for the medium-sized businesses, which sometimes don't really know what to do with this AI, it's great and so on, right? How can I use it in my company, so to speak, to ensure this transformation, or to get a concept of what it actually is? Yes, we go around a lot and we see that people often say, I can't imagine what you're offering me. Yes, and then we do a few demonstrators and show, for example, we have various AI employees on offer here. Yes, so, uh, from my perspective, this is also an important thing that the company offers, that you say, look, we have certain activities in your company, there might be no successor anymore, or the competence might not even exist anymore. Wouldn't you like to try it with an AI employee? Yes, so. So this, this is very important, that you say, we have these analogies here, yes, so that people understand, this is not something that should scare you, but we are now helping you with the automation of intellectual activities, yes, not to make people unemployed, but simply to assist them first. And then, as is the natural flow, if you will, is to say, well, we're doing this similarly to autonomous driving. Yes, there are different stages. First, assistance systems. That means, if someone has a lot to do with regulations, they get an assistant who says: "Watch out, if you want to do this now, yes, please consider the regulation, and if you want, I'll also look it up for you." So that one then enters into a sensible dialogue with such a system, which is then very focused, yes, and doesn't wander off. And so one can continue like that, so assistants, yes, then perhaps partial automation, and then finally up to, let's say, up to the point where certain activities are completely transferred to the machine. That's nothing new either, because we have that, if you, for example, at the beginning we talked about telecommunications technology. Yes, we used to have manual exchanges, yes, 100 years ago manual exchanges with many young ladies from the office, as they were beautifully called back then, yes, that was replaced by electromechanics and later by special computer exchange centers, and meanwhile, that has also disappeared. It has all, so to speak, merged into the internet. And exactly the same thing we will see here, so with our company. Uh, we want to make it manageable on the one hand, but also help companies to master this transformation. And the next step that we are aiming for is that we also want to automate the actual, the actual work that then has to be done on-site, so to speak. So I install an AI employee for you, that this is also automated, so if it's Royal Automation, second stage. Yes, and that will also be very exciting, and we assume that it will also lead to a significant acceleration for company building or company transformation. Yes, absolutely. So very important, I can only agree with that. And now I would be interested again, because you are active in this field. I have the perception that in the last one or two years we had even more of this discussion. We have this again, an AI bubble, an AI winter will soon come. In doubt, I don't need to deal with it, it will all be gone again soon. Uh, you said at the beginning, we see exponential acceleration, in doubt, the developments that one would think take 10 years, we achieve them in half a year. And with that, the question is already somewhat answered, how do you see it? What do you think is the reason why many people think or argue that the bubble will burst soon and I don't need to deal with it? Well, because that's a very comfortable position, isn't it? The only problem is that the old position leads to nothing being left soon, to put it that way. Yes, we see, we see over the decades that there have always been these winters and these trends and these hypes. Yes, absolutely correct. Uh, however, one must see one thing. Uh, there is, so to speak, an underlying pattern, namely, in the end, only one always profited, and that was the USA. Mm. They always drove it forward. There were always pioneers who then initiated the next stage, so to speak, and thus also became profitable, or one can also say, keyword AI is coming computer science. One then used new techniques and went to market very quickly. That means, ultimately, there is no alternative at all. There will certainly be a correction in value, that's always the case. Yes, so the current hype, that's a very natural hype cycle that we always see. Yes. Uh, but that will not lead to a collapse. So I don't expect that we will now have a second New Economy bubble here, but it will recede somewhat. Yes, expectations will naturally be scaled back first, but that's always the case, and if you know that, you can certainly prepare for it. For us, it has the great advantage that we might have a chance with our German speed. Yes, so, uh, well, the examples are well known. I don't need to list them again. Yes, so large construction projects and so on. Yes, we should really not afford that here, but we simply have to stay on the ball and then, so to speak, tackle things from behind. I can emphasize again that regarding humanoids, we have all the potential. Also regarding our industry, we have all the potential. It's about automation and autonomization. Yes, and, uh, and it's our only chance, and we should really understand that now and act accordingly, yes, and not retreat into our German complacency again. That won't work anyway. Yes, absolutely. So I can only agree with that, and most people underestimate how valuable that is or how much can be done by simply liking this video, sharing it. Yes, therefore, give the video a heart and ensure that we support the whole thing. And I have one last question again, because we also have many viewers who are employees, perhaps students, or who ask themselves how they can benefit from all these developments. Perhaps more and more people are also actually afraid of job loss or that their current job will no longer be relevant in the future. What recommendations do you have for people out there? What should one focus on? Which fields, professional branches are perhaps particularly exciting from your perspective? For all those who say, I also want to enter the field of AI and robotics. You also accompany many students here. Uh, what do you see? What are the most exciting fields and where? Well, I mean, it's a bit difficult from my position to give young people direct advice, like, go and become self-employed, when you yourself have had a university career, although one must also say, for fairness, that it wasn't as attractive in terms of income as founding a company. That should also be clear. Yes, everything has its pros and cons. Uh, I would, I would tell young people, I always tell them, check whether founding your own company might not be what you could enjoy. If you have a minimal entrepreneurial spirit, then now is exactly the time, because this is a truly golden moment. Yes, like perhaps Bill Gates and Paul Allen said, wow, this microprocessor, this is it. Yes, and that's how it turned out, or, well, also Steve Jobs and S. Wozniak, that was basically the same in the mid-70s. Yes, so I would say, see this chance and check it out. If that's not your thing, then look around the world. Yes, go abroad, but perhaps after your studies or in the final phase of your studies. First, get a proper degree here. If people have just graduated from high school or have a school leaving certificate, then I would tell them, or I do tell them, do what interests you. It sounds trivial, but one should not, I think, one should not make life-changing decisions based on the current job market, but look at what you enjoy. Yes, and then study it or do an apprenticeship. Yes, it will probably never be like that again. It was never really like that either, that if you study, you have a higher income than someone who builds a craft business. Quite the contrary. Quite the contrary. So a solid, let's say, activity in our world. Yes, keyword, humanoids. I can well imagine that craft businesses will indeed use humanoid robots in the future, and the question is, how will they do that? A lot of know-how will be required for that. I mean, even if you look at a craft business today, yes, that, for example, does communication technology, it's not comparable at all to 30 years ago, yes, that they deal with routers and network technology in this way. Back then, you laid cables for a telephone. Yes, that's a completely different qualification level. That means, it's entering all areas anyway. Yes, and, uh, think about it, as I said, whether that might not be something that could fascinate you, because it will certainly be a big future. So do what you enjoy. Look around the world and check, uh, uh, where you feel comfortable. And then it will work out. It wasn't like that, yes, that in our time, yes, in our time, everything was so trivial, but there was also academic unemployment. Physicists, for example, you were told not to do that at all. Yes, but physicists are highly sought after. Yes, so I wouldn't give much credence to such predictions. I would simply say, just do it and with. I would also advise to go through the training phase quickly. Yes, because then you are still young and have all the chances. Yes, especially what you say, we are essentially still at the beginning of an industrial and technological revolution, and becoming self-employed and founding a company in this area is certainly not the worst decision. So Mr. Knoll, super interesting insights into your work. Thank you again for the invitation here and for taking the time. Thank you. And if you still have questions for Mr. Knoll, please write them in the comments. I'm sure we can talk again in the future. There will also be many updates from you in the coming years, I'm sure of that. We assume so. Yes. And, uh, and as I said, we are of course always willing to engage in further discussions. Yes. Uh, but please also consider that in fact, there is simply an incredible amount going on, which is positive. Yes, we certainly can't react to everything, but if we meet again not too long from now, keyword exponential growth, yes, it's always shorter, then gladly. Yes. Yes. So write it in the comments, we'll take it all on board. See you next time, thank you. Super. Yes, thank you. Until then, thank you very much.