Lift-off for Tech Interdependence? / DAVOS 2025

Summary

This panel discussion at the World Economic Forum in Davos focused on the convergence and interdependence of emerging technologies, particularly artificial intelligence (AI), and their potential impacts across various sectors. The panelists, representing diverse fields from technology to publishing, explored how the combination of AI with other technologies like bioengineering, quantum computing, and robotics is driving rapid innovation and disruption.

A key theme was the accelerating pace of technological advancement, with AI development outpacing even the expectations of its creators. The panelists highlighted how AI is being integrated into various industries, from automotive to healthcare, enabling new capabilities and transforming business models. They discussed the potential of AI to enhance productivity, creativity, and scientific discovery, with examples ranging from autonomous driving to drug development.

The discussion also touched on the challenges and considerations accompanying these advancements. These included the need for sustainable AI development, the importance of interdisciplinary collaboration, and the ethical implications of AI deployment. The panelists emphasized the critical role of integrating AI with existing systems and data to unlock its full potential.

Looking to the future, the participants speculated on potential developments in AI over the next decade, including more human-like capabilities and the ability to learn from experience. They stressed the importance of adaptability and continuous learning for businesses and individuals to keep pace with technological change. The panel concluded with advice for decision-makers, emphasizing the need to understand these rapidly evolving technologies, invest inclusively, and consider the broader ecosystem impacts when implementing AI and other emerging technologies.

Keypoints

Major discussion points:

– The convergence and interdependence of multiple emerging technologies like AI, synthetic biology, IoT, quantum computing, etc.

– The accelerating pace of technological progress and innovation, especially in AI

– The impact of these technologies on various industries and the need for businesses to adapt

– Challenges around sustainability, ethics, and responsible development of AI and other technologies

– The future of human-computer interaction and AI’s potential to transform interfaces

The overall purpose of the discussion was to explore how various emerging technologies are converging and accelerating each other’s development, and to consider the implications of this technological acceleration for businesses, industries, and society.

The tone of the discussion was generally optimistic and excited about technological progress, while also acknowledging challenges and risks. The panelists spoke enthusiastically about new capabilities and opportunities, but also raised concerns about sustainability, ethics, and the need for responsible development. The tone remained consistent throughout, balancing excitement with caution.

Speakers

– Jeremy Jurgens: Moderator

– Aidan Gomez: CEO at Cohere

– Magdalena Skipper: Editor-in-Chief, Nature Magazine

– Aiman Ezzat: CEO of Cap Gemini

– Cristiano Amon: President and CEO at Qualcomm

– Hiroaki Kitano: Executive Deputy President and Chief Technology Officer at Sony

Additional speakers:

– Amir Hussain: World Quant Foundry

– Ken Dillon: Occidental

Expanded Summary of Panel Discussion on Emerging Technologies

This panel discussion brought together experts from diverse fields to explore the convergence and interdependence of emerging technologies, with a particular focus on artificial intelligence (AI) and its potential impacts across various sectors. The panelists, representing technology companies, publishing, and research institutions, delved into how the combination of AI with other technologies is driving rapid innovation and disruption.

Convergence and Interdependence of Emerging Technologies

A central theme of the discussion was the increasing convergence and interdependence of multiple emerging technologies. Aiman Ezzat emphasised the importance of looking at technologies in combination rather than individually to understand their potential for driving fundamental changes. He specifically mentioned spatial computing, LiDAR technology, and bioengineering as key areas of innovation. This sentiment was echoed by other panelists, with Magdalena Skipper highlighting the intersection of disciplines like material science and biology, and Cristiano Amon noting that “everything is becoming connected, is becoming intelligent.”

The panelists agreed that this convergence is leading to transformative changes across industries and scientific research. Hiroaki Kitano framed the challenge in terms of understanding and controlling “very large dynamical nonlinear systems,” providing a unifying perspective on diverse technological efforts.

AI Advancements and Applications

The discussion highlighted significant advancements in AI, particularly in areas like reasoning models and edge computing. Aidan Gomez spoke about new reasoning models unlocking “a whole new tier of capability in the technology,” emphasising their potential to be distilled effectively for practical applications. Cristiano Amon discussed the ability to run large AI models on edge devices like phones and cars, illustrating the increasing ubiquity of AI in everyday technology.

The panelists explored various applications of AI, from enabling new interfaces and interactions with computers to achieving superhuman performance in simulations. Hiroaki Kitano provided a specific example of an AI agent outperforming human champions in the Gran Turismo racing game. However, he also stressed the need for AI to handle edge cases and long tail distributions effectively, highlighting an ongoing challenge in AI development.

Integration of AI in the Automotive Industry

Cristiano Amon provided significant details on the integration of AI in the automotive industry. He discussed how AI is being used to enhance driver assistance systems, improve safety features, and enable new levels of autonomy in vehicles. The convergence of AI with other technologies like 5G connectivity and advanced sensors is driving rapid innovation in this sector.

Impact on Industries and Research

The transformative impact of AI and other emerging technologies on industries and research processes was a key point of discussion. Aiman Ezzat noted that these technologies are restructuring and disrupting entire industries. Aidan Gomez highlighted the compression of the research-to-production pipeline in AI development, noting that many AI innovations are now deployed to market before formal publication.

This rapid pace of development is changing the dynamics between academic and private sector research, as pointed out by Magdalena Skipper and Hiroaki Kitano. They discussed the challenges this poses for traditional scientific publication processes and the need for new approaches to ensure the timely dissemination of research findings. The integration of AI systems with enterprise data and processes was identified as a crucial step for businesses to leverage these technologies effectively.

Challenges and Considerations for the Future

While the overall tone of the discussion was optimistic about technological progress, the panelists also acknowledged several challenges and considerations for the future. Magdalena Skipper emphasised the need for cost reduction and sustainability in AI deployment, highlighting the current unsustainable energy requirements for AI systems.

Hiroaki Kitano raised the importance of finding practical applications for advanced technologies like humanoid robots, suggesting a need to balance innovation with real-world utility. The potential impact of AI on the job market was also briefly touched upon, with Aidan Gomez noting the potential for significant changes in employment dynamics.

Future Predictions and Advice

Looking to the future, the panelists speculated on potential developments in AI over the next decade. Their predictions included more human-like capabilities, the ability to learn from experience after deployment, and AI systems that can generate their own goals and curiosities. They stressed the importance of adaptability and continuous learning for businesses and individuals to keep pace with technological change.

The panel concluded with advice for decision-makers, emphasising the need to understand these rapidly evolving technologies, invest inclusively, and consider the broader ecosystem impacts when implementing AI and other emerging technologies. They also stressed the importance of developing centralised platforms to integrate AI effectively across enterprise applications and processes.

Conclusion

The discussion underscored the complexity of managing rapid technological advancement while ensuring sustainability and practical applications. It highlighted the significant implications for the future development and deployment of AI and related technologies, emphasizing the need for a balanced approach that considers both the potential benefits and challenges of these emerging technologies.

Jeremy Jurgens: Good morning. Welcome to our session on liftoff. for Tech Interdependence. I’d like to welcome all of our guests here in Davos, Switzerland, as well as those joining us online. We have an all-star cast today, and we’re gonna explore the topic of technology, interdependence, and how these new technologies are coming together and creating solutions that we haven’t seen or even imagined in some cases before. What does that mean, and what can we expect in the year ahead? Now, with us today, I have Mr.Aiman Ezzat, CEO of Cap Gemini, Magdalena Skipper, Editor-in-Chief, Nature Magazine, Aiden Gomez, CEO at Cohere, Cristiano Amon, President and CEO at Qualcomm, and Hiroaki Kitano, Executive Deputy President and Chief Technology Officer at Sony. So, in today’s discussion, we wanna unpack this question of these wide range of emerging technologies. I think we’re all familiar with the advances that we see in AI, and they’re coming more quickly than even the people creating them might imagine they might emerge. But it’s not only in AI. We see similar progress in synthetic biology, in IoT, edge computing, space, quantum computing. And what’s often missed when we look at these technologies individually or through a singular lens is that they’re often interconnected and feeding off of one another. And this leads to actually a kind of compounding and accelerating impact it has. So we’d like to unpack that a little bit today. And to kick off, I’d actually like to start with you, Aiman. You’re working across a range of these technologies. And what’s different? What are you seeing that’s changing here, if anything? And what are we to make of this?

Aiman Ezzat: Sure. So, of course, we are in business and technology. We’re interested in both. So we are very interested about. about how technology is shifting business strategy. And of course, not just individual technology, but the mix of technologies. We sometimes look at technologies individually. We should look at the combination or even how they intersect to be able to drive fundamental changes. I mean, if you go historically, one of the interesting one, of course, is a combination of mobile computing and connectivity, which has created the mobile industry. And the mobile industry is not just about creating mobiles, but how it created a whole new ecosystem around it that goes to creating new business models, changing customer, of course, interaction and making them digital. But beyond that, also having a whole influence on society through development of social media. And I think that’s one of the interesting thing to look at is how these changes are happening. But these changes are also happening today if you combine things like AI and spatial computing or advanced sensors and bioengineering. And just an example that’s probably I can share with you is that we got involved with the America’s Cup to try to work on making the invisible visible. What are we talking about? We’re talking about twin patterns. How can you simulate twin patterns in real time? Sounds complex. So it sounds complex. But here, using spatial computing with connectivity and with AI, we’re able to do that using LiDAR technology and a few other things. And when you look at it, yes, by making the wind visible and wind pattern, then that was for the America’s Cup, but now you can apply it to many other areas. For example, in terms of environmental monitoring, looking at forest fires and things like that, if you’re visible to visualize the wind and direction of the wind, in real time you can, of course, make completely different decision. Same thing in marine technology or you think about air traffic control to look at air disturbance in airports and knowing where the patterns are and how you should basically evolve things. So that’s the kind of thing we’re talking about. more and more is around basically how do you look at the combination between different technologies, what impact it has in actual, you know, impacting basically business and business strategy and potential users.

Jeremy Jurgens: Yeah. I’d like to come to you next, Vamp Lady. You’re in a unique position because you see this even before the technologies get deployed, right, for both, you know, scientific, academic and also helping, you know, general readers understand. Are you seeing changes in the space, both what’s, you know, coming to you and also the speed?

Magdalena Skipper: Absolutely. And so, of course, at Nature, we combine both the journalistic aspect, so as you say, writing about research, technology development, for, let’s just say, interested and to some extent informed audience. This is not a popular science publication, of course, as well as original research that appears in our pages. And, of course, things have changed. And, you know, I can talk about a very long perspective. Nature has been around for 155 years and, of course, a great deal has changed in that time. But there’s, so let me start at a high level, so conceptually, and then I’ll use some examples which actually in many ways follow on from the examples that you gave. So at a conceptual level, I like to think about the broad trajectory of our discovery, broadly speaking, which ultimately leads to implementation and technology development. So, you know, if you come with me back to 1869, when Nature was first published, we were sort of still in the age of the sort of Renaissance men doing science, and they were really almost exclusively men, of course. But the individuals who were engaging in that discovery and beginning to think how that may be implemented were not really compartmentalised the way that they were, let’s say, in the 20th century, where they were very much zoomed into their own discipline, but they were thinking more across disciplines. Now, the 20th century was very much a focus of… on a disciplinary and really drilling in, enabled by developments in technology and tools that allowed us to really look very closely at specific disciplines. And today we’re beginning to sort of come out again, empowered with all the information that the sort of specialist par excellence have developed and continuing to get these very specific, detailed information, but now coming out of it into a broader context. And of course, that’s enabled by connectivity, by greater mobility and these aspects, which are pertinent across disciplines. And of course, AI was already mentioned. That’s just one example, but there are many others. So again, extending on the examples that you mentioned, you know, the kinds of exciting things that I see submitted to our journals, and then you all see when we publish them, for example, questions and solutions that at once look at problems in material science, but also problems in biology. So there’s a bio-inspired materials, which allow us to, for example, capitalize our knowledge of how our environment is sensed in biology by biological organisms, and then use organic materials to implement this in material science. So engineered biosensors are much more biocompatible, much more flexible, can be actually part of organic organisms, bodies, or directly sometimes printed on the skin directly. Another example.

Jeremy Jurgens: Maybe if I could pause, we’ll have a chance to come back to that one. Maybe before I pass over to Aidan, I just would like to understand one thing. We hear about this compression that the technologies are emerging more quickly, and then we see this kind of compression and speed. Do you see that? where you sit on the publication side, this is actually for you, Madeleine. What do you see at Nature? Do you see actually higher quality publications coming in more quickly than previously? Or do you see a kind of continuous rate of change there with the kind of technologies you’re seeing? This is a question for me.

Magdalena Skipper: So in certain areas, progress is faster. But qualitatively speaking, there isn’t a great difference. Because, of course, you always see that a certain discipline is accelerated by certain tool development in that discipline. That becomes saturated, maybe plateaus, and something else comes in. Certainly, the scale of publication, so the scale of data sets, analysis, new knowledge that is delivered in any one scientific manuscript, that is increasing quite dramatically. If I can just add for a second, I think there is reason to pause here. Because, of course, what I’m also describing is that effectively a cost of any one scientific paper to actually generate the knowledge is rising, which in itself is an interesting challenge. How can we reduce the cost of that innovation, new tool development, for further downstream knowledge increase?

Jeremy Jurgens: Great. That’s a good transition to you, Aidan. We’re talking about both the speed. I think you’re at the forefront of driving a lot of these changes. But we also see the increasing costs. And I think there’s this discussion on whether the cost for the computes going faster than the improvement in the algorithms to drop those costs. But it’d be great to hear what you’re seeing from where you sit.

Aidan Gomez: Yeah, I think a lot has changed in the past year. So early on, it was very much enterprises thinking about POCs, and testing, and getting faster. familiar with the technology. Now we’re starting to see enterprises exit that phase and go to production. And that’s leading to new demands. Certainly, cost reduction is one of them. But I actually think that’s less important to enterprises than the actual performance and intelligence of the models. And so we’ll continue to see models get bigger, more expensive. But there’s definitely a high priority on compressing down. In terms of pace, I think from my perspective, since I’m in it every single day, it’s too slow. But every once in a while, we get a breakthrough. I think the most recent of those are this new category of model called reasoning models. And it’s a paradigm shift in terms of how the models go about their work. If you think about what a large language model does, its input space is all of language, right? So it’s literally anything. You can ask it one plus one, or to prove Fermat’s theorem. And the output space, it has to respond immediately. And so the user expectation is huge. Your input is literally everything. And you have to get the right answer on the first try. That’s obviously wrong, right? We shouldn’t spend the same amount of effort on very wildly different levels of difficulty. And what reasoning lets you do is the model actually sits down and thinks through problems. And so now it can spend a dynamic amount of energy and effort on different problems. That unlocks a completely new category of problem. These agentic tasks, you actually expect the models to fail on their first try. That’s OK. That’s part of learning the problem. And then it needs to be able to try again. So these reasoning models will just unlock a whole new tier of capability in the technology. And the other thing is, what we’re finding about them is they can be distilled extremely effectively. So that means taking it from the massive model that costs a billion dollars to train and putting them in a tiny one that can be deployed on edge is extremely effective.

Jeremy Jurgens: Great. I think that’s how you set up the transition there. You know, we have these multi-million dollar, even people talk about, you know, billion dollar data centers that are there. I think we’re not gonna be doing that at the edge, and you have one of the most widely distributed, you know, perspectives on this with, you know, your technology being embedded, you know, throughout. What do you see?

Cristiano Amon: Okay, well, it’s a very broad question, and maybe I will start, I’ll start addressing AI, but I think when you start the panel, you talk about the number of technologies that we’re seeing right now moving very fast, and I’m just gonna list quickly some of the ones we’re working on. I think we have seen that what happened, Eamon mentioned about the mobile revolution, connectivity and computing, but that’s happening now everywhere. Everything is becoming, connected is becoming intelligent. I think we saw that as computational power go to multiple different things. We probably saw revolutions happening out of automotive industry. The car right now is advanced computer, there’s a new computing surface, and it’s becoming, computing’s becoming the most important part of the car. We see that what’s happening in industries, and then on top of this, you have a number of different trends of technology. You have the ability to, computers now understand human language, and that’s enabled by those large language models and gen AI. You can actually communicate with the computer, and that’s not only how we think about it, prompting text also, and how do you see things, how do you hear things? And I think when you combine that with all the computation that is going on all the different devices, it creates incredible opportunity. And I think just to pick on what my colleague was saying, if you look off different directions that AI is going, and you have now very effective compressed SLM, smaller models, you have mix of experts coming up, which you have a model which is really focused on a particular aspect and a particular knowledge base. you have multimodal ability to process different types of inputs. And you combine that with the computing that you have on the edge, you see an incredible transformation, especially as you move from training to production. And we’re obviously very excited about this, but we are seeing, for example, the ability to run models. There were in the order of 15 billion parameters, 20 billion parameters into a phone, into 30 billion parameters into a PC, 60 billion parameter models into a car. And I think that’s revolutionary because you have the ability to fundamentally change the cost equation, especially as you go to production, but also you have things that are very important for the user like latency. If you are in the car and you literally are going to be talking to your car, I think a Gen AI interface is perfect. If you’re behind the wheel, you want a response right away. Certain things that you do and you ask a question, you want the model to respond right away. Latency becomes incredibly important. Privacy becomes important. And there are even some new techniques. It’s actually fascinating to see it. We have done a number of demonstrations of combining a small model into the device at the edge with the big model into the cloud. And what happened, you can even see it as you ask a question to the prompt, to the model, and you start to generate the tokens. All of a sudden, it deletes a couple of tokens and comes back in because it starts sending draft tokens to the cloud and the big model corrects it. And I think we’re gonna see a big change coming as we benefit of all this computer and the distributing. I think this will accelerate, I think, the use of AI and the use of computing at the edge broadly.

Jeremy Jurgens: Thank you. Really exciting and also interesting, this interplay among models. I think, Kitano-san, as your role as CTO, you’re also looking at the interplay across. these different technologies. Can you share a little bit about what is most surprising for you right now with what you’re seeing?

Hiroaki Kitano: Yeah, I think for very interesting things going on in all directions, if you look at the big picture, we are really trying to understand and control very, very large dynamical nonlinear systems. And this is something we are really going for. AI, biomedical, environmental, or predicting wind pattern, designing America’s Cup winningly boat, for example. Everything we do now is really about understanding and controlling very large-scale dynamical system. And for example, in the understanding side, we have a lot of sensor systems. We have all the measurements, laser optics, and all that. And then you have an AI system to be able to make sense out of that and provide some of the possible interventions, where experiment to verify that. And then we have biomedical. We have CRISPR-Cas9, or all kind of methods to intervene. All that technology is getting there. And then there is a combination of that. For example, it would be really powerful to have an AI scientist and a biomedical combined with the robotics and the laser optics and synthetic chemistry. It would really give us an extremely important breakthrough in drug discovery, regenerative medicines, reproductive medicines, and aging control. So it’s not really one area which is important. We really have to combine together, because we are dealing with very complex. Over 25 years ago, back in the middle 90s, I kind of proposed a field called systems biology. I published a couple of paper, which actually was a senior editor at the time. And we worked together, publishing the series paper. And we considered, let’s do the systems view on the biology. And that’s the way to go. Now, after 25 or 30 years in systems, everyone found out. Wow, that’s so complicated. So this is not the area human scientists alone were able to dissect it. We have to work with AI system, which can be able to digest a very large-scale information, come up with a hypothesis, making sense out of that, like a large language model, element of foundation model for the science is the things which moving forward, probably not the final form. We’re going to have a series of evolutions. So what we’re trying to do with the next stage is trying to find the AI scientists or AI engineers, probably I would say AI scientists, either as a tool or like a highly autonomous AI scientist to be able to help us doing the science or do the science by themselves and make things much faster. So I think that is really now the tipping point. It’s really the inflection point. And what we’re having in the scene in AI is the very early stage of industry evolutions. We’re talking about the productivity. So we have like a chat GPT or all the coheres applications and all that. And they were asked to do things much more efficiently, much more accurately. Then we’re going to creativity industry. People can do the creative artwork or entertainment. That will come. And then we’re going to have AI scientists, which will fundamentally change the way we do science. And our civilization is based on scientific discovery, how to take that into the technology. And this impact of AI propelled science in unprecedented speed will change the form of civilization for decades to come. I think that is something we are looking into.

Jeremy Jurgens: And what you’re describing is highly complex there. I guess one question I have is you start to see this interplay and interdependence among the technologies. Absent the AI scientists and the question how do you train the AI scientists, how do you actually manage to to combine these in a way and also manage the risk. I’m going to just open this up for the panel abroad and this aspect of this tension between driving the progress while also understanding the complexity that’s embedded in these kind of new emerging technologies when you start to combine them. I’m into the Magdalena, yeah. No, go ahead, I’ll come in.

Aiman Ezzat: I think what’s interesting is when technologies combine, and again, we look at it from a business perspective, industries don’t just evolve, they restructure. And I think the disruptive level that it brings makes it more and more challenging, especially in more and more technologies, and it’s disrupting many industries completely. I mean, value chains can split. I mean, look at what’s happening around mobility, and Cristiano was talking about it, you know, how the car is evolving, but it’s changing completely, even how do you use a car, what is used for the car, how it gets connected in a whole new ecosystem, providing new services to the driver, it’s not just about mobility, it’s basically you’re creating, again, a new ecosystem that’s creating that, and the challenges usually is to be able to try to predict for businesses, you know, what is new, but also what is next, because you have to try to anticipate. And I think an interesting, I mean, we talked, we heard the AI and bioengineering, interesting example is what’s happening in healthcare, you know, coming from that, because you’re transforming how treatment is administered, you’re transforming how drug is being discovered, or even, you know, the efficiency of development. So you can think about the new wearables, for example, new wearables now is medical, you know, great biosensors who are basically capturing information in real time. Now this data is being analyzed by AI systems to try to identify and flag patterns that have to help to anticipate, you know, some potential issues or symptoms in terms of diseases, which then allow you to move a lot more towards preventive, you know, versus. you know, hospitalized people. So that’s one example, but also some of, a lot of this data is also generated by AI, can now be used as well in development of personalized therapies or personalized drugs. And suddenly you’re moving to a predictable, you know, healthcare environment, which basically helps you to reduce hospitalization, but also create new models in terms of, for patients, for providers, but also for pharmaceutical companies. So we’re thinking really now you’re going to what I would call ecosystem convergence and the impact it can has on complete, redefining completely industries.

Jeremy Jurgens: Thank you, Magdalena.

Magdalena Skipper: So I think it’s worth explicitly articulating one thing that hasn’t been mentioned yet. And, you know, you just used the word ecosystem. Each of the panelists talked about generation of new knowledge, new data, modeling, iterating on that knowledge and discovery. And that ecosystem is of course, inhabited or created by researchers, engineers, developers, talent that exists in academia and in the private sector. And until recently, relatively recently, those two wells have existed in relative isolation. So, you know, and I look at it from a perspective of an editor who wants to publish new insights so that they are available to everyone in a way that is robust, that’s been tested, verified, and then it can be built upon, right? That’s a principle of how science unfolds. And until very recently, that way of communicating the outcomes of the process I just described has been almost an exclusive domain of the academic sphere, but not of the private sector. And I say almost, Hiroaki mentioned. earlier, you know, 20 years ago we were collaborating in particular and Sony is one example of a private sector research enterprise that continues to contribute to that sort of formal research communication. Now why am I saying this? Because I think that’s something that is beginning to change, it needs to change, to enable true building on knowledge and technology development. That sort of common language and publication but also open source sharing of innovation will enable much faster and interestingly and importantly, again something we keep touching on, multidisciplinary benefits across sector and across disciplines.

Hiroaki Kitano: I can actually substantiate what the McDonnell just says, like most recent work we did at the Sony AI is creating a highly autonomous superhuman capable driving agent for the PlayStation game Gran Turismo. We had the cover page of a story of nature I think two years ago and that actually have a goal of creating a superhuman agent to make it again much more creative. But when we decide to do that we have the very fundamental science in it because we have to create AI agent which can have a highly physical realistic autonomous car simulation system and to be able to push the limit of the vehicle control at maximum speed and have to do manoeuvre in racing conditions to be able to overtake. There’s all kind of tactical manoeuvres because it’s a racing game right. So we have to come up with an entirely new conceptual and theoretical framework and not just application of something already there and then do it and run it like a large-scale computing platform or the PlayStation servers and in the clusters. and then train for months, and then do the real competition against top human Gran Turismo drivers, and then get the data, iterate it, and then get to the production stage to actually deploy into the game. And then we publish the paper in Nature. And then a year later, all the agent is commercially deployable. And then if you go to the PlayStation game, Gran Turismo, you can play against this agent. So that’s actually streamlining to very, very theoretical research, proof of concept, publishable academic work, more or less, and then immediately go into the production stage. And that’s entire process of three year. And that is actually the speed of AI and some of the technology we’re going right now. It’s not just taking from something already established. Sometimes you have to really go into the basics, because the theory is not there. So I think that’s something all the tech companies are facing, more or less, is if there is available technology, more or less, it’s an engineering problem. But if you’re going to the turf that science is not there, you have to do the science yourself. Even the private company, no exception, you have to do it. Otherwise, you won’t get it. Yeah.

Jeremy Jurgens: Cristiano, you know a lot about automotives and cars. Can you just continue that transition? We’re going from theory in the virtual world, Gran Turismo. What does that look like? Is that actually continues into the automobiles that we drive around?

Cristiano Amon: Yes. So one of the, I think, many, I think, technologies happening on the car, one is assisted driving and autonomy. And in assisted driving autonomy, it’s evolving very fast, especially as you get more data. We’ve been jointly developing with BMW a stack for a BMW car. It’s going to be launching in the second half. And you do a combination of you have incredible, I think, benefits from using models with generative AI, for example, to get the input that comes from all of the AI that you have in the car. cameras from the cars, but at the same time you also have the rule base because the use case in a car is a little different. The car cannot make a mistake. The first decision needs to be the right decision, especially if you’re doing autonomous driving. And we’ve seen, I think, incredible evolution of this. I envision a future that assisted driving is going to be a feature of every single car, I think, no matter what the tier the car is. And I think AI is going to be fundamental to make that happen. But the other way to think about cars, and I want to make a bridge between cars and different devices, you know, sometimes it’s not about looking at the benefits of AI and applying to things that we do every day. You don’t have to create a new science, but you can see some examples. One of the things that is becoming incredibly popular, like we have been working with Meta on the project for the Ray-Ban glasses, and I think it surprised everybody’s expectations. The reason is when you put a smart glass, even the one with no display, there’s cameras, but because of AI, what you see, the glass sees. What you hear, the glass hears. And you start to see incredible applications. Yes, you can think about you walking around and asking questions and get answers, but for example, a man said about the issue of health care. One of the things that we see in the health care industry right now, doctors and all health care companies, insurance companies are going to tell you, they spend to the order of 30% of their time filling the medical records and say this is what happened with this patient and all that. So you can have something as simple as continue to have the medical appointment you have today, you’re wearing those glasses, the glass saw and heard what you heard from the patient, know what’s relevant, document summarization, you summarize, approve the record. move on. Like the efficiency, it’s very, very high. You apply the same exact technology to the car. We have all those cameras in the car that we use for assisted driving, but you also have a different use case. What the camera is seeing, the car is seeing, so, you know, you can say, look, I’m going to drive. I need to find this place. As soon as you see it, let me know. And the car will be able to recognize and do it. So I think the application is very, very broad. It’s a very single industry, and it doesn’t have to be complicated. I think we’re actually achieving the point that those things are going to be commercially deployed at scale.

Jeremy Jurgens: Great. And before I go to Aiden, if anybody in the audience would like to come in and ask a question, we’ll have that opportunity. Just raise your hand and we’ll get a mic to you. Aiden, over to you.

Aidan Gomez: Yeah, no, I was just going to comment on that story of research going into production super quickly and that pipeline compressing. I think within the LLM space, it’s actually inverted. And so we don’t have time to go through a publishing cycle before pushing to market. A model takes three months to train. We get the result. We see it. We’re not going to wait to publish that in a journal. And then we know it’s better than what was there before, and we’re pushing that to our customers immediately. And so now there’s a retroactive, hey, here’s what we did. We describe, we describe things afterwards. So the competitive pressures of market are so intense. We’ve actually seen an inversion on that pipeline. The second thing was around integration of these systems. One of the key bottlenecks to value in generative AI is the fact that if you think about it, these models are only as useful to an enterprise as the data and systems that they can access. And there are so many barriers, including privacy and access to data customization. So there’s really an integration project, which I think is going to take years to plug these models into all of the systems so that the models can start to drive.

Jeremy Jurgens: Great. Thank you. I see a question at the back. get a mic there, and if you can identify yourself and also signal if you have someone specific you’d like to address your question to.

Audience: Yeah, absolutely. Amir Hussain from World Quant Foundry. My question is really for the team at Sony, and you are the representative here, but I’m quite well familiar with Peter Stone’s work on the driving model. Actually, he’s a very good friend of mine. I’ve always been amazed at how Japanese companies in general, Sony in particular, but also Honda with Asimo, for years and years have really taken on this mantle of AI and have been very AI forward and have made huge investments over decades to push this work, but yet when it comes to actually now the dawn of real working AI, we see a large number of American companies take the lead. Why?

Hiroaki Kitano: Well, that’s an interesting question actually. We maintain substantial robotics and AI capability inside. Stay tuned. We’re not just disclosing it. Okay, so there’s a lot more things. Whatever form of the robotics, we have extremely fast sensor system, and we have extremely good physical realistic simulations, AI to learn it, perceptions and tactical decisions in real time, and then the robotic system to be able to make decisions and then react to that, and we have a project. Okay, so you’re gonna see we’re not fading away. Okay, we are in a game. So at the same time, it’s very interesting because as you mentioned previously, like 10 years ago, 20 years ago, if you talk about humanoids, all Japanese companies were government-researched up, and then now you see lines of humanoids. US companies, some European, and Chinese companies. I think one of the issues is when we tried a Japanese company in the university, actually tried a humanoid, we couldn’t really find a decent application at that time. Technology is interesting. It’s fascinating to create a humanoid. But where are we going to apply to the humanoid robot? Maybe some logistics, some manufacturing process. But of course, the manufacturing process, there’s like a manipulator kind of robot, right? So why you need a humanoid instead with all the manipulations? And why you have to have a biodegraded robot rather than wheels and a manipulator on top of that? So I mean, we have a real boom right now. But we have to see. We have to see in the next five years if humanoid really survive and get a general purpose robotics that we can actually see. Because the humanoid is not cheap. Humanoid is very expensive. We have gone through that. And in our decisions, most of the robotic research as the company made, we have a diversity of robotics, shape, and then all the task delineations. That’s why much less emphasis on the humanoid, and the emphasis on the more specialized robotic and specific shape. Now, we have a resurgence of humanoid. I’m very excited about this, but I’m very curious to see what will be the things that the humanoid can fit for. And of course, technology is different. We have much better sensors. We have a gen-AI. We have those. In the light of new technologies, whether we see a pretty opening of the area for the humanoid, that makes sense in terms of the business, in terms of technological evolutions. But I think many of the Japanese companies and research institutions retain the capability. So once the market is opening, I think people will jump in.

Jeremy Jurgens: Thank you. If we could get a mic to the front here.

Audience: Thank you. Ken Dillon with Occidental. So a question for the whole panel, but just a quick answer. If we’re all back here in 10 years. What is the AI topic of the day then?

Aidan Gomez: Yeah, I can try. It moves so quickly, it’s really hard to say. I think no one can see out ten years, so I’m not gonna pretend to do that. Even a year is kind of hard to see. Reasoning is a huge unlock. I already spoke about that. The next one is learning from experience. So these models today, we spend tons of money training them. Then we push them to production and they’re fixed. They’re frozen. They don’t change. They don’t learn from their interaction with the real world. That will change and unlock a new set of categories. In terms of like big picture problems, I do wonder about the impact to the job market, and we were talking about that earlier before the panel. But yeah, I’m really interested to see what happens. I can’t see out ten years, but definitely you’ll see capabilities that are much more human, much more capable. And they’ll start to be able to do these things that we view as uniquely human, and to be able to do them oftentimes better and more efficiently than humans.

Jeremy Jurgens: Magdalena, then Christiano.

Magdalena Skipper: So I have a flippant part of my answer and a more sensible part of my answer. I think the second part is more sensible. So the flippant part is we won’t be here. It’ll be our avatars here. We’ll be somewhere else in ten years’ time. But the real answer, I think a more interesting answer, is in ten years’ time, all these tools and agents we’re talking about will not be available. We’ll need to be sustainable from the energy cost perspective. You know, we’re talking about deploying AI. So the panel, of course, is about technology liftoff. AI is without a question an enabling element to it, but it’s, from a sustainability perspective today, an incredibly expensive element. We need to solve that issue. Otherwise… as we’re creating all sorts of other issues. I think that’s important. And just to build on what you mentioned as well about the fact that we put these things out there and they stop growing and interacting, really thinking about some of the other themes that came up on the panel, that interactivity across disciplines, across human and AI interface is hugely important, hugely enabling across disciplines, across solutions. No one inventor, no one organization can think of all potential ways to implement whatever it is they’re developing. So again, that communication, whether it happens through publication before or after or some other way, needs to happen in order for us to accelerate that discovery and implementation of those discoveries. And hopefully we’ll be there in 10 years’ time.

Jeremy Jurgens: Christiano?

Cristiano Amon: Yeah, you know, I agree with, it’s hard to predict 10 years, but maybe there’s a couple of things. We can probably say it’s going to, we have ability to understand how it’s gonna be in 10 years. So just to be provocative a little bit. If you look at all of us, mankind, I think as the years went by and the toys changed, but we still kind of communicate the same way. We see things, we hear things, we talk. And when you think about the computers that assist us to do things, if you just go to the history of the person, the computer, we have been adapting to the ability of the computer to interact with you. So the first personal computer, I think when IBM and Microsoft, MS-DOS, it has an ask-to character, and you communicate it to the ask-to keyboard. Then the mouse got invented. When the mouse got invented, the OS and how the computer interact with you changed. It became a graphical interface, it can point to things. Then you know what happens when the computer fell in the palm of your hand and the ability to touch. Well, now with AI, computers understand our language. you can talk to them, and you will see what you’ve seen. So I think what I can predict is how we think about computers, OSes, and applications is going to be completely different when we think about the next 10 years. And maybe it’s going to take about five years to try to get scale. But applications are going to be written different. Computers are going to function different because we just changed fundamentally the interface. And we’ve seen that happening to generations. And I think that’s one of the probably most not discussed part of AI, which is with AI, you can have a computer to interact with you. And even that can make humanoid robots more useful. And I think definitely we’re going to see that within the next 10 years.

Aiman Ezzat: Yeah, for me, I’ll bring it back to the convergence. It’s not AI alone. It’s AI with other things. I think bioengineering is going to be quite impactful, I believe, in terms of evolution. What kind of organism, what kind of evolution we can get through that. Quantum, again, if you’re really able to stabilize these quantum computers, this could have an impact. In 10 years, will we be there? Don’t know. We will see. And I would add something else, which is becoming a bit back in fashion, is nuclear, notably with SMR technologies. So when you start looking at convergence between some of these technologies, you might see new things emerging. And it’s not just AI on its own. It’s AI combined with some of these other next technologies that could be quite interesting.

Jeremy Jurgens: Maybe wrapping up here, I’d like, actually, we’ll start with you, Hiroaki. We have a lot of decision makers in the room here and also people following us online. What advice would you give to be able to unlock the kind of combinatorial power of these technologies that we’ve been discussing here? What’s the very pithy piece of advice that you would give to decision makers and leaders on? What do we need to do to actually unlock these capabilities in a way that’s safe, and as Magdalena pointed out, sustainable?

Hiroaki Kitano: Yeah, I think the key is to really get it to the core of the problem. Again, like I mentioned, these are complex dynamical systems. And the reason why some of the discussions about 10-year horizons is difficult, and also AI has not been practical before. Now it’s getting there. And robotics, we have to see, is if it can find the environment, which is like a tail part of this, a head part of distributions, the easier, because it’s more obvious, more data. Difficulties, autonomous driving, humanoid in daily life, and then most of the AI is tail side. Long tail distributions, small minor exceptions, scarcity of data, and then that piles up to a substantial part, 30%, 40%, even 6% in all the cases. How you handle that? How you AI going to perceive, and handle, and provide a solution for robotics to be able to cope with that? Robotics are expensive. You have to be able to do that. So that part is going to be the real challenge for the next 10 years. And if we get the solution for that, I think we’re going to be very good. We’re going to be revolutionized in the entire society and industry.

Jeremy Jurgens: Good, I’m going to invite the remaining panelists to compress and accelerate their answers. Cristiano?

Cristiano Amon: What I’ll say is, technology is moving very, very fast. Speaking from companies like Qualcomm that have been in the tech sector, you have to always reinvent yourself. So I think my advice to people is, a lot has happened in one year. I think he said, like, every three months you have something new. Try to understand what’s happening, because this technology is incredibly disrupted. It’s going to change your company, and I think it’s important for everyone to stay on top of it.

Aidan Gomez: I’ll be quick. So I think, from our perspective, Cohere is focused on productivity in enterprises. And the whole challenge is developing platforms. It’s not enough to just buy the AI feature inside your existing applications. You need a centralized platform to interface across all of those. That’s where the real opportunity lies for these models.

Jeremy Jurgens: Thank you. Maddalena?

Magdalena Skipper: Very quick. Think broadly, and if you are in a position to invest, invest inclusively in the full sense of that word.

Aiman Ezzat: For me, it really plays the role of orchestration, define what you want to invest in, but also think ecosystem and what you need to partner with to be able to complement your capability and evolution of how you can be disrupted and how this ecosystem can disrupt industries.

Jeremy Jurgens: Great. Thank you very much to our panelists there. I wish everybody a good continuation of the summit and also thank our viewers online. Thank you.

Agreement Points

Convergence and interdependence of emerging technologies

speakers

– Aiman Ezzat
– Magdalena Skipper
– Cristiano Amon
– Hiroaki Kitano

arguments

Combination of technologies driving fundamental changes

Intersection of disciplines like material science and biology

Everything becoming connected and intelligent

Need to understand and control large dynamical nonlinear systems

summary

The speakers agree that the combination and intersection of different technologies and disciplines are driving significant changes across various sectors, leading to new innovations and challenges.

Impact of AI on industries and research

speakers

– Aiman Ezzat
– Magdalena Skipper
– Aidan Gomez
– Cristiano Amon

arguments

Technologies restructuring and disrupting entire industries

Changing dynamics between academic and private sector research

Compression of research-to-production pipeline in AI development

Running large AI models on edge devices like phones and cars

summary

The speakers highlight how AI and other emerging technologies are reshaping industries, research processes, and the relationship between academia and the private sector.

Similar Viewpoints

Both speakers emphasize the transformative potential of AI in enabling new capabilities and changing how we interact with technology.

speakers

– Aidan Gomez
– Cristiano Amon

arguments

Reasoning models unlocking new capabilities in AI

AI enabling new interfaces and interactions with computers

Both speakers stress the importance of addressing practical challenges in deploying advanced technologies, including cost-effectiveness and finding meaningful applications.

speakers

– Magdalena Skipper
– Hiroaki Kitano

arguments

Need for cost reduction and sustainability in AI deployment

Importance of finding practical applications for technologies like humanoid robots

Unexpected Consensus

Rapid pace of technological change and need for adaptation

speakers

– Aidan Gomez
– Cristiano Amon
– Hiroaki Kitano

arguments

Compression of research-to-production pipeline in AI development

Necessity of staying informed about rapidly evolving technologies

Need for AI to handle edge cases and long tail distributions

explanation

Despite coming from different sectors, these speakers all emphasize the unprecedented speed of technological advancement and the need for continuous adaptation, which is somewhat unexpected given their diverse backgrounds.

Overall Assessment

Summary

The main areas of agreement include the convergence of technologies, the transformative impact of AI on industries and research, and the need to address practical challenges in technology deployment.

Consensus level

There is a high level of consensus among the speakers on the transformative potential of emerging technologies, particularly AI. This consensus implies a shared understanding of the challenges and opportunities ahead, which could facilitate collaborative efforts in addressing these issues across different sectors and disciplines.

Different Viewpoints

Approach to AI development and deployment

speakers

– Aidan Gomez
– Magdalena Skipper

arguments

Compression of research-to-production pipeline in AI development

Need for cost reduction and sustainability in AI deployment

summary

Gomez emphasizes rapid deployment of AI models to market before formal publication, while Skipper stresses the importance of sustainability and cost reduction in AI deployment.

Unexpected Differences

Future of humanoid robots

speakers

– Hiroaki Kitano
– Cristiano Amon

arguments

Importance of finding practical applications for technologies like humanoid robots

AI enabling new interfaces and interactions with computers

explanation

While Kitano expresses skepticism about the practical applications of humanoid robots, Amon’s optimism about AI-enabled interfaces could potentially apply to humanoid robots, creating an unexpected difference in their perspectives on future human-machine interactions.

Overall Assessment

summary

The main areas of disagreement revolve around the pace of AI development, the balance between rapid deployment and sustainability, and the practical applications of advanced technologies like humanoid robots.

difference_level

The level of disagreement among the speakers is moderate. While there are some differences in approach and focus, there is a general consensus on the transformative potential of AI and emerging technologies. These differences highlight the complexity of managing rapid technological advancement while ensuring sustainability and practical applications, which has significant implications for the future development and deployment of AI and related technologies.

Partial Agreements

Both speakers agree on the importance of integrating AI into existing systems, but Gomez focuses on enterprise data integration, while Amon emphasizes running AI models on edge devices.

speakers

– Aidan Gomez
– Cristiano Amon

arguments

Integration of AI systems with enterprise data and processes

Running large AI models on edge devices like phones and cars

Similar Viewpoints

Both speakers emphasize the transformative potential of AI in enabling new capabilities and changing how we interact with technology.

speakers

– Aidan Gomez
– Cristiano Amon

arguments

Reasoning models unlocking new capabilities in AI

AI enabling new interfaces and interactions with computers

Both speakers stress the importance of addressing practical challenges in deploying advanced technologies, including cost-effectiveness and finding meaningful applications.

speakers

– Magdalena Skipper
– Hiroaki Kitano

arguments

Need for cost reduction and sustainability in AI deployment

Importance of finding practical applications for technologies like humanoid robots

Key Takeaways

Emerging technologies are increasingly converging and interdependent, leading to transformative changes across industries and scientific research.

AI advancements, particularly in areas like reasoning models and edge computing, are enabling new capabilities and applications.

The integration of AI with other technologies like bioengineering, quantum computing, and robotics is driving innovation and reshaping industries.

There is a compression of the research-to-production pipeline, especially in AI development, changing how innovations are brought to market.

Sustainability and cost reduction in AI deployment are crucial challenges that need to be addressed for widespread adoption.

The interface between humans and computers is fundamentally changing due to AI, potentially revolutionizing how we interact with technology.

Resolutions and Action Items

Companies and decision-makers need to stay informed about rapidly evolving technologies to remain competitive.

Enterprises should develop centralized platforms to effectively integrate AI across their applications and processes.

Investment in technology development should be inclusive and consider broader impacts.

Unresolved Issues

Long-term impact of AI on the job market and workforce

How to effectively handle edge cases and long tail distributions in AI applications

Practical applications and economic viability of humanoid robots

Balancing the competitive pressures of rapid AI development with the need for thorough research and publication

How to achieve sustainability in AI deployment given current energy costs

Suggested Compromises

Combining small AI models on edge devices with larger models in the cloud to balance performance and efficiency

Retroactive publication of AI advancements after market deployment to balance competitive pressures with knowledge sharing

Developing specialized robotic shapes for specific tasks rather than focusing solely on general-purpose humanoid robots

We sometimes look at technologies individually. We should look at the combination or even how they intersect to be able to drive fundamental changes.

speaker

Aiman Ezzat

reason

This comment shifts the focus from individual technologies to their combined impact, highlighting the importance of considering technological convergence.

impact

It set the tone for the rest of the discussion, encouraging panelists to think about the interplay between different technologies rather than discussing them in isolation.

These reasoning models will just unlock a whole new tier of capability in the technology. And the other thing is, what we’re finding about them is they can be distilled extremely effectively.

speaker

Aidan Gomez

reason

This introduces the concept of reasoning models as a paradigm shift in AI, highlighting both their enhanced capabilities and potential for efficient deployment.

impact

It sparked a more technical discussion about AI models and their practical applications, leading to further exploration of AI deployment in various contexts.

Everything is becoming, connected is becoming intelligent. I think we saw that as computational power go to multiple different things.

speaker

Cristiano Amon

reason

This comment encapsulates the trend of ubiquitous computing and intelligence, emphasizing the pervasive nature of technological advancement.

impact

It broadened the discussion beyond AI to include IoT and edge computing, leading to examples of how these technologies are transforming various industries.

We are really trying to understand and control very, very large dynamical nonlinear systems. And this is something we are really going for.

speaker

Hiroaki Kitano

reason

This comment frames the challenge of modern technology in terms of complex systems theory, providing a unifying perspective on diverse technological efforts.

impact

It elevated the discussion to a more abstract level, encouraging panelists to consider the broader implications and challenges of technological advancement.

That sort of common language and publication but also open source sharing of innovation will enable much faster and interestingly and importantly, again something we keep touching on, multidisciplinary benefits across sector and across disciplines.

speaker

Magdalena Skipper

reason

This comment highlights the importance of open collaboration and knowledge sharing across academia and industry for accelerating innovation.

impact

It introduced the theme of cross-sector collaboration and open innovation, leading to discussion about the changing dynamics between academic and industrial research.

Overall Assessment

These key comments shaped the discussion by consistently pushing it towards a more holistic, interconnected view of technological advancement. They encouraged panelists to consider not just individual technologies, but their combined impact, practical applications, and broader implications for society and industry. The discussion evolved from specific technologies to overarching themes of convergence, complexity, and collaboration, providing a rich, multifaceted exploration of the future of technology.

How can we reduce the cost of innovation and new tool development for further downstream knowledge increase?

speaker

Magdalena Skipper

explanation

The increasing cost of generating knowledge in scientific papers poses a challenge for continued innovation and progress in research.

How can we effectively distill large, expensive AI models into smaller ones that can be deployed on edge devices?

speaker

Aidan Gomez

explanation

This is crucial for making advanced AI capabilities accessible and cost-effective in a wide range of applications and devices.

How can we develop AI systems that can learn from experience and adapt after being deployed in production?

speaker

Aidan Gomez

explanation

Current AI models are static after deployment, limiting their ability to improve and adapt to real-world interactions.

How can we make AI and other emerging technologies sustainable from an energy cost perspective?

speaker

Magdalena Skipper

explanation

The current energy requirements for AI deployment are unsustainable and need to be addressed for long-term viability.

How will the convergence of AI with other technologies like bioengineering, quantum computing, and nuclear energy (SMR) create new opportunities?

speaker

Aiman Ezzat

explanation

The combination of multiple emerging technologies may lead to unforeseen innovations and applications.

How can AI and robotics better handle edge cases and long-tail distributions in real-world applications?

speaker

Hiroaki Kitano

explanation

Addressing rare and complex scenarios is crucial for the widespread adoption of AI and robotics in daily life and industry.