The Role of Government and Innovators in Citizen-Centric AI

precisely this, how do we sort of build capacity in order for this technology to be applied significantly better. And in the days to come, I would really love to see a day when India and the EU collaborate much more closely to make this happen, not just in India, but all over the global south. Thank you very much for having me. Thank you very much. Don’t go away, because now I’m going to call the panel. We have a distinguished panel today, but we would like to take a picture first. So if I can invite Vice President and Secretary Krishnan to stand here, and then I invite Arthur Mensch. He’s the co -founder and CEO of the European Union.

He’s the CEO of Mistral AI, if you can just stand next to the Secretary, which is a European company developing large language model, but also Jarek Kutuloski. who is the founder and CEO of a German company called DeepL, which is on language technologies. Matteo Valero, who is a professor of computer architecture at Technical University of Catalonia and the founding director of the Barcelona Supercomputing Center. And from the European Commission, I’m pleased to announce Roberto Viola. He’s the director general of DigiConnect. And he plays a pivotal role. He’s the director general for our digital policies. Okay, so as I said, it’s a very distinguished panel from the European Union. And I would like to thank all of you for being here to participate.

I’ll start with Artur from Mistral. I repeat that he comes from Mistral, which is a European model and one of the main large language models. In your opinion, how can LLMs or general purpose models in general reshape the public sector? And as a developer, how do you work with governments to apply it in the public sector?

I’m the co -founder of a company called Mistral and we effectively train language models and perception models and we then use them to create applications for businesses and for states typically the models is never enough to actually provide value for the states we work with we have a program called AI for citizens that have multiple pillars but when we work with states the first thing we work on is efficiency what generative AI allows you to do is to delegate tasks in general and to automate certain processes that can be fairly complex, that can be fragmented, that can involve multiple people, that can involve multiple tools that can deal with IT legacy and so a state is not different, an administration is not different from an enterprise in that respect, that they have IT problems, in that they have processes that are sometimes inefficient in that they have pressure on talents because there are a lot of people that are actually retiring, so knowledge is a very big problem and management of the ledges is a very big problem.

The kind of things we do is related to that. So we deploy our horizontal platform and we create use cases. We work backward from use cases that are around procurement, that are around writing reports on the, visible in that it can show to the citizens themselves is building public services on top of artificial intelligence. And so one example is we worked with France Travail which is an employment agency in France to actually help with the matching of job employers, of employers and of people seeking jobs. And often times people would just connect and they’re looking

Thanks a lot. I now turn to Yarek, founder of DeepL, which has been a very important part of the project. Yarek has been working since 2017 in AI -based translation tools. and so there is a lot of linguistic diversity in India as well as in the European Union and so how can the AI language models help to overcome this multilingualism issue I say of course we consider it also a benefit but in administration it can be sometimes be a challenge

I would definitely try to not characterize it as an issue I think it’s something that’s actually pretty beautiful about a lot of the countries that are so multilingual and there’s a lot of differences in how deeply multilingualism is embedded in different countries and in different societies I think here in India everybody understands it extremely well but it’s not the only country in the world and there’s countries like Canada, there’s countries like Switzerland whom we’re working a lot with the public sector that have this intrinsic necessity of being able to connect to their citizens in very many languages and where partially that communication is even embedded as a part of their constitution. And here, those countries have been struggling over the years, maybe as you have indicated, on how to actually make this happen.

And AI and those kinds of frontier models that we build and the applications on top of them that are specifically tailored to bridging this communications gap, they help a lot. Nowadays, not only in written language, but also in spoken language, enabling real -time conversations maybe with citizens in a setting when they come up into an office and want to get… certain service done. So a lot of options there, but also a lot of complexity as those use cases that governments have really differ very, very much based on what you’re doing. It’s another challenge to translate legislation into different languages. It’s another challenge if you want to enable those real -time conversations with citizens. Quite a lot of exciting problems to solve.

Thank you very much. Now I turn to Matteo Valero. You are also a professor, but you’re also the director of the Barcelona Supercomputing Center. So can you maybe explain, you’re also in an AI factory, what the AI factory does and how it can help the transformation of the public sector and of SMEs?

Thank you, Lucila. Good afternoon to everyone. It’s my pleasure to be in India once more. Sorry? Sure. My pleasure to be here. You have an incredible country, believe me. So, thank you for inviting me. And I am going to start 50 years ago when Seymour Cray produced the first supercomputer, no? And this supercomputer increased the speed from 10 to 10 until now, okay? With this computer, we did simulation and we produced better results in science and engineering. In Europe, every country was alone until, thanks to Roberto Viola, we created the EuroHPC. And then, because we had the EuroHPC, we have now a reasonable amount of power in the supercomputers. So, if you look at the top 500, probably out of the first 15, we have 6 in Europe.

And we do science. We do science and this is very good. So now because the data, because the computer, and especially because the research of these guys and many others, the AI is invading us. It’s changing any activity we have. In my field, I am changing the way we do high performance processor. In the supercomputing center, let me tell you that we are 1 ,400 and we have 500 people doing hardware software, using or designing in topics related with the AI. There is no question that now the data, the control, the data, the computers, and the algorithms are dominating the world. So what we could do in Europe, we have the supercomputer, but we need to devote more energy in order to…

get the AI distribute around any activities. So the idea of the European Commission was create the factories and now the gigafactory. The AI Factory is a platform, AI platform is hardware and software, but as important as that, it’s co -located where there are people with the skills in AI, there are people with experience in transferring technology to the society. So the idea of this AI is the service is free, the people is free, is to connect as much as we can with the society to make a better world. This is the target for us. Obviously, there are many, many possible contributions, and one of them is the administration, and obviously how we can make happy to the citizen.

If we make happy to the citizen, we are successful, okay? And we can make happy to the citizen if we provide them with personalized information, accurate and fast. After that, a second question. I will give example, but I think… So this is the target for the AI factories and the gigafactories is the same but competing with the data center. Because I forgot one thing. What Europe could do is just to use this data in the platform from outside or create our platform to use our platform using this data. I think this is the right way to go.

Thanks a lot. So we have talked about what the models can do, the computing capacity that is made available. Now, Roberto, I would like to ask you, since you have reflected and designed all of this, how would you now… Mention your words. I’m your boss. Yes, he is. How would you help now facilitate the uptake of AI? By the public sector, because if we have the models, we have the compute capacity or we are building it. and we’re also building more access and more availability of data sets. But how can we make sure that the public sector actually uses AI?

Thank you. Thank you, Lucille. Good afternoon, everyone. It’s really, for me, a pleasure to be here and to be together with Lucille, with the three crown jewels of Europe, which are all very much representing what is for us giving out to citizens, society, innovation. Because you can test the Mistral on the web for free. You can use DPL for free on the web and test it and enjoy it. Translation from all Indian languages to European languages, I dare to say, yes. You can test Destination Earth. Destination Earth is the most sophisticated climate digital twin of the world, AI digital twin. You can replay the climate of the past into the future. You can zoom in in certain areas.

You can have a resolution which was an error of 200, 100 meters, three weather events, because there are already two twins which are running, the twin of the climate and the twin of extreme weather events. Again, for free on the web. So this is the first point I want to make. There’s an economist, maybe you know the name, Mr. Solow, that he expressed with numbers and, I mean, evidence a paradox. The more people invest in IT and software and other infrastructure, the less the productivity. Actually, there’s no… There’s no productivity gain in doing that. So it’s called the solo paradox because it’s a paradox because you as a user, me as a user, experience a much better user experience to have a public administration which is more digitized or an hospital where everything is digital as well or a doctor which is savvy because it has an AI co -pilot.

But in terms of productivity gain, according to the solo paradox and the numbers that he has, in a compelling way, put in front of us, there’s no productivity gain. So many economies, and of course, who solves in this room this paradox is for Nobel Prize of Economy. So, I mean, the challenge is open. So I’m not going to solve it, but I try to answer the question of Lucy. The reason why many have observed this is that because normally, IT, and that includes also now AI, overlaps what exists. And of course then it becomes very intuitive. Imagine an hospital, I mean, having all the doctors still and nurses and everyone in traditional process. So doing a bit of paperwork as they do, but also doing it digitally.

And having two systems running in parallel, of course, I mean, you imagine that the productivity doesn’t move much. Now we have seen some changes during COVID. Why? Because people, I mean, were secluded and they were forced, I mean, to use only digital. So in certain areas, sadly, I mean, you saw the productivity was in a way more linearly linked with the use of technology. The European Investment Bank has published an econometric study that shows AI as a productivity increase of 4%. Which is not the stellar numbers, I mean, some of the vendors around say, but it’s… compared to the solo paradox is not zero. And I think this sign is because with AI, especially agentic AI, you see the change.

So you don’t see the overlap anymore, one process with the other, but you could see that there’s a new process, new way of man -machine interacting and working. But Arthur, before, said something which is the key of all of this. Because if people in the public sector are not empowered, they don’t understand it. They are not part of this change. The change will not lead to any productivity gain. Because you can have the most expensive and sophisticated AI software of the world, probably absolutely not needed by the private sector, because better to have bespoke models, open source, that serve the purpose. But even if you have the most sophisticated, you still get that one. if you have someone that refuses to embrace the technology or in any case you have an organization a process that is not ready not fit for it then there’s no productivity gain.

Maybe as a citizen you can see their wonders but in reality I mean the old system becomes two times more expensive and the adoption rate is low and this is really the real challenge of artificial intelligence as paradoxically it can be. I think we can proudly say that as I see it in India and I see it now in Europe, we are developing an ecosystem which is really brilliant, self -reliant, sufficient in terms of good company producing open source, producing language technology, producing advanced algorithms. We have supercomputing center offering capacity. All of this, I mean, goes in a completely different model compared to other models, and it’s all fine. But now, I mean, we really need to work with the people and with the public administration and to make sure that we

Okay. So, Arthur, if I were to ask you, how do you get the AI accepted by the citizens and also the public administration? What kind of tools? You already provide, of course, the chatbot Le Chat. But what are other tools that you think will be easily accepted?

Well, we’ve turned Le Chat into something that we call Vibe, actually, which is a product where we can delegate tasks. We can delegate tasks fully and delegate workflows. The challenge and the reason why you don’t see productivity gains when you deploy chatbots in enterprise is that basically you’re focusing on an individual productivity gain. so that’s the case in enterprise but it’s the same in administration and so if someone can actually write a mail faster it’s not actually changing the way your business is being run when the thing starts to change if you look at a full process let’s say procurement for instance which typically you entail like multiple touch points with multiple people and you ask the agent to actually run the process itself so you move from an individual productivity endeavor to a collective productivity endeavor and you move from being equipping ICs so individual contributors to equipping managers that are going to span the same way a manager will delegate sometimes to a human it can delegate sometimes to an AI process and there are two big challenges associated to that and that needs to be solved for product but also through human interaction I would say.

The first is that you need the process automation that we run and we design them we bring our engineers in, they work with subject matter experts and they design they write the code using our coding model and then they deploy the code that is going to run the automation and that’s going to ask questions, that’s going to interact with the tools. The way we design them is to try and get the humans out of the way because the problem is that the process only brings productivity gains if you’re not bottlenecked by the humans themselves, if you’re not interrupting them all the time. A good example is coding. If you want to code faster with AI, you need to give them tasks and then disappear and then you come back maybe one hour later and the task is done.

If the thing comes and nags you like five minutes after, maybe you’re doing something else and so the thing is actually not progressing as fast as it should. You need humans to actually get out of the way of the AI automation if you want this automation to work. And then the second thing that goes with it is that once you’ve done the automation, you need to rethink the organization because once you’ve automated your procurement process, well, suddenly the people that were actually running the analysis of the procurement needs to do something else. And that’s actually… Take… some thought around how you’re reorganizing people, how you’re rescaling people, how you’re turning individual contributors into people that will effectively manage AI -operated processes.

And so you need, and enterprises need, to actually turn people that were used to do menial work into people that are delegating those work. And as you know, and as every manager in the room knows, it’s actually fairly hard to learn how to delegate and to move from being an individual contributor to being a delegator. And because the only way AI actually brings you productivity gain is through strong delegation and long execution, well, every one of us needs to become a strong delegator. And so that takes some training. We are not trained to be delegators at school in Europe, I would say, at least in France. And that’s something we will need to learn. And so we’ll need to learn it early on.

And we need to reskill the people that are, that needs to learn a new way of doing things. A new way of working. And to rewire the brain. I think a very good example, I’ll stop with that. is that we have our coding tool called Mistral Vibe. And what we see is that if you take very young developers, they use it very quickly. And so they learn how to use it. They’re excited. The way they work is inherently wired to using AI because they are 23, and that’s everything they’ve ever done in coding is through AI. Then you have the very senior people that are like 35 years old, I guess, or 33 my age. And those are still much better than the agents themselves.

So they know how to design the architectures. They can give some very precise guidance on how the agents need to rewire the code or bring a new feature. And then the problem is the people in between. The people in between, well, they got very attached to writing code. And now they need to rewire. They were very good at writing code, but they need to become something else. else. And so that’s where reskilling is very, very important. And that applies to software engineering, but that will apply to everybody working on knowledge. And so that’s the three years ahead of us, and we need to work together to make it as smooth as possible.

Thanks. And indeed, Jarek, you have developed AI agents. So when you think of applying them to the public administration with all these caveats we just heard, what do you think the acceptance is going to be like?

Yeah, I think it’s not only about the individuals and how people reskill and how people adopt AI and how they learn to use AI, but it’s also about organizations really rethinking the way that they are working. Thinking about workflows, thinking about processes, whether that’s like general purpose agentic workflows, or whether that’s something that has language at its core, rethinking of how do we do these things. We’ve gone over a couple of decades now improving those processes and maybe putting parts of AI, especially in language processes, that’s been already happening over the last years quite significantly. But we haven’t yet rethought those whole processes. Like, do we need that human review step anymore in a particular use case?

Or is it just enough to use AI? We have organizations who are translating R &D documentation for drug discovery and submitting that to the local regulators, just purely translated by AI with the appropriate guardrails. We have organizations that are translating plain maintenance records and using them as the source of truth. So there is a lot of potential in using AI, but you have to think a little bit out of the box and really forget the old ways of doing things. And the same holds for agentic AI, and I think even more so, because the potential of AI is even bigger. So it’s… It’s both for the public sector and for businesses. It’s a big redesign of how work gets really done.

And the bigger the organization, the obviously bigger the inertia that is out there. And public sectors tends to be the largest organizations in any country. So the challenge is even bigger there.

Thanks. And so, Matteo, as in the Computing Center of Barcelona, it is quite specialized also in applications for public sector, for health care. So what do you see as the main applications that are being developed on the basis of demand?

teach to the young people to understand the problem to propose solution. Thank you.

So, Roberto, you heard the challenges in terms of acceptance and implementation in the public sector, which were sometimes maybe the skills are not very strong. So how do you think that policy can really help to enable this transformation?

I think policy needs to be tuned with the transformation. So in a way, as I was trying to say also before, if you invent a digital bureaucracy, it’s a bureaucracy. It’s digital, but still it’s a bureaucracy. And you have then a bureaucrat and a digital and an AI agent bureaucrat. I mean, It would be very simple for the geniuses in this panel to produce an AI bureaucrat. And I’m sure AI can do bureaucracy even better than us, much better than us. Regulation -generating bots, yes. That would be super useful. Or regulation -correcting bots, that would be good. So you see, I think we need to be also from the legislation side disruptors and look at things with completely different eyes.

And for this, let me say that there are one thing that really does a striking similarity between Europe and India is this idea of believing in the public stock. So the idea that you can actually be managing your identity, your attribution. your capacity to sign, to timestamp, to actually exchange these attributes in an open source and an open model. This is for people and for businesses. And then in this way, the state is in your hand. I mean, it is actually, you have the bureaucracy under control because the bureaucracy, it’s you. So, if you actually reverse the logic of the citizen going to an office, that’s what you refer to, to the office going to the citizen with, of course, all sorts of nice agents, push notifications, attestations, then, I mean, you re -engineer the state.

So, my point is, if we dare, and I dare to say in India you are daring, in Europe we are daring, you can actually redesign the paradigm and then if you do that then creativity is really at work because there can be many different agents many different ideas on how to improve processes Thanks

Now we only have very little time to go but before leaving since we have these four geniuses I would like to ask you maybe a very last thought that you have on innovation in the public sector and how you can contribute

I think public research is very important in particular I think partnerships between private companies and public efforts is actually something that works because doing research takes some infrastructure infrastructure takes some capital and so I think that’s the way we can actually accelerate together

I would say that the the the the the AI is a dual use technology and we need to look for the good use in this direction I think we can do a lot in Europe because as Roberto said we have the infrastructure and then we need a little more to invest a little more and then to define common projects because don’t forget that if you look at the power at the national level between the state and China they have more than 80 % of the computer more than 80 % of the people and more than 80 % of the investment so what we can do one possibility should be being in India alliance I think it would be very good to have an alliance between Europe and India in this topic we as BSE we have alliance with the SIDAC with the super competing centre and we have with the Institute of Science in Bengaluru.

And also financed by the Commission, we have a very good project that we are very happy to collaborate with you in the end. Thank you.

Now, time is up, but I would like to know very shortly, last thought from Roberto and Jarek.

Yeah, I think we can build and we will build from the commercial side, from the business side, amazing products that are driving a lot of value creation in the AI space. I think that that’s clear. And we’re going to be trying to do that in a way, of course, that our users and our customers can be really delighted by those products. But I think there is a lot of work that the public sector can do in terms of bringing this importance of adopting this technology into the broad base of population. I think both the German and also the European countries are going to be very happy and also from all of the conversations that we had here, the European and Indian governments do understand that, but we should not underestimate this challenge.

And I think there needs to be a very strong partnership between… businesses and the public sector on driving that. Thanks.

Thank you.

I am one of the few that has been in the three summits. I mean, Blanchley, this one, and last year in Paris. And of course the size already gives you an idea how things have changed. In the kind of discussion room at Blanchley Park we were 20, including the leaders. I mean, that gives you an idea. Now, the point is, I’m so happy to be here because what I always thought a little bit is true. There’s not one future for AI and technology. And it is not written. It is not written. The thousands and thousands of people that participated to the summit this year will write the future. So those that tell you there’s only one way, I mean, there’s only one scale.

And the rest of the world should watch and applaud. and I mean adapt to it absolutely I mean this summit shows and application of AI in public service what India is doing, what Europe is trying to do shows there are many futures and as I was trying to say before the future is in our hand

Thanks a lot and with these very intelligent and smart sentences tell me to thank the speakers and thanks a lot for your participation Thank you Thank you Thank you