Open Source for AI and Emerging Technologies, continued - UN Open Source Week 2026 (Part 2)

We are off to our last two sessions with a great lineup of speakers.
But before we get to that, we'll start with a quiz with the audience.
First question is this.
Please raise your hand and then you can open your microphone if you know where the QR code for the feedback form is.
I see there in orange.
Do you want to turn on your mic and tell us? Or she's showing it.
If anyone wants to fill out the form, go to her, or you can also go to the doors when you're exiting, you have those on both sides.
Another question, what can you do if you're having a hard time hearing what the speakers are saying? Raise your hand if you know the answer.
Over here.
Excellent.
Use one of the ear pieces.
Also we have tons of space here now, so feel free to come closer.
It's time.
I'm going to head in.
I'm going to be giving the microphone to my dear colleague Filippo, who is going to be moderating in the next session over to Filippo.
Thanks, Pappy.
Hi, everyone.
I know that we are among the last things between you and the reception, do bear with us, but you are going to have a very interesting and exciting session called Open agents governing the Next Generation of autonomous AI systems.
This session will cover how AI agents are already capable of planning, acting, executing tasks autonomously across digital environments and what are the challenges and opportunities of AI agents, of course, from also an open source perspective and what are a what is the meaning and the opportunities of international cooperation on the agents.
Of course, you'll hear from different perspectives on this, but without any further ado, let me give the floor to the opening keynote, Ted Style of Professor David Shearer, who's the professor of Practice at the Imperial College of London and so much more than this, but he told me not to say it in the introduction.
Professor Shearer, please, you have the floor.
Thank you so much.
Can you hear me okay? Is this better? All right.
Hi.
I'll try and keep this moving quickly so that we can get to the panel.
What I'm going to speak to you about to tee things up is something called intelligence capital and how we create the agent based enterprise.
Around the turn of the year, I was updating some work I've done a few years ago with the investment bank Evercore on valuing the hyperscalars and valuing all the AI companies because we were all scratching our heads and saying, why are they worth so much money? This doesn't make sense based on rational norms of valuation.
And we dug in a little more and realized that the old way of looking at things was not how the world was turning and that we were entering a new era.
Um, and as I dove deeper down the rabbit hole, I came back with intelligence capital, which I shared first at DeVos and today we're going to go a little deeper into how it plays out at the enterprise.
The basic idea, next slide, is that value creation is no longer coming from looking at a stream of future cash flows, which is how we used to value companies.
Instead, we are looking at a stream of future intelligence flows, right? This is something we call intelligence capital.
For the first time, We have an asset that increases in value the more you use it.
This iPad, as soon as I took it out of the box, it started losing value, it began depreciating.
As soon as I used it more, it lost more value.
But intelligence, the more you use it, the more it's worth.
This is true for human intelligence as much as artificial intelligence, or, and this is what we'll talk about today, these hybrid systems where you combine people and agents to generate disproportionate return.
You take that unit of a person and an agent together and you give it a process.
You make it own process, and you have it go out and capture knowledge and learning, refine knowledge and learning, and then compound that knowledge.
In technical terms, you could think of that as like retrieve augmented generation, but you improve the value of that knowledge as you're using it.
If you set up a portfolio of these little units of people and agents together, you can generate massive economic returns.
I'm going to dive in a little deeper on this.
But the important takeaway here is that the objective of the enterprise is no longer to maximize cash flows, but instead to maximize intelligence flows.
Unfortunately, as people are playing around with AI, these days, they're proudly touting their AI adoption rates.
That is a vanity metric.
We should not be looking at that.
We should be looking at what are the intelligence flows that the organization is creating.
Next slide.
If you go back to you may or may not have heard of something called the theory of the firm.
Ronald Coates came out with this idea organize companies and based on the German general staff model.
In the 19 th, late 1930s, he proposed this theory of the firm where you had a CEO at the top and you had this organizational hierarchy with people in different functional specialties and little boxes underneath the functional leaders.
Well, that's the old way of doing things.
In the era of intelligence capital, instead, you organize around a portfolio of these little units that we call intelligence capital generators.
And you're constantly running thousands of experiments.
Most of them fail, some of them succeed and generate disproportionate returns.
You get what's called a power law curve of returns.
It's a new way to organize a company.
Now, mind you, or any kind of enterprise.
It doesn't have to be a for profit enterprise.
Now, mind you, this idea of a liquid labor market where people are in these little pods is actually not new.
Everyone from unilver to valve software have been playing around with this for 20 years.
What's different is the rise of the human agent system and having that be that unit of productivity.
Next slide.
So what that looks like is you put a human and an agent inside something we call an intelligence capital engine.
By the way, all of this is at intelligence generators.com, so you can download the slides there and there's also a white paper.
But you have this little intelligence capital engine that captures knowledge and learning, refines it, and then compounds it as it's being used.
Next slide.
Take that engine and stick it in something we call an intelligence capital generator.
What is a generator? The generator owns a defined business process, it retains memory and feedback, It improves its unit economics over time, and this is something my colleague, Tricia Wong is going to speak about, I think on the panel a little bit, it operates under auditable control.
You have something called proof of control.
As we start to deploy these agents and these agent human systems out in the world, we need to make sure that we have strong assurance that they were doing what they're supposed to do and not doing what they're not supposed to do.
So next slide.
You take a bunch of these generators and you stick them into a portfolio of these constantly regenerating experiments.
Now intelligence capital in your firm is governed as a portfolio and the people that you have have to be trained on how to operate in this more liquid fluid environment, which looks like a well run open source community in its dynamics.
As I was thinking about that for this week in particular, in this conversation, I was thinking about in the intelligence capital economy, what is the role of open source? We are here in the United Nations, 193 countries are right now trying to navigate between Silla and Sarbtus.
They're trying to navigate between the US and China AI systems.
Open source provides a third way.
Next slide.
Open source provides us an opportunity to explore sovereign intelligence in a way beneficial for humanity.
With that, I'm going to turn things back to our moderator and I think we're going to get the panel started.
I'm David Schreyer with Imperial College.
If you go to aolaab.org, we've got a number of white papers on sovereign AI, on intelligence capital, a little more on our work at Imperial.
Thank you so much.
We will now go and explore while I go back on stage what this third way could look like and helping us explore again, what an open source can do for AI agents and vice versa, let me introduce the panel.
Sarah Hooker, she is the Chief Executive Officer at Adoption AI and former Vice President A here, where she led Queer Labs.
Sarah, welcome onstage, and we have Yang Wa Lin, she's the Vice President and CTO of the Beijing Academy of Artificial Intelligence, formerly the Director of IBM China Research Lab.
Going to the UN colleagues, Mustaf Accardi, Senior Cybersecurity solution Officer, our UNICC that is the UN International Computing Center in Valencia and the co chair of the open source Community of Practice.
Last but not least, Tricia Wang, she is the Chief Executive Officer of Advanced AI Society and she's a public interest technologist.
Of course, David, welcome back onstage.
Yes, sir.
To start with, I think we will go in the order of calling and I will invite you to share in 3 minutes some observation reflection on what we heard from David and on the topic of this panel, and then we will go into back and forth with some follow up questions.
And you should be able to date to you mind just pressing.
Fantastic.
It's really lovely to be here.
I think this is a very needed forum.
Although I just learned that Filippo, our moderator, is not being BUN and his support of World Cup teams.
Since Italy didn't make it, he's not supporting anyone.
But it's fantastic.
Let me maybe kick things off.
I like to make things relevant.
This conversation is bringing together stakeholders and people who care about open source.
For that, it's a lot about who gets to own technology and who gets to shape intelligence.
My careers as a computer scientist, I feel like I've done my tour of duty at different frontier labs.
I was at Deep Mind and then I led the research arm at cohere.
For the last ten years, the dynamic has been such as came across actually in your slides that you place immense amounts of capital and you build the biggest model.
In fact, the primary ingredient of progress for the last ten years was scaling model size.
What's quite refreshing about this moment in this panel is that I wrote a grumpy paper about this, which was late last year, the slow death of scaling.
So Scaling model size is largely, I would say, at its end as a formula.
Instead, so much of progress is going to be how a model interacts with the world.
Why does this matter for this gathering? The dynamics are completely different.
For pre training, you need massive amounts of compute, you need to co located in the same place because you don't want to distribute across data centers, and there's massive barriers to entry.
Most rate of return for compute, where you get bang for your buck is now moving to post training.
It's moving to interaction with the world, it's moving to a gene This is a way more distributed form of compute and you can get much more with less.
Why that's interesting is that we often end up in these rooms gloomy about the state of open source and who can shape, but the dynamics of shaping are completely changing.
I'd also say another reason why I care deeply about this topic of how does a gente shape who gets access is that there's another component to this.
One of the things that's really accelerating is how we do scientific innovation.
One of the things that we're working on is, how do we train models that learn from researchers how they can be optimized.
One of the biggest obstacles to sovereign AI to people owning AI is the know how, which is currently concentrated in a handful of labs.
A lot of what we're doing even with auto scientists and other projects is, how can we empower innovation cycles that are much faster outside of this handful of labs? There's a few themes there, but one, I would say the barriers to entry of people shaping innovation has dramatically changed over the last two years.
People are not going to for exer size of that model anymore.
The second is, we're actually learning how to learn, which means that we can really give people the possibility of shaping their AI outside of these labs.
But with that, let me pass on who is next? Ha, please.
Okay.
This is Yu Hwang from Beijing Academy of AI.
For our institute, we are the non profit institute focused on all the whole stack of AI research from the model to the data eeration and AI compute and all of things actually the open source.
Back to the Davis talk, I think it is very impressive and I also feel that people might notice that he mentioned Sometimes currently for AI have the two very important campus and China and US.
Actually, we also notice that we also want in the global, we don't need to do this kind of to choice, and we can benefit all the outcome from these different countries.
Currently, on one way is the model, the model from the different countries, they change the model and open source the model.
The people, you can choose any model you want.
That's one way through the open source, you already can solve that.
But on the other hand, you might also need to think about how to run the model on the different hardware and really today the computer resource is very expensive and also for different countries including the US, Europe, and also China, different country, they also building their hardware.
So on the other hand, people might also think about which hardware should I buy from which country.
We also want people don't need to make this hard decision.
That's why we also build a open source stack called flat OS.
That is to allow the different model can run on different hardware without any coding.
So we actually have the two can allow the model download and if that model changed on some US devices, then through the tool, it can very easily to migrate onto the different hardware which generates or built from China or on the other hand, the model, if it changed on the hardware from China and it can migrate onto the the US, for example, a media hardware.
That means that for the user today, you don't need to struggle.
We also don't want you to struggle and choose which country's technology because we want the AI to be inclusive.
Thank you so much.
Let's go in order, Fish Thank you.
Yeah, I think this is a really interesting time.
I agree with Sarah that there's a lot of room for innovation right now and especially when I hear about David's talk on intelligence capital and saying this is the future and we're all going to be yielding all this benefits from intelligence capital.
But I think the most important thing we have to figure out is that no one's going to want to participate in this world if we don't trust and we don't have a way to figure out how do I know this agent is acting on my behalf? That my behalf can be an enterprise, it could be an individual, it could be a nation state.
It could you know, be a government.
And I think this is the most important thing that we have to figure out.
And at the Advanced AI society, you know, our mission and our purpose is to make sure that as a industry association building out the public goods for verifiable AI, that this is one of the key pieces that are missing from the story right now because there's a lot of discussion we need responsible AI or ethical AI or AI for good.
But all this is about what AI should do instead of, well, what can AI do right now? There's all this incredible technology that's been developed in the decentralized tech space or in cryptography that is really a latent market that can be applied to making AI secure.
Because the biggest thing that's missing right now is all this hype over AI.
We're not really talking enough about AI security and the need for us to trust and for us to trust AI, verification is at the basis.
It's a precondition for trust that we have to be able to know how do we verify this AI did what it said it was going to do.
Right now, we live in a claims based AI world where you don't really know that the AI did what you wanted to do.
It's based on assertions.
That's how these models work.
And what we need to move into is a verification based world where it's based on evidence and that evidence, actually, the most tamper resistant kind is cryptographic.
Cryptography I think it's going to have its moment right now.
Usually cryptographers and cybersecurity leaders have been left out of the conversation when it comes to innovation because usually people are scared.
Engineers are scared to talk to the cybersecurity people because they're seen as a no people or locking down the perimeter.
But now I think the cybersecurity industry what their job is, is in order to make agents useful and valuable and they have to figure out how do you decrease the risk.
But for an agent to be useful, you have to be multi cloud, you have to be multi vendor.
The job right now of cybersecurity is to figure out, how do we make AI secure? One of the greatest toolboxes and the tools that we're seeing from the companies that are the innovations are coming from companies that are adopting verifiable AI, which means that you have cryptographic security at the core as one of the core pieces.
The reason why this industry is so important as a latent market is that cryptography has really figured Um, how do you? How do you verify a fact? How do you verify something happened without releasing all the information? This is going to be super important now, which is if you want your intelligence to be shared, to be used in any way, you need to be able to move that intelligence.
That intelligence has to be portable.
Your agent has to go and move into other clouds and vendors or spaces or other nations, other warehouses to do stuff.
But you need to be able to confirm the agent's identity, its privacy, who authorized it to do that? What was the chain of delegation? What was it? What's the agent's provenance? Well, all of these things are now being solved through all of these various mechanisms using cryptographic means, through ledgers, distributed means, de centralized means.
I'm really excited I think this is one of the most exciting moments for open source is to actually think about now, not just open data or open standards and open hardware, but this is really, I think the moment to shift to expand open source into thinking about open verification.
And that verification has to be open.
It cannot just be like anyone saying, Well, I can verify that because verification cannot belong to the verified.
Thank you, Mustafa.
Amazing.
Listening to David's presentations today, the one thing that struck me the most is the idea of the intelligent, um, the intelligence harnessing components of your framework.
I believe this is actually one of the most critical components that most enterprises will eventually realize.
I think every agent, first of all, I want to separate AI models from agents from anything else.
I'm only talking about agents right now.
The part that strikes me the most is true that every agent on its own working on different things, building their own memory in a way that they are not talking to each other.
But if you have that closed self evolving loop eventually, this harnessing of all those different intelligence sources into one becomes even bigger and stronger than the total of all the components of it.
So this is key, and I think this is going to be one of the most critical design patterns when you're talking about agents in the workforce moving forward.
I think there's already some of these in the market today, but this is going to become the future norms.
Another area that I wanted to highlight here is open source.
We are an open source sup, and open source is actually an opportunity to solve some of the problems of the AI world.
One of the biggest issue was the transparency of AI.
How do we know that the actual logic or the code behind the blocks is actually doing something for our benefit as individuals or also as an enterprise? Uh, so open source, basically, is transparency.
So you open the box and you let everybody inspect and see what's in there to make sure and you can trust the system.
This is key.
The more we work with agents over time, the more we need to actually trust it.
If you are not able to trust the system, you most likely will not use it, or you'll be hesitant to use it in sensitive areas.
With that, I think, those two points is key.
David, do you want to say a couple words? Just to pick up on Mustafa's point, I think the question of orchestration is going to be one of the most interesting and complex challenges that we can take on of how we manage not just one agent, but a whole array of agents in a symphony that then does great things for an enterprise, for society, for a group of people, for a bunch of friends.
I'm really looking forward to see how people build these orchestration systems in a way that is safe, reliable, and trustworthy.
Thanks so much.
Now that we have all the elements laid in front of the table, let's try to make a meal and a decent one with all these suggestion and all these elements that the panelists brought up.
Let me start with Sara that you mentioned your paper this low of scaling and now we have the rapid rise of an agentic AI.
I wanted to bring this from a user perspective, like what changes we can expect when we're moving from just humans, all of us using AI tools, to humans interacting and working alongside agents.
Yeah, maybe to kick off, I'll comment because there was some fun topics there.
I want to just briefly touch on harnesses because you know what's crazy? Harnesses can either compound proprietary advantages or they can open up.
Frankly, some harnesses that proprietary firms build don't work with any other model but that firm, so you have more lock in.
I think the question of harness is super interesting because so much of also what we're seeing now, this push, the momentum in this room is largely because in recent weeks we've seen decisions to withhold certain models.
People have realized there needs to be a diversity of options and an ability to have a pool of models for access.
I think this forum is happening at a very interesting time.
Key to that, if you care about how do you leverage pools of models, which is much more akin to our own intelligence, by the way.
We all showed up here, we all showed up at the same time.
Human intelligence is rarely individual, it's much more collective, frankly.
But this idea of how to harness it, one, I'll say this.
The fact that there is no consensus on how to do it is very interesting.
A general harness is typically built for all tasks, which means it's very inefficient token wise and it's very inefficient on your task.
Um, for me, it's a meta optimization problem.
I think that a lot of what we'll see is, how do you auto learn the right harness for the right task? I think the burden placed on the user right now is pretty extreme.
I actually think that needs to shift much more to not place in.
Ideally, a user isn't doing all the acrobatics of today like massive prompts, trying to figure out how to piece things together.
We should be auto learning a lot of that and then just really getting on with the user's task.
What is the place of how humans should interact with agents? It's interesting because it's actually pretty critical.
The two areas which have seen the most development, frankly, I'm trained as a computer scientist have been code and design.
I'll share why because those had interfaces that we were used to, that we typically use and we gave a ton of feedback along the way.
All of the tasks are completely underserved by current interfaces and they're instead presented with a chat and a thumbs up, thumbs down.
I see our vision is that the whole stack should be adaptable, but for AI to work for everyday tasks that outside of computing and design, you have to have dynamic interfaces that allow people to get feedback and someone has to feel like the model actually changes real time when they give feedback.
Right now, if you give feedback in a chat, who knows when that model will change? You have to wait for the next version.
I actually think for the first time ever, the most intelligent system will be the one that cares about interface and cares about how people give feedback.
There are other computer scientists who have the opposite approach.
They believe in browser agents and crawling how the Internet is.
I frankly don't think that will be very successful.
I think that we already know from tasks that it's where humans find interfaces useful and are intervening frequently along the way that it's very possible to build systems that are useful.
I'm much more interested in that.
How do we build dynamic adaptable interfaces? We do a lot of work on and dynamic visualizations, dynamic, tools, which is really to allow humans to be in the front row and to be in the driving scene.
Thanks so much and also thanks for raising the issue of that is, I think it was an underlying theme of capacity.
You mentioned the use of token that of course, it's like the data, like the GP.
Yeah, it's kind of like so unequally distributed.
Yung, you mentioned a very important aspect talking about stock, that is the hardware component, and of course, not just software component, it's hardware component.
Talking again, expanding on this stock that you mentioned, I wanted to dive a bit deeper on what are the technical foundation that are the most critical to make sure that you can build an open source in Agenda K models that are reliable, but they are also scalable.
Yeah.
Thank you.
Actually, what I mentioned just now is the concept about open compute.
Yeah, because we do want different model can run on different hardware.
That means that if you have more choice, then you have the chance to lower the price.
Also in today's very complex agentic system is not only the single model.
You can apply the different model, large model or smaller size model for different purposes to work together.
So that's why we think that to allow the different hardware can support the different model.
This is very important to make the Gene system more scalable and the So this is very important, the first point and the second point actually is about the reliable.
I do think that reliable, on one hand is, of course, because the information security issue, then we need to need to consider how to deploy the overall Agente system together with the model in the proprietary way.
So that's why we need to consider how to put the model into different hardware.
But on the other hand, reliable also means for how will you benchmark verify it.
Today, we do see that from the different kinds of benchmark, one is to develop the framework to evaluate the successful ratio of the completion The second is the efficiency to finish the task in terms of the time and the last is how much token you will spend on this age system to finish the task.
For different angle is also very important to consider for the sustainable and also the reliable.
Thank you.
Since you touched upon verification, let me turn to Mustafa and also echoing what Trisha was saying that we live in a claim based AI world.
What I wanted to ask you Mustafa is two different things actually, what does it mean verification for the architecture of AI in Asian ecosystems? Also, just before getting on stage, you mentioned a UN acronym that is always good to learn.
That is Bosa that is open source agentic research architecture.
Maybe while you cover the verification, you can also say a couple of words on Asara.
Yes, absolutely.
So for us to be able to have some accountability on the transactions of the agents, we need to be able to trace all the transactions back to the agent, the source of all of that.
And every single transaction has to be in a way where you cannot basically change it after and you can be able to defend it in a way.
So verification is key here.
If you cannot verify that the agent done those things at this time, you are not going to be able to trust it, and then you'll have all kinds of problems down the line.
So that's why we have to get this right, the identity components of the agent.
This is something that everybody is talking about.
If you've seen any futuristic design or proposed design for the new harness, it's always going to have that identity component.
Also, going back to what Sarah was talking earlier about the human plus agent as a new operating model moving forward, I genuinely believe that this is going to become the reality, not because there's a choice.
This is the best actually.
It's not a choice.
After going through a whole lot of research in this, by the way, I come from a very long history of software architectures and cybersecurity framework development in the past.
So I totally understand what the value is for identity and being able to verify that and capture it over time.
Coming to OSRA as a part of the work that we are doing in the Open Source United is a community of practice of the United Nations responsible for open source projects and to facilitate all the work across the different agencies.
One of the initiative that we just have is called OSRA's Open Sovereign Agent reference architecture, is actually a two tracks One is a specification, documentations explaining what are the main components inside the hardness of the agent that must remain open.
There are others that it could be open and it could be restricted or closed, whatever you want it.
It also separates the intelligence source all the AI models outside.
It treats it like a battery in the car.
You can swap it out anytime you want.
But that's not going to change the functioning of the actual agent itself.
It also comes with all the protocols and the different layers of the agent.
We believe that the person is going to have their own personal agents, the enterprise is going to have their own agent, the brain that will be functioning on their behalf, eventually also a G agent where it will function on behalf of all the different government entities.
We had a roadmap, like the entire thing is probably going to be become reality within the next two or three years, maybe longer.
I don't know.
But the other part of that is actually a documents called AI Bill of Rights.
This is documents almost like a policy or guidance, if you would.
It explains all the roles that the person will have to do to take care of their agent because that's going to be representing them.
Also for the enterprise, what responsibilities they are, if they're going to allow a person to bring their own agent to work in the future, what are the different things that they have to ensure as far as compatibility goes? Then, and most importantly, the role of government in this area, how are they going to facilitate and what are the building blocks that they must have at at the country level in order to have or facilitate that ecosystem between all different types of agents.
Some of the aspects of it already exist today, but we were able to put all that together in a way that it gives you the design of the car and also the roads and the dimensions and everything in one go.
This way, we don't have that chicken and egg problem that we have in right now with regulations and design.
Thanks, Natrisa going back to you, Mostafa, raise an important point that is eventually the question who is responsible for the actions all sorts of agents will be taking.
Tricia, you cover the dimension of safety.
My question for you, what we're hearing the conversation at the UN, from member states, from the private sector, from stakeholders and we are talking about open source and accountability and transparency.
So there is almost a manic view.
I either open source automatically creates accountability through transparency or Open source is terrible for accountability because of transparency.
If you could elaborate on these.
Yeah.
I mean, this is a problem that's been solved by the open source world already.
I don't know why we have to keep litigating it because for example, enterprise solved this.
There was a time where enterprise said, Oh, we don't want to use open source because we haven't solved for the security stuff, but the whole point of open source is that we are able to inspect the code and how something was built.
But what enterprises had were their own security concerns.
And so, OSI, Red Hat, we have Linux Foundation, everyone came together and then we figured out open core where you could fork it.
There are ways and we also learned then that it is not binary.
It's not like if you build open, you can't have things closed.
We know that we can build closed on top of open.
I think the challenge now is that we are in the open source moment for verification and that what many people did, important open source leaders did in the late 90s, is that they came together and they said, Hold on a minute.
Anyone now saying everyone is saying they run open source.
That term is being used so loosely and this is history that Brian Bellendorf knows well and taught me.
He's one of our advisors.
He and many others saw that we have to come together and actually give a definition.
There was a definition that's already out there, but the community had to come together and say, this is what's open source, yes or no.
And then they were able to figure out how to institutionalize and create governance around it.
Now, we're in that moment for verification where anyone now is increasingly starting to say, Oh, we do verification.
But really, what is verification? How do you know that, you know, does verification for us, I don't think it counts if it is not independently viewable of the operator because you cannot trust that verification process.
And so this is super important that right now we're in this moment where if we want to be able to trust what agents did, we need to be able to see evidence, but we need to be able to understand the nature of that evidence, and we have to be able to trust it.
And so we have to figure out how to define in this moment and I'm excited to talk with Mustafa and my other panelists about this, is that how do we define verifiability right now in the same way that open source leaders came together in the late 90s and took that messy landscape and gave it a bit of order to make it publicly definable, commercially viable and procurable and legally defensible.
We're in this moment now.
It's a little different and that open source was easier to define.
It was like either yes or no.
You either met these conditions, it was binary You're open source.
Now, there's stages.
There's different types of verification, there's self verifiable, there's mutual verification, there's independently verifiable, there's cryptographically verifiable.
There's centralized means to do cryptographic verification, there's decentralized means.
We have to as a community come together now in this moment and this is why at the advanced AI Society we've launched most often we're talking about this of the beauty of ASARA which he taught me that the word means essence.
The essence in Arabic.
Oh, so beautiful.
What we're doing at the society is called the proof of control architecture, which is the technical foundations for verifiable AI to be able to tell how do you confirm an AI did what it was supposed to do? The purpose of this architecture is that we have to be able to define this in a way that makes sense for what Sarah was talking about, which is You know, right now, we have the messy part of figuring out how do we use AI post deployment? And this is why we're in the time of AI governance.
This is why AI policy is so important because we have to figure out how do we actually trust these things if we're going to make AI inclusive, like a was saying, right, is that we have to develop governance around it, of how do you combine different AI models or how do you figure out these tricky questions of privacy and protection? And I know we can do this because we've done this before.
We give power, we trust agents all the time.
Like when you get on a plane, I don't know how many people came on a plane here, but we have developed systems of trust and verification To entrust someone else to be an agent on my behalf.
Every time I get on a plane, there's already been, I think, whatever certification training process that the whole industry has gone through.
Then if something fails, there's accountability, there's insurance mechanisms, there's a way that the industry has price risk, and we have not done any of that yet.
It is literally being figured out right now.
I think a lot of times people are very critical of AI being like, Oh, the industry hasn't figured out itself out, but things are moving quickly.
And we are figuring things out right now.
We are unfortunately flying the plane and trying to build at the same time.
I'm not saying it's right, but it explains, I think some of the messiness and we should be realistic to say, look, we haven't figured out the messy and boring stuff around procurement, around actual science, around this, how do we price risk and getting the insurance industry with all these agreements and governance and nation states involved.
So I think it really makes sense right now where we're at.
But really figure out safety, we have to agree verification, and we have to agree on what open verification looks like.
Otherwise, everyone's just going to claim it, and we're not going to get any further along with, how do we actually trust that an agent acted on my behalf.
So much.
I guess it doesn't harm to repeat some things that are obvious.
I think that all of you being here, you're making the case for open source.
Thanks so much for that.
Turning back to David.
Again, Tricia, something else that you were saying is that, again, the conversation is always about what AI should do rather than what AI is right now.
It pains me a bit to ask this to David, but based on what we're discussing, what do you think talking about agent based systems, what do you think are the main challenges, I would say medium term over the next five years.
That again, It's an entirely different geological era in terms of AI, but what do you think are the main challenges out of us? Well, I'm going to slightly reframe that.
I'll talk challenges and opportunity.
The main challenges that we're facing, one of them is a concentration of know how.
There are maybe a couple of thousand people in the world who know how the most advanced systems work, and most of them have been captured by eight companies.
And so what can we do to bring more skill and knowledge to a greater array of people away from that concentration of know how? More broadly, one of the big challenges is we have a hyper concentration of intelligence capital in the hyperscalars, right? So today, as of close of market yesterday, they were worth about 25 trillion US.
By 2033, my forecast is they're going to be somewhere 52000000000000-80 trillion of market capitalization.
Um, and so how can we sort of pull out more value for more people rather than have it be in the handful of a few oligarchs? And so sovereign AI is, I think, part of the answer to that.
And so I'm very excited.
This is in the opportunity space of how can we bring more people working on sovereign AI out to those 193 countries and unlock the intelligence of entire populations of entire countries.
India, for example, is acutely aware of this.
India rode the outsourcing boom and did very well off of that, and now is facing the outsourcing collapse and from the federal government down to the individual states is making a concerted effort to retool the economy around intelligence.
Pakistan, also, my former student, doctor Sahil Mouner is chairing the National AI Council as they try and reorient the entire economy around intelligence.
In fact, 50 countries at least that we're tracking are building sovereign AI initiatives from Sweden to Singapore and more should be getting in on the act.
And one of the promises of sovereign AI is that you can have the intelligence of a population instead of all that know how and knowledge and that increase in economics that I've talked about going into the hands of an individual hyperscalar magnificent seven plus one, it'll now go into the hands of the people.
That's a really exciting opportunity.
As we conclude, let's go rocket fire.
I want to add one opportunity that I think in order as a mechanism to get to this world that you've outlined, David, is that we have to get the opportunity and the challenge is we have to get over this mindset that you primarily use only hyperscalars.
But I also think on the other end, all these responsible AI and ethical AI critiques of hyperscalars to tell everyone to get off of them and that they're bad and evil.
I don't think that works either.
I don't think you can penalize people for wanting to use tools that just work.
Sure, you don't have enough transparency, but you can't shame people.
One way to get people there, and I think it's a mindset shift is that we're missing the design mechanisms that we're implementing advanced AI society is that we talk about this concept of a portfolio AI approach, and in the same way that you would approach your financial portfolio as you don't put all your eggs in one basket and just one company Well, what's exciting about companies now that are coming up and that's why I've been looking at so excited when I saw Sarah was on my panel is that I think companies like hers and you're going to see more is that we're now past finally, okay, we know there's the frontier AI lab and hyperscalars, but there's actually so many more companies, partially because of open source has allowed this, but a lot of great entrepreneurs are starting to come up where they're making all these other companies for us to choose from so that we can take a portfolio AI approach and we can say, Hey, based on as a nation state or as an individual or as a company, based on my data needs, my regulatory needs, what my use case is, what my purpose What is the best portfolio for us in this business context or in my daily life or use case and how does that change? We're going to see a lot more different kinds of AI applications.
It's going to look more inclusive.
People are going to be mixing models.
I think we need to really move out of this shame based idea because that's not going to work and there's all these companies like Sarah's and others that are coming companies in China that we're just going to be able to choose from around the world and everyone's going to have more choices.
Thank you so much.
We're almost at time, but I wanted to check with the panelists if someone has a quick.
No, you must.
No, you must.
I love it.
Everyone is actually ready.
I'm sorry.
I just wanted to highlight one thing here that actually when you architect this new era of agentic AIs, you have to have cybersecurity in your mind, software development in your mind and architecture, and also cryptography at the core of everything that you do.
If we don't get that right from the very beginning, I don't think we'll be able to get another chance at this for another 30, 40 years.
I was really commenting on everyone had their finger at the ready.
I'll only say this.
Honestly, gatherings like this were very depressing a few years ago.
Most of the conversation was around, open source should focus on evals, which was absurd.
The conversation is entirely different.
The mechanisms by which we can own intelligence have dramatically changed.
Part of that is a shift of the rate of return or compute, but part of it as well, we're starting to automate AI training itself and AI optimization, which means barriers to entry are totally different.
I think that's the main reason there's so much momentum in this room and it's a totally different environment from even a few years ago.
It's the right time for this forum.
David, go for it.
Yeah.
Very quickly, since open source was pioneered, let's argue in the 1980s, consistently what we've seen is on the one hand, people are always going to be paying for state of the art models that are closed source, expensive ones at the top of the pyramid.
Open source brings in more access and allows more people to take advantage of the benefits of technology.
It's important that we continue that work.
Since today's discussion is about open source and also in this United Nations building, I want to say today actually, if you look at the overall different countries, the computation installation, 90% of them might be already out of date to support the latest model.
So the goal for the open compute with the open source is we want the AI AI agente system can run on most of the computation resources which they might already under utilized.
That means that we want to use those underutilized resource deployed in the data center to support the AI and also AI Ante system so that we can have the chance to let the whole global different country, developing country, they can benefit from that.
Actually, in the past one year, we worked together with the United Nations and also the African Union to try to try this open compute for AI education for talent and we provide this underutilized free resource as the online resources support Africa Universities, they can launch the classes to teach their students.
That's very meaningful for them to get our next generation ready for AI.
I think open source and open compute are very important for us to consider how to expand the AI capacity and make AI inclusive.
Thank you.
Thank you so much.
I know we are running out of time, but just checking the room.
I wish that the Q&A was slightly longer, but that's the time we have, we have time for one or two questions.
If someone wants to take the floor, please press on the mic.
I should be able to see the light and let me see if I see any.
I see one over there.
So my question is primarily around may I ask you to introduce yourself? Sure.
Thank you.
Sure.
My name is Mirza Bag and I am from Kindrel.
My question is primarily around open hardware.
It's really great to hear about cryptography on this forum, and we've been talking about software quite a lot.
I would like to hear about open Cloud, open hardware, and what kind of open stack is available out there and what the committees are thinking about it.
Thank you so much.
Any other intervention I'm I'm seeing let's go to my right.
Thank you so much for a wonderful session.
My name is Habib Norby.
As a professor back home in South Africa.
In the context of research methodology, a lot of the data that we use, if, for example, it is not valid, we would not classify it as reliable.
In the context of open source data, my question would be around the fact that if the data is accessible, if it's open, and even if it is verifiable, how do we ensure that the data that we are having access to in the open source context would still be valid or reliable in a similar context to research methodology.
Thank you.
I think we might have time.
For the last one.
Myself, Rad Shaker, I work for a company called Fraznius which is healthcare plus med tech company.
My question is regarding it's open source AI.
But more and more countries are going in the direction of data residency.
But when you talk about open source AI, for AI, the bedrock is the data.
When countries are going in the direction of data residency, you guys want to or at least I would like to get the opinion of the panel how it works.
Thank you so much.
Teresa volunteered, I think for open hardware question.
I can take the first one to answer the question of what's happening with open source or verifiability with hardware, is that there's a lot happening right now and it's a great time for cryptographic works one of the things that we're seeing with verifiable AI, some of our members, is that can now do, especially for sovereign AI or any sovereign AI needs, is you can do provable compliance.
This is a great time for governance, for policymakers, for anyone who's trying to figure out how do you figure out where you are in the compute in your queue for trying to get compute? How do you prove that your data was residence in this specific geography? How do you even prove that your data was computed onto this very chip? Right? And so we have members from Lucid to equity who are figuring this out, who can build out the entire data supply chain and you can track cryptographically just tamper resistant, and you can check it's independent of them.
You can confirm that this data was computed here and on this hardware.
So I think there's a lot that's happening in that area in terms of the intersection of hardware and software.
So that's one example.
I have one more comment.
Actually, today, the software that's already there called the Flat OS Act it is built together with tens of different research institutes, university and also the enterprises together.
It already can support more than 330 different type of hardware, and we also collaborate with the open source community like the Linux Foundation, PyToch Foundation, Eclipse Foundation, this together.
Today I think for the open compute or open hardware, I think we are ready.
Thank you.
I don't know if someone else wants to come in from the panel.
The last question the answer is you can't.
There's no way.
Just to add to that, I don't know if you noticed, but the US issued an order that says, we don't care if your data is resident in your country.
If it's operating on a cloud run by a US company, we can look at it.
That's a problem.
How we work forward from here is really messy and the data ownership, data flow data providence question is one that we all need to work on, but I'm not going to even pretend that people have begun to solve it because it just got a lot messier.
This brings our panel to an end.
Thank you so much to all of you for being here, to the panelists for the great presentation.
We wish you all the best for the rest of the week.
We have a fantastic program, great receptions and the UN conversation on AI will also continue not only in New York, but in Geneva with the first AI dialogue that will take place on the sixth and 7th of July.
I want to acknowledge the Estonian minister and in the room, Estonia and El Salvador are leading the efforts of the first dialogue on AI governance and open source will be part of that conversation as well.
Thank you all and on to the fantastic MC Pepiv to transition to the next session.
Thank.
Thank you so much.
I love whenever we're doing tech events and then there's technical issues.
I think it's quite ironic.
Thank you, Filippo, for finishing us off on time.
That's a miracle whenever you attend UN events.
We're going to start off in a minute or two with a very exciting closing session.
So let me hand it over to our dear colleague, Medy again.
You saw him earlier.
He graciously stepped in to moderate a previous panel, and he'll be back onstage to introduce the speakers of the closing session.
So over to you, Medy.
Colleagues, guests, I'm very happy to wrap up this open source first open source and AI day at the UN and we'll be discussing in very formal, informal chat with firstly, UAG Ame Gil, the Under Secretary-General especially envoy for Digital and emergent Technology, Mr.
Alberto Caggio, the Director General of the Spanish agency for the Supervision of Artificial Intelligence, Asia and Her Excellency, Lizlpcosta No.
Ts and Pakosta Minister of Justice and legal affairs from the Republic of Estonia.
Sorry if I'm misspelling any of your names.
I invite you all to join me on the discussion.
Microphone are not working, so we are getting the discussion in a very formal setting for a very informal discussion.
While waiting for the speakers, we'll make it very short.
It will not be a very long discussion.
Thank you so much.
Oh, okay.
Oh, I don't have the list.
Yeah.
What? Okay.
I forgetting, Er Excellency, Gabriela Gonzalez, the DPR of Uruguay.
Very sorry.
I'm using all notes.
I need to update.
I need to start using Google Docs or whatever.
Things that get updated automatically.
Today we got a various sessions all along today.
We've been talking about robots.
We've been talking about the future of AI with the opening keynote.
We've also discussed the development of AI, and then we got a first of kind panel with only women minister showing us the pathway on how to develop the deal in real way in an effective manner.
Uh, the three of them are very excellent and bringing some concrete results in their country with very impactful, tangible outcomes.
We discussed the robots, we discussed, the agent and all these very complex and technical subjects.
Now, we'd like to wrap up this discussion with some very insightful outcomes from all of you.
And then I'll start with a very simple question for all of you.
When first you heard about open source and AI, what is the immediate things that you are thinking about? I'll start with our Spanish colleague, please.
Open source and AI in few words for you.
Up to me it's openness and transparency.
I think those are two major values that we need to include in our development of AI.
I like the previous speaker also said that there's nothing wrong with companies trying to look for their own economic benefits, but there's also nothing wrong to ensure that these economic benefits are also supported by a transparency and responsible development of technology that protects the rights of people.
Boss, please.
Very quickly, for me, it is the usability, the reusability, and the community dimension that you have a community that has your back.
Your Excellency.
Thank you.
I fully agree what has already been said.
From Estonian side, I would add that it is a governance issue.
It's not a technical issue.
Open source is something that you have to put on the table as a default solution from the governance side.
The only then you can fulfill all the values and goals you are looking for.
Thank you.
Thank you so much for giving me the floor.
For us, I'm going to repeat also what is being said by the previous speaker, Madam Minister and the USG.
For us, it's not only a technical principle, it's a strategic tool for trust, inclusion, innovation, and international cooperation and also open standard, open technologies, and collaborative governance that can help bridge digital divide across the countries and regions from developing and developing countries also.
Thank you.
Thank you so much and I'll stay with the bridging the gap.
I know how involved you are in making sure that no country is left behind.
There is a big effort on the multilateral level, the global digital compact, the Pact of the future, the global dialogue, the panel.
There is a lot of efforts happening at the UN.
From a member state perspective, do you think that open source is getting enough space? How do you see that should be enforced within the multilateral, either within the multilateral or how the multilateral could enforce the open source? Well, of course, It's never enough what we can do to get bridges between regions and using the open sources.
But I think with the work that the audit is doing is good.
We are going to the good path.
Of course, we can improve, but for that, we need many mandate, we need also support the offices.
We have to create also the tools in our own governance at the UN for improving our work with us the member state.
But indeed is needed, and I think we are in the right way here working with the multilateral system.
Thank you so much, UAG, about the mutilar ADT was cited.
What do you think AD can do more? Well, get the tech to work better always in the room with our colleagues from OICT.
But I think today has been a good day, the quality of the discussions, some of the new topics that have come in, robotics, for example, and the fact that we didn't lead with a policy heavy agenda, we allowed the community to surface its own insights, its own concerns, the community of technologists.
So in a sense, we successfully close the gap between some of these high level considerations and multilateral forums and what is happening on the ground in terms of the work of the people actually building those systems.
So I think that is the big lesson for multilateral forums at a time of rapid change and questions about value add that we need to bridge that gap between the abstract and the, uh, uh, practical.
So, you saw with the turnout today, very strong member state turnout.
In the morning, we had a panel with three women ministers leading the digital transformation in their countries, Jamaica, Morocco, and Sierra Leone.
That tells you something about the potential, the energy of this uh uh week and we had the keynote from n Lakh who I thought gave us a lot of food for thought in terms of what's the direction of the technology, the different angles this trajectory could take.
A very powerful lesson was that we should have hope that things could default to the small, the smart, and the distributed way of doing things, something that the open source community knows quite well and I totally agree with the distinguished ambassador from Uruguay about the need for open standards, open infrastructure, collaboration to be catalyzed through more targeted work, clear mandates, and adequate resources for the UN system.
Thank you so much, Suji.
Excellency, at the multilateral level, we need and it's clear that more effort is needed.
But what's at the member state level? Yeah, I would say it's the same.
Member states need interoperability within the technological world, and that means open standards and that means open source.
It's simple as this.
In Estonia, we have had a fully digital state already from 1990s and we know very well from our experiences that the system we have based on X rode, that is an open source, of course, that the interoperability allows us to go further on with whatever AI solutions, but we have to be not vendor locked.
That is important for every other country as well because we are a small country, but whatever country in the world needs this interoperability and open standards.
Let me give you just one example.
In Estonia, we have e voting.
We just celebrated 20 years of that.
National voting and local voting is based on the technological possibilities and even this is basically open source audited by international independent auditors and that shows the possibilities we have with open source and open standards.
Thank you.
Thank you so much.
Mr.
Alberto, we discussed about open source member states, multilateral, UN, But at the end, we need to implement that on the ground.
I know that you are leading a big effort in Spain around that.
What does it really mean finally concretely to bring that on the ground? Let me also compliment the previous speakers, but I think the first of all, is that open source, it's a public policy or it needs to be a public policy in member states.
When we are developing technology that it's basically serving the interests of our ministries, for example, we need to do that in cooperation with open source communities because that's the way to bring back control to what is happening internally, digitally speaking.
And And then we need also to ensure that the say the public sector is also putting forward resources for these technological communities to get together to create more things.
One example of what we have done in Spain, we have created our own AI model, which is trained with almost 20% of data in Spanish in order to counterbalance, let's say, the power that it has commercial models from big tech companies that are mainly trained in English.
This way, we are able to somehow also reflect the values, reflect the history and reflect the nuances of our language, which is Spanish and it's also spoken across the world by 700 million people.
So it's also encouraging member states or states to put forward the resources that can help these communities create more AI system.
Then if you allow me, I think I need to commend the work that you know that is doing on this open source and bringing international community on open source here because you guys matter.
What you have to say matters very much to the multilateral system and to the international discussions that we are having.
I mentioned in my previous intervention, we are gathering in Geneva about the global governance of AI.
What you have to say here needs to be also translated into what member states discuss, into what ministers discuss and heads of the government of the state.
It's important that we do know that we close the gap we build bridges and this is a great work that you're doing and needs to continue.
Thank you so much.
We heard about these different experiences.
I'm looking at the Uruguay experience.
Is this a very solid approach? It sounds like very solid approach.
Every country brings their own experience in building that.
How Uruguay is building experience? Is there any digital cooperation that is framing that experience, building up on that open source inside of the digital transformation, digital cooperation, how it's being framed in Uruguay.
Sorry.
First of all, cooperation is always necessary.
It's an umbrella that we have to work with between member state and also with the civil society, with all the companies, with the private sector.
Cooperation is not only needed inside the government, and with other member state, with you here, it's very important that there has to be a back and forward feedback because we have to talk about an ecosystem and not about isolated.
A member state isolated have no sense because at the end, what we want is the benefit of our population.
Development, more resources, more connections, and technologies and emerging technologies, AI, and that are a challenge for us, especially if I'm speaking from the global South.
Talking in national capacity as Uruguay, we have create an agency to deal with this matter that is dealing with digital agenda that include digital government, open government, cybersecurity, data governance, digital citizenship, national strategies on AI, data and cybersecurity.
We also understand that digital transformation is not a journey that countries should undertake alone.
As I said before, Uruguay has sought to share its experience through international cooperation, including South South and triangular cooperation.
In this regard, we participate also with our agey of cooperation itself in initiatives such as agreement between Uruguay, Dominican Republic, and Spain financed by the European Union, which support the creation of the ICT Observatory in the Dominican Republic, the strengthening of Uruguay An observatory, drawing on the Spain digital development experience.
We also work with our agency with NDP supporting implementation of the digital transformation of our government, our state, reflecting the importance of the partnership between national institution, international institution, cooperation agency, and the United Nations system.
And this experience show our openness that openness is not only a principle, if not a method of cooperation, and I don't want to steal the time of the speaker.
But that is more or less a general idea how we are working and our conception of cooperation in this area.
Thank you.
This sounds very inspiring for the Global Digital Compact and the digital cooperation work we are leading.
I hope that this is a good example to be taken into consideration and showcase it because frankly, this is the first time I hear about that work you are doing.
It sounds very great.
Let me look at my colleague from Spain.
You heard about the cooperation on that.
It's coming, it's being implemented.
The digital cooperation, it's clearly the path for accelerating the uptake of the digital transformations and AI.
Is the open source also beyond the digital cooperation catalyzer for the digital cooperation? Absolutely.
I mean, the model I just mentioned is open source.
It's available to everyone.
You can download it, you can improve it, you can upload it again with your improvements.
You can specialize it in different sectors.
This is something that we need to get the communities to work on because it's a public interest.
It's value for everybody.
In particular, Spain has been working with all the other American countries in order to ensure because you see, Spanish has very many variants.
The Spanish of Uruguay is not the same as the Spanish of Spain or the Portuguese of Portugal is not the same as the Portuue of Brazil.
What we're trying to do is to ensure that the mothers of the system, they can somehow build on our history, on our tradition, on our nuances of the languages.
But also that we need to go to the concrete level of the problems that people have in their daily lives, meaning restaurants, meaning shops, how they are adopting AI, how they can create AI systems and the open source community can bring a lot of value into it and at the international level as well if you allow It's also important that there is more and more cooperation between countries, that the UN serves as a meeting point for all of us.
For example, Spain has supported, they know that in the creation of the laboratory called AI governance for humanity that is working on a number of areas for the global dialogue on AI, but it can do much more.
It can go into concrete solutions for countries that maybe don't have compute capacity or maybe they need extra support, and then this can be somehow an enabler of creating more AI systems that solve real challenges of countries.
Thank you so much.
Before going to USG, I'd like to question the Estonian Minister on.
You've been citing the XRP project, one of the most famous project in Europe when it comes to the data change and it was replicated across Europe and then at some point it was even a question of upscaling it at the European level.
I know the challenges you face on intensive data.
We are talking about AI, but AI with no data, there is no real discussion.
Open data was one of the light motive of building the rode.
I'd like to hear about all the bottlenecks, mostly on bottlenecks issues that you can see happening now that we are moving into the AI era with the data exchange, which bringing a lot more value into the data.
How do you think the future of the AI extrudes? Okay.
XRD is an excellent extremely cyber secure way of data interoperability and it is perfectly suitable for AI solutions because it has not only a pipeline between different databases, but it includes computing capacity within this pipeline.
Estonia uses already more than 200 government services based on AI.
But you asked about bottlenecks and we have always been very open about the mistakes we have been doing throughout this long story.
If anybody is interested in mistakes, then you can just ask Estonia and you get good lessons from us.
One of the lessons was that we have actually stopped using all the sensitive data.
Before the sovereign solutions with AI because the data protection comes so much important and turns to be the absolute most important point that there should be no backdoors, the full sovereign solutions for the sensitive data.
The second lesson learned is something I think we face all together.
This is that in the democratic countries where we value very sincerely the rights of authors about their history books and whatever other information, we actually see, for example, the Russian bots spreading very intensively the false information and the propaganda to all the ums.
We are We all democratic countries are facing a situation where the bad guys are filling the AI systems with the propaganda and false information that might affect younger generations on a totally wrong way and the good countries who are protecting the office rights and all these things which are very important.
So the LLMs are closed.
I was listening with all the pleasure, the Spanish example about your history and culture.
If we can't find together good ways to fill in the LLMs and the systems with accurate information about cultures, histories, narratives, identities, and again, about histories, then we are losing our values, we are losing our most sacred feelings and understandings to the bad guys and throughout the history, this data falsification has always been used to prepare the great wars.
This is one of the lessons learned that we have to act in a different way to feed the large language models with not only different languages, but also the histories, the cultures, their sounds, their voices, their pictures, the colors of each cultures.
This is one of the reasons we very highly appreciate the work that you are doing at the United Nations because this is the very right and timely way to operate.
But sorry, now I had such a negative information.
You asked about bottlenecks.
I will end with a positive note.
Because we are so widely using open source in Estonia, we have been able to first to protect our cybersecurity is extremely strong and this is because of open source.
This is one of the values.
The second thing is we have had the possibility to export and upscale our solutions to 140 countries.
140 countries in the world the governments use at least some components of our open source.
This is really that we are very proud of.
Thank you.
Thank you so much.
Very insightful.
U, you heard there is a multilateral happening at the small scale.
How to take that upscale it at larger scale and ensure that no country is left behind when it comes to this global digital cooperation.
I think the Global Digital Compact called for these partnerships, this networked multilateralism.
It's interesting the examples that were mentioned.
In the old days, we used to call them South South and triangular cooperation, south south north Spain, Uruguay, Dominican Republic and others.
The digital and emerging technologies turn is making such partnerships more interesting and more possible.
Uh, so a concept from the past is being reborn in a good way with this, uh, technological, uh, turn.
And that's why, you know, uh, some of the other examples you could, uh, think of, you know, uh, Estonia's X road, it traveled through the, uh, technical community to a much larger setting, which was India.
The data exchange layer in the India stack is inspired by XRde and the carrier of that was the open source community, the coding community.
These kind of partnerships, these unusual partnerships, and sometimes some of this flows back to the global north from the global South.
So those partnerships are becoming possible now.
They're encouraged by the Global Digital compact.
And I think for AI, open source AI, open development AI in particular, they will be crucial because computer is scarce.
It is locked up in a few geographies.
It's costly.
Talent is also highly concentrated in a few places, countries, whether they are rich or poor, small or big would have to network to be able to access compute, to be able to develop their own talent and perhaps work together on datasets, pool datasets, share use cases.
If something has worked in Estonia, can it work in another place? If the Spanish supervisory agency's experience with startups is instructive, for another country.
These things can travel through these networked approaches and the open source community can act as a conveyor belt between these different geographical centers.
Thanks so much, Suji.
Yeah, indeed, there is a lot of to be done and hopefully we can fill in the gaps and then when it comes to the AI, avoid the same issue that we faced when it comes to the digital transformation.
I don't know a lot of about the digital strategy of Uruguay, but I heard that you are also collaborating with Spain and the Dominican Republic.
That is an international cooperation, it looks like it into the DNA of the digital strategy, but the government strategy.
How does it impact the decision making process, this design of the strategy.
What is the component of international cooperation that is taken into account while building all these national strategy, meaning being under constraints of sovereignty, being under constraints of lack of computing and all these things, how this is taking into consideration? That is a very wide question.
But I can tell that the government, we have a central role in a role to play.
We are not only a regulator or we collaborate with these ecosystem, public investors, standard setter, and major user of technology.
Through procurement, public policy, international partnership, as you mentioned, government can steer innovation toward the public interest.
This shows that the openness is not only about sharing technology, but also sharing knowledge, institutional models, and lesson learned.
In the case of Uruguay, because everything is connected, and I know many of you are from private sector.
Unfortunately, I only see on you open source AI or something like that.
Next time we should know better who are all of you in order to also interact better.
But for example, in the case of Uruguay, we export services We work.
We have a very small country in the South for those who doesn't know between Argentina and Brazil.
Our trade is based mainly on food, but also we export services to around the world and especially to the US.
We have to collaborate with the private sector creating the infrastructure and the cooperation in order also to have the capacity building assistant to improve that sector.
Everything is connected.
We have a role to play as government, to give the regulations inside the government, but also collaborate with our partners in the region.
But also since we have developed for the last 30 years, a good private sector taking the hand of the government both together, we can export this model in the region.
For that even in the South, we have different stage of development in this area.
Sometimes even we are not so developed as European Union countries or in the level that Estonia it is, we can also share our experience with countries like Dominican Republic.
Also, we want to do it with countries in Africa to start to do interregional cooperation.
But for that, we also need partners could be member states like Spain, could be the UN system itself.
We also have a lot of cooperation with European countries, even with the EU at itself, but also with different members.
I think we are doing We try to do not just in general, with all the technologies, with the whole umbrella, all the emerging technologies, including AI, of course, and of course, regulation also trying to have a international regulation of the cyberspace, taking account with Excel mentioned before the bad use of these technologies.
Thank you.
Excellency, thank you so much.
That's very insightful and very, very impressive what you've been saying on this openness from the way on Africa, the Caribbean region, South America.
But I think that also the notion of South is a little bit moving from the geographical definition, seeing some of the north country coming to the south because when it comes to the AI, they are very delayed and then we'll welcome them at the south in the definition.
But that's very impressive to see how the digital cooperation is a strong part of I want to add something, please.
About what you mentioned, developing countries, we must ensure that we are not merely consumer of digital technology, but also active participants and contribute and digital solutions.
Even we know that not the main countries involved, companies involved in AI are not.
I think with my fingers, I can count.
We can collaborate and do partnership with those countries too.
Yeah, we need to start building that indeed.
It's an imperative.
We will be only consumer.
What we consider today as very scarce and very available like the AI, probably will be much rave in the future and the access will be very limited to all the developed countries, and then we should not reach that point.
From we get back to Estonia because I know how advanced and beyond the Este have been many attempts also into building the DPR stack, building the open AI.
There is a lot of lessons to be learned from the Estonian experience.
The EVT, I remember when first it was launched and then all what open it behind it, meaning the ID and all these elements and the security and the cybersecurity.
If we need to pack the stack or that vision and make it open.
If we take the experience of Estonian and say, let's make it open source.
What does it require? What does it need? How to take it forwards, but also how to bring member states to adopt this kind of approach or your solution or the way you have designed your solution.
Thank you.
As I said in the beginning, I truly think we are talking about the governance issue, not the technological issue.
That is where the agreement to find the common space between the member states is the work that United Nations can help us all to find the common ground on that and also to understand that open source is very much not to be vendor locked.
What you very rightly said from Uruguay that really we have a lot of countries that are not extremely rich.
We have to serve our people in the best way.
This is globally extremely important that nobody is left behind and that we all have the opportunities to use the artificial intelligence for the good of the people.
Otherwise, the speed is so high that there would be a total loss if we can't solve this issue.
It is very important that we find a common ground here and open source is definitely useful for every company actually who does it.
This is something I think we can learn from Estonia because Estonian digital path has been built knowingly from the very beginning already from 1990s on the open source.
That has been our political decision throughout the governments that open source is the solution by default.
This is a political decision.
If we can have this decision spread and made a little bit better example for public procurement in all the member states, then we have a solid ground not to have a vendor lock.
And that already means that we have a solid ground for interoperability of whatever services whatever company is producing.
It is economically good for the business side actually to go on with open source.
That is our lesson.
Because of open source, you have a possibility to export to upscale.
Thank you for this wonderful example from India on the X road.
Is exactly the example.
If you use open source, you can upscale from a country with 1.3 million inhabitants to your country, which is slightly bigger.
This is the economical benefit that is there.
Of course, we all understand the social benefit, the political benefit, et cetera, but there is a clear economical benefit as well.
Thank you.
Thank you so much.
Alberto, just a technical question.
I know how Spain was behind the AUAI Act that is stronger in building that framework.
What is the part of the open sourcing the AUA Act, meaning the audience would be curious to know about that and me too.
Yeah.
No, for everyone to know, the EAI Act took for the political agreement, 36 hours, 100 people in one single room in order to agree on all the obligations of the regulation.
I was lucky to be there with my Secretary of State in that moment, Carl Martinez who then came to be one of the corporate co chairs of the UN high level Advisory Board.
On the regulation, since you are putting forward this, I believe there is a message that I would like everyone to take home and is regulation doesn't stop innovation.
This is something that is important.
If you do regulation properly, innovation doesn't need to stop.
What we did in the UA Act, it was not to regulate AI, it was to regulate the uses of AI that significantly impact on the fundamental rights of people or on the security of people or on the health of people.
In that regard, what we finally regulated is ten, 15% of all of the AI systems that are put in the European market, which is a very targeted regulation and we believe it's a very powerful one.
When you come into the open source, open source in general is excluded from from the regulation as long as you don't put it in the market.
So why is it excluded when you put it on hag in face on JiHub? Because there is transparency possibilities.
There is auditability possibilities to do on the weights on the model on how the system is operating.
That is giving this extra, let's say, transparency on this extra support in order for people to be able to work with something that is truly open.
Then another thing is when you create that as a solution and then you place it in the market and then that is actually affecting individuals, in my if it's in a high risk sector, the obligations will apply will start applying.
But because there is so much transparency, you will understand better how to construct the system once you put it into deployment.
Because the problem and this is another problem that we see is that, for example, with models, the big models that are now available, to what extent can we ensure that there is no bias in how the model was training with that data? If you don't have the access to that information, how are you ensuring that your system does not discriminate? So this is an important aspect that we looked into when we regulated AI in Europe, and we believe that open source, it helps bring this problem to a lesser extent.
Thanks so much because we don't have a lot of time with this.
I'll go for a very quick question.
I see.
Very quick question.
U sorry.
I'll give you the, the final question.
So for your excellency, on the final words regarding what is the openness, how this openness, what is your perceptions of the openness in the lens of the AI and digital and digital transformation? If you can give us a closing sentence of thoughts regarding that.
On a sentence the I'm going to just refer to AI in general, open source for us as I'm speaking now from the national perspective, from the South.
You asked me from the G 77 earlier.
No.
Well, I'm not going to speak on behalf of all of them, but from us, but I see from our perspective from the South, openness is not enough, is not accompanied by capacity building, investment, cybersecurity, data readiness, compute capacity, financing, and public and private cooperation, as I mentioned at the beginning of this closing remark.
Government have a central role as regulator, investor and standards setter, ecosystem building and using of their technology.
But if you allow me 1 minute more, the three closing messages that I would like to share with you is also that first, openness can democratize, the digital transformation.
Secondly, must be grounded in trust, security, human rights, and responsible governance, and third, that you also asked me earlier, openness requires cooperation.
No country or sector can build the digital future alone, not even the UN or a country or a region or political system by the EU, we need to work all together.
The goal is to move from a fragmented digital solution to share ecosystem, from our perspective, from dependency to capacity, from digital divide to digital cooperation.
Openness should be the pathway to share progress, ensuring innovation became a force of development, dignity, and opportunity for all of us.
Thank you.
Very powerful.
Thank you so much.
Very powerful message.
So we know what we are taking with us after the decision.
Minister, the same question regarding the openness.
I mean, we started with what is open in AI or open AI and what is the openness? Openness is a ground for trust and without trust, the digital democratic digital world is not possible.
So everything has to be grounded related, built on trust.
That is, again, a governance issue and trust is needed so obviously that, uh, Even when I'm usually against bureaucracy, I fully agree with my Spanish colleague that you need some kind of regulations to build trust, and then you can build a lot of innovation on these regulations because if we share some of the regulations, then you can build whatever innovation on that.
This is again, one of the issues where we trust United Nations to build trustable, um Agreements, at least agreements where the innovation can be built on and openness is so much related with trust, it allows controllability, it allows interoperability.
I fully agree what my colleague from Uruguay said, but the central focal point is trust and trust with the AI issues, it's moving so fast, we have to act much more faster.
Thank you.
Thank you so much.
Yeah, well deserved at least.
Alberto in 30 seconds in 30 seconds, strest on people, but it's also legal certainty for innovators and that is very important because you know what is going to happen if something was wrong because there are rules and then you can rely on those rules.
So legal certainty, it's super important in order to make your commercial offer really viable and really scalable.
I think this is what regulations like the AI Act and the work that Estonia and Spain we are doing also in making AI responsible is so important to be based on those two pillars, trust and legal certainty for innovators.
Thank you so much.
You, you heard about all the calls for the UN.
I will let you close this session and close the day, please.
Thank you.
There's a few words between you and the reception.
I think I agree with whatever has been said.
There's not much I can add there.
I think the call for action for us is to ensure inclusion in our own processes, work very hard for that and then promote openness, which leads to trust, which would hopefully lead to adoption and applications and hopefully those applications would be done in a in a well governed way so that they lead to impact for the maximum number of people on this planet, the positive impact.
Thank you for a wonderful day today.
Thank you for your presence here and looking forward to seeing you at the reception.
Thank you very much.
And see the microphone works again.
I think it's because the USG said we have to fix it.
Thank you very much for the whole day.
Can we do a round of applause for the audience for sticking through the whole day? It's not easy in these windowless rooms.
So as was mentioned by USG, there's a reception that will be on the fourth floor.
Unfortunately, there's a limited capacity, so please check your badge.
If there's a sticker, you'll have priority.
So please make your way to the fourth floor and then see you there.
Thank you.