High-Level D4SD Panel on South-South and Triangular Cooperation for Digital and Artificial Intelligence for Sustainable Development

Hello everyone, good afternoon. Excellencies, distinguished guests, and friends and colleagues, it's an enormous privilege and pleasure to have you all in the room with us today for a very important and very dynamic discussion. I appreciate you're all here to hear from our fantastic panel and not from myself. By way of background, I'm Callum. I work on AI at the United Nations Development Programme, UNDP, and I will be moderating the discussion today. We have representatives from several governments, from UNDP, and also national— from going to global to national to local, so we should have a very diverse and dynamic discussion. I'd just like to briefly start the discussion by making 3 broader points to frame why we're here today. So we're here to unpack a very important and complex question: what does it take for African and Arab states the Global South to become leaders and not only consumers in the open-source AI ecosystem. And there are 3 points I want to make in, in that context. First, AI leadership is not just about technology. In fact, technology is often the comparatively easy bit. Leadership is very much about building the skills, the culture, the opportunities, and the platforms to drive AI opportunities and exploration. And open source is a very important catalyst throughout this. Second, AI leadership is also not solely about generative AI. In the same way that we need to recognize that AI is not synonymous with generative AI, AI has been around for well over 60 years, and leadership also now means identifying and leveraging new pathways and new opportunities in the AI landscape. And again, open source is woven throughout this journey too. And then very, very finally, we often talk about AI as artificial intelligence, but as, as important as this, we also need to be amplifying intelligence. So this is about strengthening and leveraging the digital talent across governments, the private sector, civil society, and our populations in countries and communities around the world. With that in mind, the way we're going to run the discussion is I'll hand over to Her Excellency in a moment to open the discussion. We then have a series of fantastic panelists, and then we would very much invite contributions and perspectives from the floor. With that in mind, with no further ado, it's my enormous privilege to invite opening remarks from Her Excellency Professor Amal El Falaa Sogrouchni, Minister of Digital Transition and Administrative Reform from the Kingdom of Morocco. Excellency, the floor is yours.

Excellencies, ladies and gentlemen, dear partners and colleagues, it's a pleasure to take part in this high-level roundtable at a time when artificial intelligence is becoming deeply embedded in our economies, institutions, and public administrations. The question is how do we shape it, govern it, and build the capabilities that allow our countries to benefit from it on our own terms. From Morocco's perspective, open source AI may help achieving sovereignty since states, researchers, startups, and administrations can understand the technologies behind, audit them, adapt them to their languages and legal frameworks, and ultimately contribute to them. This is especially important for Africa and the Arab states. Many foundation models are designed, trained, and governed outside our regions with data, languages, and value systems that do not always reflect our societies. This increases our dependencies to foreign technologies and shape public knowledge services and decision-making without our, our full participation. Today, open-source AI may offer another path. When models are auditable, reusable, and adaptable, they allow us to move from access to capability, from conception to contribution, and from dependency to shared sovereignty. But openness alone is not enough. An open model without trusted data, compute, skills, governance, and institutional ownership will not create sustainable impact. For Morocco, this is precisely the spirit of our roadmap, AI Made in Morocco, which is a national AI roadmap towards 2030, Building AI capability and AI governance together from the outset. Our ambition is to contribute to the global effort on AI. Through the AI Made in Morocco Roadmap, we are working to position Morocco as producer of trusted, sovereign, and inclusive AI solutions adapted to our languages, our cultures, and our development priorities. This means, first, trusted data. Morocco is working on a national data factory to transform public data into structured, secure, and AI-ready assets while preserving data protection and citizens' trust. It also means building digital commons, including a national software forge and critical public APIs so that administration and innovators can use trusted components instead of rebuilding them in isolation. Second, sovereign infrastructure. AI requires cloud capacity, computing power, cybersecurity, and resilient hosting. Without this, openness remains a a theoretical matter. Third, talent and innovation. We need engineers, researchers, developers, entrepreneurs, and public servants that can not only use AI but understand it, secure it, and improve it. The Jazari Institutes are designed to connect universities, startups, industry, administrations, and territories around applied AI and technology transfer. This also includes national AI models adapted to Moroccan realities, including Arabic, Darija, and Amazigh languages. This is also the spirit of Rali AI Future Lab, a flagship initiative bringing together thousands of young talents, researchers, entrepreneurs, and innovators around a shared ambition: transforming Africa and the Arab states from users of artificial intelligence into creators of AI. Headed in Merzouga last week, Rally AI demonstrated that talent is the most strategic infrastructure of the AI era and that the future of AI will be built not only through its technology, but also through its communities, collaboration, and shared innovation. Fourth, governance. Open source AI must be responsible, auditable, explainable, secure, rights respecting, and aligned with cultural and linguistic diversity. This approach is aligned with Morocco's vision of Administration X.0, a proactive citizen-centric and platform-based state built on interoperability, sovereign hybrid cloud, open APIs, multilingual access, and the once-only principle. At the regional level, the Morocco Digital for Sustainable Development Hub Collide with UNDP gives this ambition a cooperative dimension. Morocco's unique position as a bridge between the African Union and the League of Arab States places us in a position to foster stronger South-South and triangular cooperation around digital transformation and artificial intelligence through the D4SD Hub, We seek to move from isolated national initiatives to shared regional capability. This means connecting government, academia, innovators, development partners, and the private sector around common priorities and concrete solutions. Our countries face common priorities: education, health, agriculture, public services, digital identity, inclusion, language technologies, and Youth employment. Open source AI allows us to pool expertise, build shared datasets and models, and create digital public goods that each country can adapt to its own context. It also allows our regions to move from being rule-takers in the global AI ecosystem to becoming contributors and rule-makers— capable of shaping technologies and governance frameworks that reflect our realities and aspirations. This vision is fully aligned with the Global Digital Compact, which calls for artificial intelligence that is inclusive, transparent, and accountable. Openness remains one of the most effective mechanisms available to strengthen auditability, trust, accountability, and public oversight of AI systems. Africa and the Arab states should not be only users of AI systems designed somewhere else. We must become builders, contributors, and rule makers. For Morocco, openness is not the opposite of sovereignty. It is one of its conditions. Thank you very much.

Thank you, Excellency. Inspiring remarks, and I think a lot of fruit for discussion later as well. I'm now going to hand for a broader development perspective to Francine Pigott, who is Deputy Assistant Administrator of the UN Development Programme. Francine.

Thank you very much, Callum, and thank you, Minister, and friends, colleagues. It's really a pleasure to be here, an honor to be here as part of this discussion. UNDP works in 170 countries, and we work with governments with different ministries, whether it's finance, digital, public administration, planning. We work with those ministries to help them make choices around the use of digital technology, the use of artificial intelligence.

We—

I don't really want to talk to you today about technical assessment, but I can certainly talk to you about where countries stand, where governments stand, uh, in terms of their use of, of AI, the choices that they're making, the gaps, the capacity gaps that, that governments are facing, and the support that they need. And I want to start with, um, some numbers. Because I think these numbers illustrate the unevenness of the picture that the minister is, is talking about how we can address. If you look at the latest Human Development Report that UNDP came out with last year, which was focused on AI, we, we stated that 1 in 5 people use AI. But 2 in 3 expect to use AI in their lives. AI in the last 3 years has reached 1.2 billion people, 70% of those in developing countries. So the reach is far. And here comes the but, but just 5% of AI innovators in Africa have access to the compute that they need. And if you look at the AI landscape assessments that UNDP has done with over 25 governments around the world, understanding, you know, where they stand on AI, what we see is that the political ambition far outpaces the implementation, right? The capacity to steer AI responsibly. So when it— when we look at, um, where governments and countries stand, we see on the issue of open source AI, we see 3 different postures that I want to flag to you today. The first posture, which is exemplified, I think, by the government of Morocco, is where governments are moving deliberately and with strategy with open source or open weight AI. Okay, and Morocco is a good example. I think some of the countries in the Gulf, uh, Rwanda also. That's the first posture. A second posture is where governments are very interested in exploring open source AI, but they're constrained by implementation challenges. And then the third, which is the third posture, which I think reflects the situation in many countries, um, is that they're adopting AI by default, right? They're not making a deliberate decision about whether it's open source or closed. There's not a deliberate or clear governance or development strategy with regard to the use of AI. So the question that we see is not, you know, whether open source, open weight AI will reach— it has arrived. But the question is whether governments are going to address these issues by design, or by default whether to use open or closed sources. So the minister mentioned the Global Digital Compact, and that, as you mentioned, Minister, calls for open, inclusive, and accountable AI. And this Digital for Sustainable Development Hub that we're going to talk about today is really a good example of how to support countries and governments in addressing these issues and addressing some of these, these gaps or this unevenness. This hub is a 3-year partnership between UNDP and Morocco, the Ministry of Digital Transition, and it was built precisely to address these gaps that we're discussing. And as the minister said so eloquently, Morocco is not waiting to be governed by AI norms written elsewhere. It is helping to write them. In March 2024, Morocco co-led the first UN GA resolution on AI alongside the US, the only developing country to do so. And in July 2025, it hosted the first national conference on AI, producing a governance roadmap that the minister talked about, which is now being turned into law. And then in January this year, it launched Jazari Root, a national network of AI centers for excellence. So we've all— we have already, as UNDP, as I mentioned, deployed these AI landscape assessments, uh, sort of looking at the lay of the land with regard to AI in a number of countries, Rwanda, Malawi, Côte d'Ivoire among them. And these give governments a clear picture of where they actually stand before they make deployment decisions. We also have, alongside this AI landscape assessment in UNDP, we have this solutions marketplace which shares tools across borders. And we're accompanying, uh, 3 African pilot countries: Senegal, Malawi, and Nigeria. Through this first Digital for Sustainable Development Acceleration cohort. And the logic here is simple. A solution that works in Morocco travels to its neighbors far more easily than anything that's being designed further away in, in high-income contexts. Same languages, similar legal traditions, comparable fiscal realities. And this is the South-South principle in practice. So what we need from this room is to help us convert some of this very exciting momentum into delivery. The political foundations are clearly in place, and the question that we hope to grapple with today is implementation.

Thank you, Francine. Moving back to the government perspective, I'd like to invite His Excellency Bakytar Mukhametkaliyev, Vice Minister of AI and Digital Development of the Republic of Kazakhstan, to share his remarks. Thank you.

Yeah, first of all, thank you for your presentation. And we really appreciate before we did the math and discuss very interesting topic. In Kazakhstan, before we had the, like, this meeting and when I show our results and the plans and I try to briefly show this and maybe tell. In our country, our president announced this year is digital and artificial intelligence year. And all our activities that follow to implement artificial intelligence in the digital sphere. And we have some achievements in the government platform. In the previous year, we launched supercomputer cluster with the NVIDIA GPU. We have the international AI centers when we working to the under to developing on the human capital. We have the big lands, we have the natural resources, we have the energy system. And today we try to build some comprehensive approach. For example, the first level for us, it's the energy system. And the second, we— before I discuss about the data centers, we, we build the data centers to the two ways. First of all, to internal, to our demand, uh, to our country. And the second, the export-oriented. We have a strong agreement with the Firebird NVIDIA company to build 100 megawatts at the first step, but scale up to the 1 gigawatt data centers in the northern part of our country. It's only export-oriented centers. And the second, we have maybe one of the good e-government platform because we make 1,300 services in online format. And today we try to implement artificial intelligence in the difference here. For us, it's so important. But the same time, we are happy to the achievement for the other governments, and we also they open. If you maybe need to discuss about something, and we also do try to share experience, and all country have that the big achievement, some difference here because it depends from different culture, different political structure, and For us, it's technological is one of the important thing to our agenda. And with the open available 24/7. Thank you.

Thank you, Sir. Very inspiring work. Huge investments in data centers, in compute, in energy as well. Shifting further east, actually, I'd like to turn to our representative from Japan. Mr. Masayuki Yamada is head of the data unit within the digital agency of the government of Japan. Thank you.

Distinguished delegates and esteemed colleagues, thank you Morocco and the UNUTP for hosting this vital talk. To ensure absolute precision, please allow me to refer to my prepared notes today. Morocco holds a special place in my heart. In 1993, I, as a student, I traveled from Tanje across the Atlas Mountains to the Merzouga Desert and on to Tunisia via Algeria. That journey taught me a deep respect for local cultures. To be invited by the Kingdom to speak today is a great honor. That respect for local context is the core of digital sovereignty. As our delegation will share during the day 4 country spotlight on the Japan model in the Open Source Week, our journey began with a struggle 40 years ago. When global rules couldn't handle Japanese-specific vertical typesetting, our engineers built a local patch and gave it back to global CSS standards. To our surprise, that base code also helped support Arabic and Mongolian scripts. When we solve local needs without overfitting to our own context, our solution generalized to empower others. By doing so, we established our active presence in the global ecosystem. And when you give something back, a door opens from being a user of open source to a co-author of it. This might be the essence of the sovereignty. This same rule applies to open source AI. For those of us working in real government service, let me share our reality. Japan is rolling out our government AI, GenAI, to 180,000 civil servants this year. In April, we put part of it on GitHub with open commercial licenses. Open models gave us the base, but we quickly learned having the model is just a start. Using AI in government brings a real hurdle. You must test it. You must prove it follows your laws and culture. To do that, you cannot just download test data from the web. You must build your own local data. You must define your own ground truth. And this is where the real work hits you. In an AI training data verification program I directed, We needed to check if the model had correctly run our local laws. We asked top lawyers to create 140 multiple-choice questions in complex areas that require cross-domain interpretation. But during testing, 3 different AI models agreed on answers that contradicted the official answers key on 7 of these questions. We asked the experts to re-verify. The result was very humbling. They realized that 4 of those 7 questions actually contained human errors or ambiguities. The blocker was not the AI. It was our human ability to perfectly interpret our own rules. This shifted, um, our view on AI sovereignty. We now believe that building your own local test data may matter more than possessing computing resources. If you run a local model but rely on outside data to grade it, you let others define your truth. True AI sovereignty means taking the duty to describe your nation in your own words. You cannot outsource self-understanding. To read in open source, you must own— sorry, I'm repeating. From Japan's experience, today I have leaned toward the data side of that story. There is more to it, but data is where I wanted to start. This week has a full day of digital public infrastructure. I would simply add data curated and owned is part of that infrastructure too. Thank you.

Thank you. Knowledge cannot be outsourced. I think it's a very valuable point. Now, I'm going to move from the global to the local. So, my pleasure to introduce Dr. Elinor Fournier-Toum, who is Chief AI Officer of New York State. Elinor, the floor is yours.

Thank you so much. Thank you for having me here, and I've already learned so much, so thank you for all your contributions. Esteemed Excellencies and guests, it's a real pleasure to be at the UN, back at the UN for Open Source Week, an event that has really grown in importance over the last few years due to the increased relevance of the subject in today's world. So I remember speaking at the first iteration maybe 3 3, 4 years ago, um, and now we're speaking about open source AI, and I think it's, it's a very different context. Um, for the last 6 months, I have been working as New York State's inaugural Chief AI Officer, tasked with the mandate of deploying responsible AI across the state government with a focus on, uh, the 57 executive agencies and a workforce of approximately 130,000 people. Our AI team is composed of three pillars. So we have innovation, which works on the deployment of general-purpose AI tools such as Code Assist, generative AI, and predictive analytics and more. So not just limited to generative AI. Governance, which is tasked with the implementation of the state's acceptable use of AI policy, which legislates AI from a risk perspective. And training, which ensures that the workforce is upskilled in ethics, engineering, and literacy. This all operates in a spirit of taxpayer accountability where we must make sure that we are spending money effectively and with an objective of improving the delivery of services. In this context, open source AI is becoming more and more interesting for a number of reasons. The first is data protection. In a government context, data protection is essential. Agencies manage residents' data that are highly regulated, such as, uh, here, um, HIPAA, which relates to personal health data, CJIS, which relates to criminal justice data, and more. Ensuring that the tools can be used without risking data exfiltration, uh, which is a major issue, of course, with closed-source models, or oversharing, uh, has become a really big concern of ours. Open-source AI, because it is run locally on government premises, has the potential of reducing the risk of data leaving the state tenant and going to a third party. Secondly, costs. In experiments by our team with closed models, especially CodeAssist tools, we have seen that the more models are used, the more they cost, which seems obvious, but it's not a subscription-based, it's based on use. And so your super users will be the most expensive users. Um, and we've, we've learned the hard way. So while this may be less visible in the case of general-purpose LLMs like your personal ChatGPT, um, that's used to summarize text or the like, it is certainly visible in the case of an engineer who's using a closed model to develop a new web app application, to refactor a code base, or perform a vulnerability scan on critical state infrastructure. Because OpenAI tools don't rely on external APIs, there are no token costs. Which in my experience can really increase the cost of running a model. This doesn't mean that there are no costs at all because these tools must be hosted and run on-premise. However, for high-volume use cases such as coding and engineering, the open-source tools are certainly much more affordable for a cost-conscious government. Finally, there is the question of flexibility and empowerment of in-house engineers. Our team has developed some tools from open source AI, namely a custom procurement support tool, which was very well received internally because of its flexibility and motivated my team also because we developed it ourselves. So what are some of the current challenges for local government deployment? Critical risks include cybersecurity, where models can behave unpredictably, insert code vulnerabilities into critical systems, risks that are usually at least partially handled by the vendor in the case of closed models. There is also generally technical debt where since there's no vendor support at all, the tools need to be maintained internally by government workers. Finally, there are also licensing issues. Out of it, not always clear on what data the models are trained and if that data was obtained ethically from an international perspective. The most relevant work that can be done globally to help local deployment can include the development and continuous testing of open-source models. They've improved a lot in the last year, and I think they will continue to do so. The funding of local infrastructure that can store and run the models, and the creation of policies on which local governments can base their own governance frameworks. The world of AI is changing very quickly, And open-source AI is so much further along today than it was last year. I look forward to seeing how we can use these tools for public benefit in the coming year. And thank you, and I look forward to our conversation.

Thank you, Adrienne. Great insights into implementation. And we're seeing the word tokenomics pop up quite often now. It's— this is not, uh, yeah, it's a new level of, of cost that we're seeing in, in closed source. Um, Just to wrap up our panel discussion before we open to the wider floor, I'd like to invite Deema Al-Khatim, who is Director of the UN Office for South-South Cooperation, to share her remarks. Deema, the floor is yours.

Excellency, we're very happy to join you in this discussion, and I've been also listening very carefully to the different interventions and wanted to share from our perspective, a few messages as it relates to the subject. First one on AI governance, that it must be grounded in national capacity, trust, and development leadership. We have recently undertaken a foresight exercise on the future of South-South and Triangle cooperation that shows that the Global South has the capacity to lead on sustainable development end use of AI. And as noted by Her Excellency, countries need to further develop the institutional, technical, and policy capacities to shape and audit and adapt and govern AI systems in line with their own languages, institutions, rights frameworks, and sustainable development priorities. And under one of the foresight scenarios developed by our office, namely the AI dividend scenario, Countries strengthen infrastructure access and institutional capacities at the same time. And this means that sovereign AI becomes possible not because every country builds the full AI stack alone, but because countries cooperate on shared compute, local language models, data governance, procurement safeguards, skills, and implementation capacity. But the second thing is that South-South and Triangle cooperation can turn the AI divide into an AI dividend by strengthening institutional capacities. And the divide is not only about technology, as you mentioned at the opening. It also concerns whether public institutions can access needs, procure responsibly, manage data, and oversee risks, audit systems, and scale AI solutions safely. And cooperation can help countries pool resources, share safeguards, and build the institutional— build the institutional readiness needed to move on from pilots to durable, durable public value. The third message is that OpenAI models are, I mean, a strategic pathway to sovereign and inclusive AI in the Global South. Open, auditable, and adaptable models can reduce dependency on closed systems and support local language and domain-specific applications, improve transparencies and accountability, and make AI more accessible for public service delivery education, health, agriculture, and climate resilience. Noting the experience of Japan and what you shared with us, I think it's also, you know, an opportunity to look at the lessons learned and learn from that. The Digital for Sustainable Development Hub is a unique and practical regional platform for AI governance, cooperation, and leadership, and we see the hub as a platform that connects existing ecosystems in Africa and the Arab states. It connects policymakers, technical communities, and partners, exchange expertise on open-source AI, data governance, procurement safeguards, and responsible deployment, and of course translate the global principles into practical regional action. And we look forward to have the D4SD as a member of the Global Alliance on South-South and Triangle Cooperation. Thank you.

Thank you, Dima. Um, I'd now love to invite participants' perspectives from the, from the floor. Um, we only have about 10 minutes or so left, so would kindly ask you to briefly introduce yourself, the institution that you represent, and then any questions or points you want to raise for the floor. Thank you.

To, um, NDP in the Kingdom of Morocco, um, and, um And we have the pleasure to work very closely with Her Excellency and her team, not only in what really Morocco has led the way, really, in a lot of innovations that we believe are so important for the efficiency of public administration and the delivery of public services, and therefore showing how how technology is so central to development and to inclusion, in fact. And now we have the pleasure of also co-creating together with the government of Morocco this ambitious model for the D4ST hub. And for me, we are working very intensely on this project. And we have developed specific offers that go with it, and it's— and they really respond to some of the challenges that the different panelists have highlighted. And it comes down to really capacity and collaboration. So I think maybe the question that I have for the panel, because we are working on this partnership, is we know what the challenges are, we know what the needs are for capacity building, for talent development, for capacity building, comparing administration, for this collaboration on solutions and models across borders. So from your perspective, what do you see would be the most important entry point on what would be the types of solutions that most urgently need international cooperation to develop open source, which can be, can be, can benefit from the open source AI models. When we look at development, you know, across the regions also that we, we see, I mean, we see common challenges on, for example, on on water, we see common challenges on climate change adaptation, we see common challenges on energy efficiency. But what are the— from your perspective, and we heard different perspectives, the city perspective, the country perspective, if you had to identify one entry point that would make the most transformative effect on leapfrogging on this, What would be the entry, the thematic entry point that we should really look at? Thank you very much.

Thank you. So thematic entry points and AI opportunities. Excellency, would you be happy to speak from Morocco's perspective first, and then we can get— take the panel?

Please.

All yours, sir.

Thank you.

It's very difficult to say, to speak after the eminent panelists and Her Excellency the Minister, but all I want to say is that artificial intelligence is an opportune tool to enforce cooperation, especially South-South cooperation and trilateral cooperation, because now artificial intelligence is evolving in all aspects of our lives. But I'm not talking about the human rights, privacy, or cybersecurity. I'm talking about the implementation of SDGs. And I think what we have here as hub is a model of cooperation. And we are very happy that UNDP trusted Morocco and worked with him to prepare this hub for African countries and Arab countries. And it opened the window, if not the door, for enforcing interconnection and also enforcing the kind of cooperation that we have either in our Arab region or Africans. Fortunately, and as you know, Morocco have a roadmap of cooperation with Africa, African countries, and this question will be integrated, if not it's already integrated in the plan of action that we have with several countries. It's the same time for the Arab region. Morocco is also privileged because it has adopted, under the leadership of Madame the Minister, its national strategies. And when we see how many countries have prepared their national strategies in Africa, It's less than 10%. So I think we are— Morocco doesn't— never claimed that it's the model. No, but we are showing the road. And for this, we would like to do so. And of course, whenever we are requested to help, to build with them, and to cooperate, Morocco is already available for this purpose. Because we believe that the race for the exploitation of artificial intelligence is huge and Africa should not be left behind. Africa should invest. It has the human resources, it has the energy, it has the potential, the markets, and what is needed is may be funds, finance. But there, Africa can aggregate its efforts financially. And of course, there is already plan or strategy for African countries that have been adopted by the African Union. There were several meetings. I think the last one were in Kigali and Addis Ababa. But personally, if you If you ask me, I will say that it's not enough because Africans— we can see it here in New York. When we are calling for meeting on artificial intelligence, few African countries are present. Why this disinterest? I cannot explain it. It's the same thing with the last summit Kenya, there was less African countries. So, and we see that India and United States have already taken, you know, a long way in their strategic cooperation. It was the same thing when the summit took place in Paris. That's why I think having such event here and this week of open sources is an occasion for mobilizing political will, mobilizing financial resources, and also human resources. Engineers, African engineers and Arab engineers who are working in the north are numerous and it's thousands of people, but they need to think also about their respective countries. I'm not saying that we have to tell them to come back. No, they can stay there, work there, improve there, advance there. But in same time, if they can dedicate just small time of their life, of their profession to help their respective countries, I think Africa can gain a lot of space. I'm sorry I had to improvise this because I was not ready to make speech. Thank you.

You improvised very, very well, so fantastic. And we are actually seeing countries— I think Nigeria is a very interesting example about how they leverage their diaspora to write their national AI policy. So diaspora, very important. Um, I don't know if the rest of the panel would like to jump in on this topic at all. Everyone's very polite.

Very good question. And just a quick— I'll come up with 3 very quickly, things that could really help accelerate. I think first one is that, you know, openness is not the same as equity. And I would build on what the ambassador said and just say we— what's really important is building capacity to— within governments to understand and use AI. So just building that government capacity is critical. A second area that I think could really accelerate is building that compute infrastructure and making sure that it is energy aware. So I think that's a second area.

And I—

the third, I think, building on what you said, Ambassador, is around language data. We know that Africa has more than 2,000 languages, but the vast majority are absent from any training data. So that means the models that are developed are— don't use these languages. So I think that's another critical infrastructure that could really make a difference.

Thank you for giving me the floor. I think I will be back to one idea we have discussed with Ilaria a few months ago. I think all what you present, like building capacities, infrastructure, languages, etc., are very important. But if I want to answer your question, Ilaria, about the unique entry point of— to enter the— to solve, I mean, to find something that can solve these problems, I think it's research and development, R&D. Because today, we, for example, I will explain. If we want to run our models, if we want to build these languages, these small models or large language models, if we want to understand open sources, etc., we need to grow up a bit in capacities. Because this domain of AI is not like other domains. It requires very high-qualified people. If we want to solve real problems, we cannot solve them by doing just copy and paste. This means that we need to think about R&D. You gave some numbers, at the beginning, and I agree on all of them. I would like to add that research publications in Africa represent less than 1% of global publications. So this is the point, I think. Building capacities means that we have qualified teachers, qualified trainers. We have to work on very difficult problems. That these problems have no solutions, or at least will not find these solutions somewhere in the GitHub. We have really to push R&D at the African level. This means, if you remember, Ilaria, we talk about like European Commission for Africa and Arab world, that fund collaborative programs across Africa and Arab world. But you have to push on the— to put on the table and to push topics that can bring added value to our countries. And this is possible. I have been working on Horizon program in Europe for a long time, and it was very nice to see people from all the countries around the table solving European problem. I think we can do the same because— not because we would like to stay in Africa or Arab world, but because we have the same geography, we have the same problems. The problem of water in the south of Morocco is not the the same in France or Germany. So we have to find the flagships that are very important for our countries and to fund research and development in this. And I think this is the unique way to solve the problem. Otherwise, we will struggle in doing the same but badly, not in the same— the right direction. Etc. So I will be very interested if there are people around the table that want to create this advisory board. We will talk on this next Tuesday— Thursday, I think. In this advisory board, to look to how to push research and development within G4SG. That could be one starting point.

Very important points. You're absolutely correct. We see publication rates, patenting rates, citations much lower across the African continent. I think increasingly as well, we're seeing a challenge where if you are an AI researcher, even from the Global North, and you want to continue AI research, you go to the private sector because they have the money, the compute, and the power and the place for you to do research. Being in the public sector or being elsewhere, it's getting much, much harder to be a researcher now. So you make some very, very important points. I think we have time for one final reflection or comment from the room before we wrap up.

Thank you for giving me the floor. It's an honor and a pleasure to be among these very esteemed panelists. I'm here on behalf of ITU. I represent Doreen, our Secretary-General. So I wanted to also like share what ITU is doing in this space. The way we would like to think about it is— I love this analogy by NVIDIA CEO who's describing the AI stack as a multi-layer cake. I think here we're also addressing like different areas in that cake where we have like the infrastructure area including energy, network, et cetera. We have the compute layer, then followed by the regulatory and governance layer, and finally the application layer. With ITU, we try to support our members across all the board and across the different layers of the stack. In one aspect, we provide like tools for infrastructure design. And infrastructure deployment cycle understanding and data evidence-based decision-making. We just launched a tool called the Connectivity Plan and Platform, CPP, which is kind of an open source but available for our members at this stage, where we help our member states and regulators make decisions based on actual data with the data validation pipeline. We've got like many programs also through i2 Academy, Gov Stack on the application layer where we're trying to design sandboxes and try them with local data, with local initiatives. And of course, like our flagship conference, AI for Good Summit, taking place in a couple of weeks from now in Geneva, where I invite you all to hopefully to be able to see you there. So yeah, it's a multi-pronged problem with different pieces. AI governance dialogue is very much needed at this point also to kind of trigger anything. So the biggest hope here for us, we're here in the Open Source Week, is that open source is not an end goal, but rather a catalyst, a trigger to kind of like push in the different areas. So we're hoping to continue the dialogue and looking forward to working together to support you in your endeavors. Thank you.

Thank you. Before I hand to Your Excellency to wrap up proceedings, a few shared points I think have come out from the different presentations and the points raised. I think first, for me, open source, it's clearly open source AI is a valuable way to build the skills, the expertise, the capacities, and institutionalize that knowledge within government and beyond government, but also build the kind of ecosystems and the sustainability and reduce the dependence on external actors, resources, and similar too. Similarly, we've also heard that AI is not homogenous. It has different strands, different technologies, different tools.. And it's about making it real and relevant to local populations, whether it's attuned to national languages and local languages or to national priorities. But then also, despite that, there's also an exciting opportunity where it is both local but also opportunities to drive shared collaboration, shared solutions, and development as well. And then very finally, I think what's coming through very much from all the panelists and the points raised is that the aspiration is huge. Countries are already running very, very quickly. They're not waiting for the global infrastructure to keep up. They are leading the way, and I think hugely exciting opportunities to learn from what's happening across the Global South. An example I always go back to is that Kenya had mobile money 15 years before London did, and we're seeing some amazing stuff happening across the Global South around AI that I really feel like the Global North and others can be learning from as well. New pathways, new opportunities, new technologies, and new direction as well. With that in mind, I'll hand over to Her Excellency to wrap up the discussion, and again, thank you all for your time today as well.

Thank you very much. To wrap up, I would like to say that we are discussing the opportunity to use open source to improve our development in AI, and this is something very useful, I think. What you have said is that using technology is very important, but we should be at the origin of the development of this technology. If I would like to simplify the message, our future of AI should be written by us in Africa and Arab world because we have common concerns, we have common problems, and we have also many similarities in needs, languages, culture, health, etc. I think as His Excellency said Cooperation is very important, whatever the shape of this cooperation— South-South, multilateral cooperation. But which is very important is that Morocco has opened the window, the door to this cooperation. And we can provide— I think at this stage we can provide the model many aspects of developing AI, like the Institute of Jazari. Like, a very exciting experience we had last week is putting together 1,000 youth people and in the desert, and they had to build solutions on AI and 45 degrees Celsius, a lot of water. They were awake all the time for 4 days, and the result is just amazing. And I think collective intelligence is something we can count on in our countries. Solidarity also. We have seen many situations of solidarity, very amazing. People with handicap, for example, were completely taken in charge by the others. And, well, I think we can provide very nice stories and experiences within Morocco. I think we count on all of you to, for the next meeting, to create something very useful. Still, I'm still convinced that R&D is very important and we will see how to do that. May be. We need a very strong advisory board and also a board for scientific staff. I think it's very important. Just to say thank you very much for being here, for this very interesting and nice insights, and see you on Thursday. Thank you.