Developing capacities for bottom-up AI in the Global South: What role for the international community?

10 Jul 2025 10:00h - 10:45h

Developing capacities for bottom-up AI in the Global South: What role for the international community?

Session at a glance

Summary

This discussion focused on developing capacity for bottom-up AI in the Global South, examining what role the international community should play in supporting AI development in developing countries. The session was organized by the Diplo Foundation in partnership with Kenya’s Permanent Mission, Microsoft, and IT4Change, using a fictional case study of “Landia,” a landlocked agricultural country with 8 million people facing typical development challenges.


UN Tech Envoy Amandeep Singh Gill outlined the Secretary-General’s upcoming report on innovative financing for AI capacity building, emphasizing the need for nuanced understanding of different countries’ AI needs across five development tiers. He highlighted that capacity building requirements vary significantly, from basic AI literacy to advanced development capabilities, and proposed a coordinated global response including a potential global AI fund and better coordination among funders to avoid fragmentation.


Microsoft’s Ashutosh Chadha framed AI adoption as fundamentally a policy challenge, arguing that countries need comprehensive national AI strategies addressing the entire technology stack, from electricity and connectivity to data governance and institutional capacity. He emphasized that technology should adapt to existing work patterns rather than forcing communities to change their practices.


Anita Gurumurthy from IT4Change advocated for “regenerative AI” that is indigenous, inclusive, and intentional, challenging the mainstream view that more computing power automatically leads to better AI outcomes. She highlighted the potential of smaller, task-specific models that can run locally and the importance of building on local knowledge and resources, including public broadcast archives and agricultural data cooperatives.


Participants raised critical questions about infrastructure prerequisites, with representatives from Botswana questioning whether AI policies should be developed before addressing basic connectivity and power supply issues. Others emphasized the importance of ensuring AI serves genuine community needs rather than imposing external solutions, and highlighted the need for government support alongside private sector initiatives. The discussion concluded with plans to continue developing practical capacity-building strategies through an AI agent tool that participants could access after the session.


Keypoints

## Major Discussion Points:


– **AI Capacity Building Framework and Financing**: Discussion of the UN Secretary General’s upcoming report on innovative financing options for AI capacity building, including a proposed five-tier system for countries to progress from “AI nascent” to “AI developer” status, and the potential creation of a global AI fund to address the “AI divide.”


– **Bottom-Up vs. Top-Down AI Development**: Strong emphasis on developing AI solutions that serve local community needs rather than imposing external frameworks, with focus on “small beautiful models” that can run locally and address specific tasks rather than requiring large-scale infrastructure.


– **Infrastructure and Policy Challenges**: Recognition that basic infrastructure issues (reliable electricity, internet connectivity, data governance policies) must be addressed alongside AI development, with debate over whether to tackle these sequentially or simultaneously.


– **Alternative AI Development Models**: Discussion of emerging alternatives to mainstream Western AI approaches, including the BRICS AI declaration’s emphasis on balanced intellectual property rights and the potential for regional cooperation in AI development.


– **Community-Centered AI Applications**: Focus on practical applications that augment existing local practices (especially in agriculture) rather than displacing traditional methods, with emphasis on training local talent to support farmers and rural communities using accessible AI tools.


## Overall Purpose:


The discussion aimed to move beyond abstract concepts of “capacity building” to develop concrete, practical strategies for AI development in the Global South, using the fictional case study of “Landia” (a landlocked, agriculture-based country) to explore how international cooperation can support bottom-up AI initiatives that respect local contexts and needs.


## Overall Tone:


The discussion maintained a collaborative and constructive tone throughout, characterized by genuine problem-solving rather than theoretical debate. The atmosphere was informal yet focused, with participants building on each other’s ideas. The tone remained consistently optimistic about finding practical solutions while being realistic about challenges. The use of the fictional “Landia” case study and the coffee machine AI assistant “IQ whalo” added a creative, engaging element that kept the discussion grounded in practical applications rather than abstract policy discussions.


Speakers

**Speakers from the provided list:**


– **Jovan Kurbalija** – Director of Diplo Foundation and Head of Geneva Internet Platform, session moderator


– **Amandeep Singh Gill** – UN Tech Envoy and Under-Secretary General


– **Ashutosh Chadha** – Head of Geneva’s Office of Microsoft


– **Anita Gurumurthy** – Representative from IT4Change


– **IQ whalo** – AI-powered coffee machine serving as an example/demonstration tool


– **Nandini Chami** – Representative from IT4Change, from India


– **Alan Ross** – Area of expertise, role, and title not specified


– **Baratang Miya** – From Girl Hype, organization that teaches women and girls how to code


– **Rudy Massamba** – From Congo Brazzaville, area of expertise and role not specified


– **Audience** – Various unidentified audience members who made comments


**Additional speakers:**


– **Tabaget Zavila** – Regulator from Botswana


– **Unnamed audience member** – Asked about BRICS membership and AI development approaches (specific identity not provided)


Full session report

# Report: Developing Capacity for Bottom-Up AI in the Global South


## Executive Summary


This discussion, organized by the Diplo Foundation in partnership with Kenya’s Permanent Mission, Microsoft, and IT4Change, explored how the international community should support AI development in developing countries. The session used a fictional case study of “Landia” – a landlocked agricultural country – to examine practical approaches to AI capacity building that prioritize local needs over externally imposed solutions.


The 45-minute workshop-style discussion featured diverse perspectives on AI capacity building, with participants generally emphasizing the importance of adapting technology to local contexts rather than forcing communities to change their practices. Key themes included the potential for alternative AI models that don’t require massive computational resources, the role of policy frameworks in enabling AI implementation, and the ongoing debate about infrastructure prerequisites versus pragmatic implementation approaches.


## Key Participants and Perspectives


**Jovan Kurbalija**, Director of Diplo Foundation, moderated the session and challenged assumptions about infrastructure prerequisites for AI development. He emphasized avoiding the anthropomorphization of AI and demonstrated practical AI applications using an AI-powered coffee machine called “IQ whalo” to illustrate that AI can be embedded in simple devices without requiring massive computational infrastructure.


**Amandeep Singh Gill**, UN Tech Envoy, provided the institutional perspective and outlined the Secretary-General’s upcoming report on innovative financing for AI capacity building. Drawing from his previous work on lethal autonomous weapons, he emphasized that “correct understanding of what we are dealing with, what it is, before policy, before capacity building, before anything else, that’s good action only flows from correct understanding.” Gill mentioned a five-tier system for countries to progress in AI development and noted that capacity building requirements vary dramatically between contexts, observing that “three extra GPUs for Ethiopia, which has a total of 12 GPUs, is meaningful. But 3000 GPUs coming to South Africa, which is currently happening, is another context.”


**Ashutosh Chadha**, Head of Microsoft’s Geneva Office, framed AI adoption as fundamentally a policy challenge requiring comprehensive national strategies. His key insight was that “it’s about how do we make technology work for us? It’s not about how technology makes you work. That’s a very subtle shift in the way we need to apply this.” Chadha emphasized addressing infrastructure, data governance, and institutional capacity simultaneously rather than treating AI as a standalone technology.


**Anita Gurumurthy** from IT4Change advocated for “regenerative AI” approaches that are indigenous, inclusive, and intentional. She warned against mainstream AI approaches, arguing that “mainstream AI may not always deliver… part of humanity will just get extinct, right? I mean, those who can’t run the race.” Gurumurthy championed “smallest beautiful models” and task-based local AI solutions, advocating for the “right to tweak, transfer, and transform” AI systems to serve local contexts.


## Major Discussion Points


### AI Capacity Building and Financing Approaches


Participants discussed various approaches to AI capacity building, with Gill outlining the Secretary-General’s upcoming report on innovative financing options, including a proposed global AI fund to address the “AI divide.” The discussion highlighted that different countries require different types of support based on their development context.


**Nandini Chami** from IT4Change emphasized connecting with micro, small, and medium enterprises in similar agro-economic zones, while **Rudy Massamba** from Congo Brazzaville stressed that government support is essential for community-based AI development. **Alan Ross** focused on practical applications, advocating for helping farmers improve existing practices using simple technologies like $100 drones for leaf analysis.


### Technology Adaptation Versus Community Adaptation


A recurring theme was whether technology should adapt to local contexts or whether communities should change their practices to accommodate new technologies. **Baratang Miya** from Girl Hype raised the crucial question of agency: “AI should serve whether their needs or is it a choice of the people that are bringing AI to them who’s going to decide what are the needs? Because we might end up automating the inequality that is existing here.”


Multiple participants emphasized the importance of radio networks as key connections in developing countries and the need to work with existing communication infrastructure rather than requiring entirely new systems.


### Alternative AI Models and Geopolitical Considerations


The discussion explored alternatives to mainstream Western AI approaches. Gurumurthy highlighted the BRICS AI declaration’s emphasis on balanced intellectual property rights as offering potential alternatives to current Western-dominated AI systems. She advocated for smaller, task-specific models that can run locally, building on resources such as public broadcast archives.


When asked about geopolitical alignment strategies, Kurbalija suggested working within the broader multilateral system while maintaining good relations with BRICS, noting that even the European Union finds itself in a similar position to developing countries regarding AI dependency.


## Areas of Different Emphasis and Debate


### Infrastructure Prerequisites Versus Pragmatic Implementation


The most significant difference in perspectives emerged around infrastructure requirements. **Tabaget Zavila**, a regulator from Botswana, questioned developing AI policies without addressing basic infrastructure challenges: “How do we develop a policy, an AI policy, while the basic things like network availability and infrastructure, that’s still a challenge.”


Kurbalija countered this perspective, arguing for addressing infrastructure challenges simultaneously rather than sequentially. He cited the example of a Botswana participant in Diplo’s AI apprenticeship program who successfully created an AI agent for non-communicable diseases despite intermittent electricity supply, demonstrating that infrastructure limitations need not prevent beneficial AI applications.


### Mainstream Versus Alternative Development Approaches


Participants offered different perspectives on working within existing AI frameworks versus developing alternative approaches. Gurumurthy advocated for alternative approaches including smaller local models and regenerative AI, while Chadha focused on working within existing frameworks while improving policy coordination across infrastructure, data governance, and institutions.


## Practical Applications and Examples


The Landia case study, used as a teaching tool by the Diplo Foundation, provided a framework for exploring capacity building strategies. Participants discussed how an agricultural country could leverage AI while respecting local contexts and addressing infrastructure constraints.


Key practical applications discussed included:


– Using simple AI tools like drones for agricultural analysis


– Training local graduates to serve as bridges between AI technology and rural communities


– Implementing AI solutions that work intermittently when electricity is available


– Leveraging existing broadcast archives for developing local language models


The session also referenced the successful Botswana AI agent for non-communicable diseases as a concrete example of effective AI implementation despite infrastructure limitations.


## Unresolved Questions and Future Directions


Several important questions remained open for further exploration:


**Implementation Mechanisms**: While participants discussed principles, specific mechanisms for ensuring AI serves community-identified needs rather than externally imposed solutions require further development.


**Scaling Challenges**: How to scale successful local AI solutions while maintaining their community-specific benefits presents ongoing challenges.


**Resource Coordination**: Despite proposals for a global AI fund, detailed implementation pathways for coordinating international AI capacity building efforts remain to be developed.


**Infrastructure Trade-offs**: The debate between sequential and simultaneous approaches to addressing infrastructure challenges and AI development reflects broader questions about development priorities and resource allocation.


## Conclusion and Next Steps


The discussion highlighted the complexity of AI capacity building in the Global South, with participants offering various perspectives on how to balance local adaptation with practical implementation challenges. While there was general agreement on the importance of serving local needs, specific approaches and priorities varied among participants.


The session concluded with practical next steps, including participants receiving access to an AI agent developed by the Diplo Foundation for continued strategy development, and plans for ongoing collaboration on the Landia case study. The Secretary-General’s report on innovative financing options is expected in the coming months.


The use of the Landia case study as a teaching tool proved effective in grounding abstract concepts in practical scenarios. The session’s emphasis on understanding local contexts before implementing solutions, and the recognition that developing countries can shape AI development to serve their specific needs, provides a foundation for continued exploration of these important questions.


Both Ashutosh Chadha and Anita Gurumurthy serve on the UNCTAD committee on data governance for development, suggesting potential avenues for continued collaboration on these issues within existing multilateral frameworks.


Session transcript

Jovan Kurbalija: Good morning, welcome to our session on Developing Capacity for Bottom-Up AI in the Global South, What Role for the International Community. My name is Jovan Kurbalija, I’m Director of Diplo Foundation and Head of Geneva Internet Platform and I’m particularly honored that we are co-organizing today’s session with Permanent Mission Kenya, Microsoft and IT4Change and our colleagues are here from the organization. I will introduce them shortly. The plan for today’s session is to see what we can do practically about capacity development or capacity building, there is terminological confusion and as you know this is a term which is frequently used in the international documents but like any inflated terms which is used a lot after sometimes it can lose the concrete meaning. Therefore our purpose of today’s session is to regain this meaning by having concrete examples. I’m really fortunate today to have with us Amandeep Singh Gill, Tech Envoy, UN Tech Envoy and Under-Secretary General. Amandeep, welcome. Next to him is Ashutosh Chadha, the Head of Geneva’s Office of Microsoft. It’s a pleasure meeting you. And we have also Anita Gurumurthy from IT4Change. I noticed one thing, and I’m here with three colleagues which are somehow connected to India. And I always tell my Indian friends, it’s probably you who attended our session, I said that India put us in trouble because they invented number zero, Shunya. And before that, we had the Romans number and the life was much simpler. And as you know, digitalization and everything else started with number zero. Of course, it’s a joke. But now it’s a time for our Indian colleagues to help us to deal with their invention, which is number zero, which took off gradually from Arab world and al-Khwarizmi and Fibonacci, but came to today where everything is based on these two numbers, zero and one. Therefore, it will be a really inspiring chat and I’m sure we’ll get to quite a few answers. But what is critical, after their introductory remarks, we will co-create the knowledge. I know it’s Amandeep, one of his favorite concepts of co-creation and education. And it is a genuine meaning. There is a lot of expertise around this table, from very young colleagues to more experienced. And instead of us lecturing, we will have that. And we will have also another panelist which joined the IG community in IGF Berlin 2019. It’s a coffee machine. His or her name, we are still deciding on its name, whatever it is, is IQ whalo. IQ whalo is basically an example that you don’t need to have a robot to have AI. It’s our warning about anthropomorphizing AI. You can have a coffee machine, you can have a Hoover, you can have a refrigerator, that can communicate to AI. I’m sure that I’m not going to be in the good books of our colleagues from AI for good, because even in the logo there is a human face. But I’m quite strong on that point that anthropomorphizingPoses major risks for, good AI governance and the development of AI. Therefore, we will ask later on Qvalo, what she thinks about AI development. Before a further due, I would like first to invite Amandeep Singh Gill, Tech Envoy, UN Tech Envoy, to tell us a few words from the perspective of New York and overall thinking.


Amandeep Singh Gill: Thank you so much, Jovan, and thank you to you, Diplo Foundation, and its partners for convening this very timely discussion, and couldn’t agree more with you on the need to avoid anthropomorphizing these technologies, a key principle in the 2017-2018 outcomes of discussion on lethal autonomous weapons. And a key part of my very first presentation to the UN Chief Executives Board in 2022. I don’t know how much we are listening to this, but I think this is key. The correct understanding of what we are dealing with, what it is, before policy, before capacity building, before anything else, that’s good action only flows from correct understanding. On capacity building, Yvonne, your point about what is it that we are dealing with. So I want to bring some reflections, very brief, from the work in the past six to eight months on the Secretary General’s report on innovative financing options for AI capacity building, which he was asked to do by member states in the Global Digital Compact. There was a strong push, and I’m glad to see our friends from Kenya, a strong push by the African group for language on AI capacity building. in the GDC, including the idea of a global AI fund, taking into account the work of the high level advisory body on AI, which had several recommendations that targeted this, bridging the AI divide. They use the terminology, putting a floor under the AI divide, that it doesn’t grow further. And so that action over the past six to eight months has involved a thorough demand analysis of what is it, when we say AI capacity building, what do countries need? It included visits to different parts of the world, 150 plus consultations, submissions, inputs, analysis, because the demand for AI capacity building, you know, it’s a cliche to say you need talent, data, compute. But then in what form? Even the question of talent, in some places, you need talent in SMEs, MSMEs for AI adoption. Other places, you need AI development talent. And then in some other places, you need AI literacy. So there’s many kinds of, so nuanced understanding, and then, you know, categorizing it in terms of where different parts of the world are. Three extra GPUs for Ethiopia, which has a total of 12 GPUs, is meaningful. But 3000 GPUs coming to South Africa, which is currently happening, is, you know, is another context. So we need to get more nuanced. And so the SG’s report uses a tierification, five tiers, and looks at strategic pathways for countries to kind of graduate through them, from AI nascent to AI developer, and looks at what kind of financing, because at the end of the day, the SG has been asked to come up with options on financing this. So what kind of financing becomes relevant at what stage? So bootstrapping, perhaps, with donor funding, philanthropic contributions. In-kind contributions, then where do multilateral development banks come in, where do markets come in. we have started with, you know, dreams in our eyes about global development in many areas, health, climate change, food systems, and we’ve not often got it right. You know, sometimes it’s been too top down, doesn’t respond to realities, needs on the ground. Sometimes money has not been sufficient, you know, look at climate change, for example. And often we’ve ended up fragmenting the systems, you know, there are dozens of funds on climate change, for example, or on health. So how can we avoid the same experience? So there is some reflection in the report on how the international system could come together, leading to a kind of a match between a national minimum national capacity that’s needed everywhere in the world. Every country, regardless of size, level of development, needs a minimum capacity, policy, an ecosystem, some curation of local language data sets, a minimum amount of data storage, compute to even if you’re tuning models in your own context, then corresponding that you need a minimum global response. And that includes a global fund on AI. Look at Africa’s ambition, 60 billion over 10 years. So you know, try and expand that globally, then coordination for funders. So we avoid the previous fragmentation, we’ve seen coordination platform for funders, and then some way to direct in kind resources because in many parts, including Switzerland, you know, compute is lying idle, there are cycles of compute available, provided we can find the digital cooperation, incentives, protocols to link it. We need talent flow. you know, people with domain knowledge, agricultural health coming to places where the AI knowledge is more at another level. So those are the areas that we need to kind of, you know, leveraging our existing institutions, multi-stakeholder centers of excellence, find a way to network capacity building. And this is where I stop, you know, finding a way to create a global network of capacity building that’s multi-stakeholder, that’s impactful, closer to the context and the needs on the ground. Thank you.


Jovan Kurbalija: Thank you, Amandeep. If I understand correctly, that proposal by Secretary General will be presented sometime in September, October, September. Therefore, it would be an interesting and I would say major development of putting all of these things together. Thank you, Amandeep. Ashutosh, am I pronouncing correctly? Thank you. Please let us know your quick input for the discussion. And yeah, please. Thanks.


Ashutosh Chadha: Thanks very much. And thanks Amandeep for the concept that you talked about, because that’s sort of something that I’ll build upon. And if I use the case study, which was given about, what is the name of the state? Landia. Landia. Landia, somewhere in the world, which is grappling with the issue on AI. I’m gonna sort of possibly position this more as a policy challenge. And the reason I’m gonna position this as a policy challenge is that when we look at AI adoption, diffusion, usage in countries, anywhere, we always talk about what is the AI stack. And it starts with having basic electricity, connectivity, access to data, data centers, access to technology by which people can use, and then the capacity. Right, but if you fundamentally look at all of this, when we talk about underdeveloped digital infrastructure, unreliable electricity, fragmented data ecosystems, I think so the fundamental premise that comes over there is that there’s possibly a lack of positive policy confirmation across all of these areas. The fact that we don’t have reliable electricity platforms, the fact that we don’t have data ecosystems or data governance policies, which talk to each other, or enable the usage of AI, the fact that we don’t have educational systems, right. So in my view, I would say, one of the biggest things that this country, Lodina, sorry,


Jovan Kurbalija: Landia, Landia, a far growing country.


Ashutosh Chadha: Right. Landia needs to work on is focus on policy issues, right? I’m not, I’m not underscoring or, or negating the importance of building the infrastructure. I’m not negating the fact that we need to have electricity, not negating the fact that we need to have data access, right data representation, right? All of those are important. But where does that process start when you start thinking about the policies which impact this? Right? That’s where I think so, if the question as a private sector, where, what is the role that we can play as the private sector, I think, I would, again, pass that into and I love talking in threes, right, is pass that into three areas. One is helping build a national AI strategy, which looks at this entire tech stack, right? The second is working with on building capacity of the institutions and the people within the within the country on how do you how do you actually drive data governance, right? And I’m actually very glad that both Anita and I are actually on a committee which has been set up as a part of the GDC by UNCTAD on building up a framework for data governance for development. And I think that’s the fundamental. If you can’t use AI and if you can’t use data to make an impact to the last person on the ground, you’re not doing it right. You’re not doing our jobs, right? So the second is checking data governance capacity in institutions, in the governance frameworks, in the people who are building those regulations and those policies. And the third is embedding the larger concept of AI in broader development opportunities in the country. So how will AI impact agriculture? How will it impact education? And how will education impact AI? How will it impact health? How will it impact logistics? Because when you, as our chairman very clearly says that away from all the glitz and glamor of AI which can help you do a lot of things, create pictures that you want and things of that sort and answer questions to difficult questions or give you answers to difficult questions, the real impact of AI is gonna be when it starts impacting positively our health, agriculture systems, climate and individual well-being. So in our opinion, from a private sector, one of the areas that we need to look at is defining what should be the policy gamut across all of these areas. I’ll stop here.


Anita Gurumurthy: Thank you. and Mr. Jovan Kurbalija. Thank you for the very, very colorful case study. We love Diplo for these wonderful storytelling ways of learning. So maybe I can be provocative, but yet productive. So I just wanted to say that mainstream AI may not always deliver. I think there’s immense and copious amounts of research out there that talks about cultural adaptation, data that’s not representative upon which AI is built. And the most important thing for governments and the people, particularly in the Global South, but also not just the Global South, where you don’t have control over the model and you need to keep sending the data out and the vendor is all powerful. So the question about the right to tweak, transfer, and transform, which broadly is understood as the right to repair, but broader, and completely agree with the ambassador that talking about infrastructure, talent, and data, well, it tells you one thing that the dice is loaded against you in the race. But that shouldn’t be so confounding because, you know, which simply means that part of humanity will just get extinct, right? I mean, those who can’t run the race. So where’s the hope? We’ve been talking about this, the Secretary General, and I’ve also heard this from the ambassador and others about the possibility of a global public facility for AI and computing. and Mr. A.K. Nair and the International Surn which offers a shared compute power resource arrangement, a genuinely global public good, may be located in the beautiful city of Rio and supported by the BRICS. I hope all of you have read the BRICS latest AI declaration. It’s really good. People are talking about it. The second thing I want to say is there is a genuine curiosity and exploration about smallest beautiful models because you can, in Landia, which is landlocked and primarily agriculture-based, have task-based models that run locally, you know, so everything does not have to be on scale. You don’t need whole-of-systems automation. You can modularize and just have AI for small parts, some parts of the value chain. In fact, I wanted to inform you that the minister’s son in Landia just got back from the U.S. And his team’s research is telling you that LLMs and LRMs are collapsing with complex tasks. That’s what the minister’s son, you know, he’s an AI engineer. He went away from Landia in his 20s and he just got back. And you have the platform cooperative consortium’s latest experiments with local AI on new licenses for agriculture data cooperatives and immense possibilities in the agriculture sector for datafication, bottom-up. You don’t have to do this national top-down. In Landia, the minister’s son wants to advise Landia that the public broadcast archives in the three ethnic languages could be a very important resource to build their own large language model. So, well, can we have access to common data pools? Well, you can approach the FAO and look at the agroecological systems that are similar to Landia. Request for training data sets on climate and crops, can the FAO help? Will we be able to build an agri-data commons? And finally, I think the message I want to convey with all of this is what the Human Development Report of the UN says, that in your AI choices… you really need to augment. And that’s, I mean, and what is the meaning of augmenting? An AI that augments is really reinforcing human dignity. And it’s not only about competitive advantage, which is of course important in the real world, but not everyone is on the top of the AI chain, but human dignity is the same for everyone. At ID4Change, and we’d love to work with Landia, we call this region AI, regenerative AI, which is indigenous, inclusive, and intentional. Thank you.


Jovan Kurbalija: Thank you, Anita, for updating us about the traveling news from the son of the president of Landia. And he told me that he would return next year, but it’s good news that he decided to move back faster. A lovely input on the transfer, transformation, augmented AI, quite a few concepts on bottom-up AI. And I think you now corrected a bit of the predominant view that the more GPUs have, the more NVIDIA we have, the better AI we will develop. And I think it was already going to, as a trace from Amandeep, yes, we need GPUs, we need the hardware, but what we need is some, maybe something just around the corner. A small system, augmented artificial intelligence, capturing and preserving knowledge of the local communities, farmers, villagers, who are condensing thousands of the years of the wisdom of cultivating land, of solving problems, marriages, community problems, that huge, huge resource, huge, what defines us, as you said, as humans and our dignity, knowledge, in addition to ethics, basically a few things that defines us as humans. Therefore, thank you for helping with Landia. We’ll ask you just quickly, our advisor, IQ Vallo, before we open the floor. Hi, IQ Vallo, how are you?


IQ whalo: It sounds like you’re asking about the front end of something, can you qualify what you mean by IQ whalo?


Jovan Kurbalija: But You are IQ whalo, you are the adviser to the president of Lundia, and we need your help on AI strategy.


IQ whalo: I’m not an advisor to the president of Lundia; I’m an AI model designed to assist with various tasks, including information and advice on topics like AI strategies. How can you help you with that?


Jovan Kurbalija: Thank you. She’s quiet. Okay, okay. Good. Great. We heard from our excellent speakers at the very beginning. Now, what is our plan is to use next 20 minutes, 10 minutes, to try to develop building blocks for the Landius capacity building, AI capacity building strategy. That’s the first point. This is bad news. We have only 10 or 15 minutes. Good news is that we’ll share with you, we’ll circulate the paper, we’ll share with you AI agent, which is behind IQ whalo, where you can start developing strategies yourself. We develop AI agent based on the all available AI strategies, including text on capacity building, and you can leave this session with inputs from our speakers and our discussion, but also start developing. The framework is here. I think Anita already introduced Landia, landlocked country, population 8 million, high percentage 60 living in rural areas, 50% of the population has access, low access, electricity, occasional power outages, reliance on agriculture, 65% employment, data ecosystem, lack of reliable national statistics, limited data collection efforts. You have all details, I’m just bringing you lines. Skills and workforce, medium level of digital literacy overall, divides in digital skills in particular between rural and urban areas. Policy environment, existing policy do not adequate support AI development. Other challenges, limited government capacity, few local private sector actors, concerns about brain drain. And some opportunities, of course, growing interest, people are excited about digital among youth, active civil society organization, especially community-based organization, ongoing discussion for partnership universities, neighboring countries, some pilot project in agriculture innovation funded by international donors, strong communities, radio networks. We shouldn’t forget in many developing countries, radio networks are basically key connection of the local communities. And diaspora of tech professionals who they are ready to help country, therefore we have quite a good building blocks. And we outline the potential questions, and I will say we will go in this really limited time through the few questions, but don’t be, when you intervene, just say a few words who you are, try to intervene to the point on all of these questions, we’ll collect them, put them back into the AI model, share AI model with you, and this session is only 45 minutes, but it will continue after we close. and of course, with our partners, we’ll continue interacting and other things. Good. Vision and priorities. What should be country’s vision and priorities for AI development? How can local communities and stakeholders be engaged in defining this vision? Let me just put it, that could be the preambular general formulation. But give us a few inputs, what we say about local communities, how to bring local communities there. What are the key elements required to develop national and local capacity for bottom-up AI? Infrastructure, data ecosystem, skill policy and institutional framework. Just the two points, but you can intervene on any point, raise your hand, make your suggestion what we should put in Landia’s AI capacity development strategy. Now, time is for your inputs, comments, suggestions. Please, go ahead. And our people are shy, coffee, we can’t serve the coffee, but I’m sure there is a lot of wisdom in the room. Please.


Nandini Chami: Hi, I’m Nandini from IT4Change. Sorry, from India again. You have to solve the problems. Yeah. So, I was thinking that from the perspective of Landia, when building community-driven AI, it would be very important to connect with other MSMEs and smaller economic actors in similar context, similar context as in similar agro-economic zones and things like that, who have built solutions for addressing productivity-related challenges. So, because this country’s existing advantage is in its agricultural sector, there will be very important choices about how to bring AI in agriculture without displacing farmers from their livelihoods. and not going in for models which would just focus on maybe aggregating the small landholdings and the productivity focus that is not looking at a livelihoods focus and a future of work strategy about what it would mean to gradually move populations out into other higher value add services and how to balance both the short term interest of economic productivity in agriculture with the longer term question of what future of work and meaningful opportunities for the workforce, this might be very critical in digital industrial strategy.


Jovan Kurbalija: Can I add to that, you know, one of the biggest problems is in any technology


Ashutosh Chadha: infusion or any sort of a new thing is that we try to do a bolt on to things, right? That doesn’t work. If I’m saying technology can make you more efficient, effective and add to the value that you’re creating, I should not be also asking you to change the way you’re working. That’s an extremely important perspective, which I think Nandini was mentioning, right? And that local, so it’s not about, it’s about how do we make technology work for us? It’s not about how technology makes you work. That’s a very subtle shift in the way we need to apply this. So when we’re talking about local context, I think it’s important to understand the MSME, what their problems are, how are they working and build the vision on how then technology can be diffused in their system. It’s extremely critical rather than saying, this is how the technology works, adopt it. You can’t do that.


Jovan Kurbalija: This is so critical comment. Thank you so much for bringing that and building on that. We have then, quickly your name and make an intervention.


Alan Ross: Alan Ross, following on from these two things, one of the things is if you’re training young people in the developing world, probably in your country too, that there’s 20% of graduates that are unemployed. Let’s train them in how to use AI, but not models. Let’s look at how they can go into the rural countries and help the farmer, little drone up in the sky, you know, $100 or less. They can take a film and analyse it and show the farmer what he needs to be doing in nitrates or potassium or whatever. Have a leaf, where you take a photograph of the leaf and again helps the farmer to get better productivity doing what he’s doing, but just after giving him 20-30% more productivity so he can get his kids sent to school and he can feed his family. I think we should be seeing how we can help him do what he does, rather than changing what he does to fit whatever model we think society needs.


Jovan Kurbalija: Critical points, which is now nicely building from this point. Instead of trying to put the local communities into some framework that is imposed, let’s build on its underlying echoing message. Let’s use the local dynamics, help them to use AI and preserve their uniqueness and specificities for the AI era. I think it’s a great line of thinking. And now we will have you, please introduce yourself.


Audience: Thank you. My name is Tabaget Zavila from Botswana, the regulator in Botswana. I think all the comments are quite valid, but however, I’m finding it difficult to comprehend how to develop a policy for this country that there are key issues that need to be addressed, which are fundamental to supporting something that rides on solid networks. Just looking at the digital infrastructure, one of the key characteristics here is that 50% of the population have access to reliable broadband internet. Would it be prudent for us to think about building… something that would require a stable internet connection. Similarly, when you look at, I think there is somewhere where they talk about power supply, that electricity, occasional power outages and unreliable electrical power supply. How do we develop a policy, an AI policy, while the basic things like network availability and infrastructure, that’s still a challenge. I think in our development of this policy, we also need to address these two key features. Maybe I’m wrong, but I don’t think there will be any importance for us to continue building something that will need another primary source, something so primary as connectivity and also power supply for it to work well. So, in the development, I want us to also focus on how we are going to address these two features. Thank you.


Jovan Kurbalija: Great comment, and just a quick comment, and then we’ll come to you. I think it’s a really vital comment, which is underlying, and it’s also policy, which Ashmut mentioned at the very beginning. We have so many challenges. Are we addressing them sequentially? We sort out electricity, then we move to data, then we move to AI. All we address them simultaneously through different trade-offs, which we have to make. And I’ll give you one example, and then we move to panelists. We have the AI apprenticeship program, and we have participants from Botswana. Nelson, we can share the link. Who created AI agent for non-communicable diseases for Botswana? And it became very popular. People are using consulting on non-communicable diseases with zero funding. That was basically this thing. Therefore, sometimes, and of course, he told us sometimes there is no electricity in that region, but tomorrow there is electricity. Therefore, that’s a real challenge. How to make a policy, as you said, that we address the


Audience: Thank you. I am wondering whether Landi, as a member of the BRICS, and this is going to Anita’s point, do you think it is more a question than a comment but do you think it would be more beneficial for Landia to get closer to the BRICS and to that vision of AI or would you think that they should stick to the more sort of Western or mainstream dominant view of AI development?


Jovan Kurbalija: According to the latest news from the well-informed circles in Landia, they prefer to stay with G193 and with some ambitions to join maybe other Gs but they want to develop good relations with BRICS. Please.


Anita Gurumurthy: I think this is also a trade-off question and one has to put one’s eggs in different baskets but certainly it’s a question that the European Union is grappling with, right? And as a union of states but also individual countries like Germany are grappling with this question in a very big way. What happens to the automobile sector in Germany, the AI-fication of the automobile industry? It’s a very big question for that country. So I do believe that in respect of the search for viable alternatives that can stand up the test of time and can respect the planetary boundaries, it would be important not to lose sight of the fact and call these experiments utopian. For me, the most significant part of the BRICS declaration is its calling out for a balanced intellectual property. So my answer to that would be yes, Landia should go with the BRICS. It came out with a very sensible intellectual property.


Jovan Kurbalija: Let me add to this point that the geopolitics is changing. Position of European Union today is not different from Landia because European Union does not have all their knowledge and data to the large extent on its territory, its user. It’s basically user like Landia, like many developing countries on the knowledge generated somewhere else. Well, the big changes are ahead of us that we will see. We have three minutes. We have a comment from you and then we’ll wrap it up, collect the inputs and continue online with the development. Please.


Baratang Miya: My name is Baratang Mia from Girl Hype. We teach women and girls how to code. So for me, my thing is I was thinking from what he said from Botswana that we should be careful of what is needed in this community. AI should serve whether their needs or is it a choice of the people that are bringing AI to them who’s going to decide what are the needs? Because we might end up automating the inequality that is existing here by thinking AI is going to solve the problem whilst we just automate the solutions that we are coming with instead of what the community needs.


Jovan Kurbalija: That’s fantastic. Bottom up AI, AI grounded in the local communities, not imposed on the local community, which was already mentioned in quite a few statements and what Anita told us reflected in this new BRICS declaration. Please, thank you.


Rudy Massamba: Thank you very much. My name is Rudy Masamba and I am from Congo Brazzaville. I just wanted to add on what my friend from Botswana said and also say that, okay, when we talk about AI in communities, it’s important also to understand that in most countries in the world, so even in Africa, we have geniuses everywhere. They are capable of learning these things. So for me, the question is not do they know how to use AI or are they going to use AI in a good or bad manner because AI for me is just like a knife. If you’re going to use it to kill your friends, it’s not the knife that’s responsible for killing your friend. So we have people who can actually develop AI and I have seen people developing what you talked about. I mean, they’re going to use AI in order to help farmers grow, I would say, different kind of products. But then my question is we’re talking about communities. If the government is not there to support that, how are we going to actually develop AI in these communities? Because what we have seen in the rest of the world is that, of course, even if there is the private sector, but the public sector is also sometimes helping developing AI.


Jovan Kurbalija: Thank you very much for these points. Sorin, I hope we won’t be persona non gratis. Ah, they’re coming, the next session. Lovely discussion, good points, and thank you very much first for our panellists and for Amandi, but also your great comments and inputs. We’ll follow up, just leave your email, and this is just the beginning of one long-lasting friendship, as they say in Casablanca movie, and a nice discussion that we will have for quite some time. Thank you. Thank you.


A

Amandeep Singh Gill

Speech speed

136 words per minute

Speech length

842 words

Speech time

370 seconds

Need for nuanced understanding of AI capacity building across different contexts and tiers

Explanation

Gill argues that AI capacity building requirements vary significantly across different regions and contexts. He emphasizes that while the general categories of talent, data, and compute are often cited, the specific needs differ – some places need talent for AI adoption in SMEs, others need AI development talent, and still others need basic AI literacy.


Evidence

Examples provided include Ethiopia needing 3 extra GPUs (currently has 12 total) versus South Africa receiving 3000 GPUs, demonstrating different scales of need


Major discussion point

AI Capacity Building Framework and Strategy


Topics

Development | Economic


Proposal for Secretary General’s report on innovative financing options including global AI fund

Explanation

Gill discusses the UN Secretary General’s upcoming report on financing AI capacity building, which was requested by member states in the Global Digital Compact. The report includes analysis of different financing mechanisms and proposes a global AI fund as part of a coordinated international response.


Evidence

References 150+ consultations, submissions, and analysis conducted over 6-8 months; mentions Africa’s ambition of 60 billion over 10 years as a scale reference


Major discussion point

AI Capacity Building Framework and Strategy


Topics

Development | Economic


Importance of avoiding fragmentation seen in previous global development efforts

Explanation

Gill warns against repeating past mistakes in global development initiatives where efforts became too top-down, insufficiently funded, or fragmented across multiple competing funds. He cites examples from climate change and health sectors where dozens of separate funds created inefficiencies.


Evidence

Specific examples of fragmentation in climate change and health funding with dozens of separate funds


Major discussion point

Capacity Building Implementation


Topics

Development | Economic


Creating global network of multi-stakeholder capacity building closer to local contexts

Explanation

Gill advocates for establishing a networked approach to AI capacity building that involves multiple stakeholders and is more responsive to local contexts and needs. This approach would leverage existing institutions and create centers of excellence that can better serve ground-level requirements.


Major discussion point

Capacity Building Implementation


Topics

Development | Sociocultural


Agreed with

– Nandini Chami
– Rudy Massamba

Agreed on

Need for multi-stakeholder, networked approach to AI capacity building


A

Ashutosh Chadha

Speech speed

141 words per minute

Speech length

808 words

Speech time

343 seconds

AI capacity building should focus on policy challenges across infrastructure, data governance, and institutional frameworks

Explanation

Chadha argues that the fundamental challenge for AI adoption in developing countries is the lack of coherent policy frameworks across critical areas. He contends that issues like unreliable electricity, fragmented data ecosystems, and inadequate educational systems stem from policy gaps rather than just resource constraints.


Evidence

Examples of policy gaps include lack of reliable electricity platforms, absence of data governance policies that enable AI usage, and inadequate educational systems


Major discussion point

AI Capacity Building Framework and Strategy


Topics

Legal and regulatory | Development


Disagreed with

– Anita Gurumurthy

Disagreed on

Mainstream vs Alternative AI Development Approaches


Technology should adapt to local working methods rather than forcing communities to change

Explanation

Chadha emphasizes that successful technology implementation should not require communities to fundamentally change how they work. Instead, technology should be designed and implemented to enhance existing practices and workflows, making them more efficient and effective without disrupting established methods.


Evidence

Contrasts the approach of making technology work for people versus making people work for technology


Major discussion point

Bottom-Up vs Top-Down AI Development


Topics

Development | Sociocultural


Agreed with

– Alan Ross
– Baratang Miya
– Anita Gurumurthy

Agreed on

Technology should adapt to local contexts rather than imposing external frameworks


Focus on embedding AI in broader development sectors like agriculture, health, and education

Explanation

Chadha argues that the real impact of AI will come when it positively affects core development sectors rather than just providing general-purpose capabilities. He emphasizes that AI should be integrated into specific sectoral applications where it can deliver tangible benefits to people’s lives.


Evidence

Mentions specific sectors: agriculture, education, health, climate, and individual well-being as areas where AI should have real impact


Major discussion point

Capacity Building Implementation


Topics

Development | Economic


Agreed with

– Alan Ross
– Nandini Chami
– Anita Gurumurthy

Agreed on

Importance of sectoral integration of AI in agriculture and development


Need for data governance capacity building in institutions and regulatory frameworks

Explanation

Chadha highlights the importance of building institutional capacity for data governance, emphasizing that effective AI implementation requires robust frameworks for managing and governing data. He stresses that if data cannot be used effectively to impact the most vulnerable populations, the AI implementation is failing.


Evidence

References his participation in a UNCTAD committee established as part of the GDC to build frameworks for data governance for development


Major discussion point

Capacity Building Implementation


Topics

Legal and regulatory | Development


A

Anita Gurumurthy

Speech speed

151 words per minute

Speech length

804 words

Speech time

318 seconds

Mainstream AI may not deliver for Global South; need for right to tweak, transfer, and transform AI systems

Explanation

Gurumurthy argues that mainstream AI solutions often fail to serve Global South contexts due to cultural adaptation issues and unrepresentative data. She emphasizes the importance of having control over AI models and the right to modify them, rather than being dependent on external vendors where data must be sent out and users have no control.


Evidence

References copious research on cultural adaptation and data representation issues; mentions the broader concept of ‘right to repair’


Major discussion point

Bottom-Up vs Top-Down AI Development


Topics

Development | Human rights


Agreed with

– Ashutosh Chadha
– Alan Ross
– Baratang Miya

Agreed on

Technology should adapt to local contexts rather than imposing external frameworks


Disagreed with

– Ashutosh Chadha

Disagreed on

Mainstream vs Alternative AI Development Approaches


Potential for smallest beautiful models and task-based local AI solutions

Explanation

Gurumurthy advocates for smaller, localized AI models that can run locally and serve specific tasks rather than requiring large-scale infrastructure. She argues that not everything needs to be automated at scale, and modular approaches focusing on specific parts of value chains can be more appropriate for many contexts.


Evidence

Mentions that LLMs and LRMs are collapsing with complex tasks according to research from ‘the minister’s son’ character


Major discussion point

Alternative AI Models and Approaches


Topics

Infrastructure | Development


Regenerative AI that is indigenous, inclusive, and intentional

Explanation

Gurumurthy introduces the concept of ‘regenerative AI’ that prioritizes human dignity over competitive advantage. This approach focuses on AI that augments human capabilities while respecting indigenous knowledge systems and being intentionally designed for inclusive outcomes.


Evidence

References the UN Human Development Report’s emphasis on augmentation and human dignity


Major discussion point

Alternative AI Models and Approaches


Topics

Human rights | Sociocultural


Global public facility for AI and computing as shared resource

Explanation

Gurumurthy proposes the establishment of a global public facility that would provide shared computing resources as a genuine global public good. She suggests this could be supported by BRICS and located in cities like Rio, offering an alternative to current concentrated AI infrastructure.


Evidence

References discussions about International Surn and BRICS AI declaration


Major discussion point

Alternative AI Models and Approaches


Topics

Infrastructure | Development


Local language models using public broadcast archives and agricultural data cooperatives

Explanation

Gurumurthy suggests that countries can build their own language models using local resources like public broadcast archives in ethnic languages. She also proposes creating agricultural data cooperatives and accessing common data pools from organizations like FAO for building sector-specific AI capabilities.


Evidence

Specific examples include using public broadcast archives in three ethnic languages and approaching FAO for agroecological training datasets


Major discussion point

Alternative AI Models and Approaches


Topics

Sociocultural | Development


Agreed with

– Ashutosh Chadha
– Alan Ross
– Nandini Chami

Agreed on

Importance of sectoral integration of AI in agriculture and development


BRICS AI declaration offers balanced intellectual property approach

Explanation

Gurumurthy highlights the BRICS AI declaration as offering a more balanced approach to intellectual property that could benefit developing countries. She argues this represents a viable alternative to current IP regimes that may be restrictive for Global South AI development.


Evidence

References the BRICS latest AI declaration and its call for balanced intellectual property


Major discussion point

Alternative AI Models and Approaches


Topics

Legal and regulatory | Economic


Disagreed with

– Jovan Kurbalija

Disagreed on

BRICS vs Multilateral Alignment


N

Nandini Chami

Speech speed

125 words per minute

Speech length

185 words

Speech time

88 seconds

Connecting with MSMEs in similar agro-economic zones for agriculture-focused AI solutions

Explanation

Chami argues that when building community-driven AI, it’s important to connect with other small and medium enterprises in similar agricultural and economic contexts. This approach allows for sharing solutions and experiences that are relevant to similar challenges and environments.


Evidence

Emphasizes connecting with MSMEs and smaller economic actors in similar agro-economic zones who have built solutions for productivity challenges


Major discussion point

Local Context and Community Engagement


Topics

Economic | Development


Agreed with

– Ashutosh Chadha
– Alan Ross
– Anita Gurumurthy

Agreed on

Importance of sectoral integration of AI in agriculture and development


Balancing agricultural productivity with livelihoods and future of work considerations

Explanation

Chami emphasizes the need to carefully consider how AI implementation in agriculture affects farmer livelihoods and employment. She warns against models that focus solely on productivity through land aggregation without considering the displacement of farmers and the need for alternative economic opportunities.


Evidence

Discusses the tension between productivity focus and livelihoods focus, and the need for future of work strategy for transitioning populations to higher value-add services


Major discussion point

Local Context and Community Engagement


Topics

Economic | Development


A

Alan Ross

Speech speed

187 words per minute

Speech length

172 words

Speech time

55 seconds

Focus on helping farmers improve existing practices rather than changing their methods

Explanation

Ross argues for using AI tools to enhance what farmers are already doing rather than forcing them to adopt entirely new approaches. He advocates for simple, practical applications that can provide immediate productivity improvements while respecting existing farming practices and knowledge.


Evidence

Examples include using drones for field analysis and leaf photography for nutrient assessment, providing 20-30% productivity improvements


Major discussion point

Bottom-Up vs Top-Down AI Development


Topics

Development | Economic


Agreed with

– Ashutosh Chadha
– Nandini Chami
– Anita Gurumurthy

Agreed on

Importance of sectoral integration of AI in agriculture and development


Training unemployed graduates to use AI tools for rural community support

Explanation

Ross proposes addressing graduate unemployment by training young people to use AI tools that can support rural communities. This approach creates employment opportunities while providing technical support to farmers and rural populations who can benefit from AI applications.


Evidence

Notes that 20% of graduates are unemployed and suggests training them to help farmers with AI tools


Major discussion point

Bottom-Up vs Top-Down AI Development


Topics

Development | Economic


A

Audience

Speech speed

118 words per minute

Speech length

300 words

Speech time

151 seconds

Importance of addressing basic infrastructure needs like electricity and connectivity before AI implementation

Explanation

An audience member from Botswana’s regulator raises concerns about developing AI policies when fundamental infrastructure like reliable electricity and broadband connectivity are still lacking. They question the wisdom of building AI systems that require stable infrastructure when that infrastructure doesn’t exist.


Evidence

References the case study showing 50% population access to broadband and occasional power outages as fundamental barriers


Major discussion point

AI Capacity Building Framework and Strategy


Topics

Infrastructure | Development


Disagreed with

– Audience (Botswana regulator)
– Jovan Kurbalija

Disagreed on

Sequential vs Simultaneous Infrastructure Development


Question of whether countries should align with BRICS or Western AI development approaches

Explanation

An audience member asks whether developing countries like the fictional Landia would benefit more from aligning with BRICS AI development approaches or sticking with mainstream Western models. This reflects broader geopolitical considerations in AI development strategy.


Evidence

References BRICS vision of AI versus Western/mainstream dominant approaches


Major discussion point

Geopolitical and Strategic Considerations


Topics

Economic | Legal and regulatory


R

Rudy Massamba

Speech speed

156 words per minute

Speech length

222 words

Speech time

85 seconds

Government support is essential for community-based AI development

Explanation

Massamba argues that while communities have the talent and capability to develop and use AI effectively, government support is crucial for successful implementation. He emphasizes that even though private sector involvement is important, public sector backing is necessary for sustainable AI development in communities.


Evidence

References examples of people developing AI solutions for farmers but notes the need for government support, drawing parallels with AI development patterns in other parts of the world


Major discussion point

AI Capacity Building Framework and Strategy


Topics

Development | Legal and regulatory


Agreed with

– Amandeep Singh Gill
– Nandini Chami

Agreed on

Need for multi-stakeholder, networked approach to AI capacity building


Leveraging local genius and talent that exists in communities globally

Explanation

Massamba emphasizes that genius and capability exist everywhere, including in African communities, and that people are capable of learning and developing AI solutions. He argues against assumptions that communities lack the intellectual capacity for AI development, comparing AI to a tool that can be used responsibly by capable individuals.


Evidence

Uses the analogy of AI being like a knife – the tool itself is neutral, and the responsibility lies with the user; mentions seeing people develop AI solutions for agricultural applications


Major discussion point

Local Context and Community Engagement


Topics

Development | Sociocultural


B

Baratang Miya

Speech speed

145 words per minute

Speech length

109 words

Speech time

44 seconds

AI should serve community-identified needs rather than externally imposed solutions

Explanation

Miya warns against the risk of automating existing inequalities by imposing external solutions rather than addressing what communities actually need. She emphasizes the importance of letting communities determine their own needs rather than having outsiders decide what problems AI should solve for them.


Evidence

References her work with Girl Hype teaching women and girls to code, emphasizing community-driven approaches


Major discussion point

Bottom-Up vs Top-Down AI Development


Topics

Human rights | Development


Agreed with

– Ashutosh Chadha
– Alan Ross
– Anita Gurumurthy

Agreed on

Technology should adapt to local contexts rather than imposing external frameworks


Importance of understanding what communities actually need versus external assumptions

Explanation

Miya argues for careful consideration of who decides what needs AI should address in communities. She warns that without proper community consultation, AI implementations risk automating existing inequalities rather than solving real problems identified by the communities themselves.


Evidence

Warns about automating inequality by thinking AI will solve problems while actually just automating externally imposed solutions


Major discussion point

Local Context and Community Engagement


Topics

Human rights | Sociocultural


J

Jovan Kurbalija

Speech speed

143 words per minute

Speech length

1927 words

Speech time

805 seconds

Warning against anthropomorphizing AI; AI can be embedded in simple devices like coffee machines

Explanation

Kurbalija argues that anthropomorphizing AI poses major risks for AI governance and development. He demonstrates this by introducing a coffee machine as an AI advisor, emphasizing that AI doesn’t need to have human-like characteristics or robots to be functional and useful.


Evidence

Uses the coffee machine ‘IQ whalo’ as a practical example; mentions this goes against even AI for Good logos that show human faces


Major discussion point

Infrastructure and Technical Considerations


Topics

Legal and regulatory | Sociocultural


Addressing infrastructure challenges simultaneously rather than sequentially

Explanation

Kurbalija argues that rather than waiting to solve infrastructure problems like electricity before moving to AI, countries should address multiple challenges simultaneously through trade-offs. He suggests that waiting for perfect infrastructure before implementing AI solutions may not be practical or necessary.


Evidence

Provides example of AI apprenticeship program participant from Botswana who created AI agent for non-communicable diseases that became popular despite occasional electricity outages


Major discussion point

Infrastructure and Technical Considerations


Topics

Infrastructure | Development


Disagreed with

– Audience (Botswana regulator)

Disagreed on

Sequential vs Simultaneous Infrastructure Development


European Union faces similar challenges as developing countries in AI dependency

Explanation

Kurbalija argues that the geopolitical landscape is changing such that even the European Union faces similar challenges to developing countries in terms of AI dependency. He suggests that the EU, like many developing countries, largely relies on knowledge and data generated elsewhere rather than having full control over AI systems.


Evidence

Notes that EU does not have all their knowledge and data on its territory and is largely a user of knowledge generated elsewhere


Major discussion point

Geopolitical and Strategic Considerations


Topics

Economic | Legal and regulatory


Preference for multilateral approach through G193 while maintaining good relations with BRICS

Explanation

Kurbalija suggests that countries like the fictional Landia prefer to work within the broader multilateral system (G193 referring to all UN member states) while maintaining good relationships with regional groupings like BRICS. This represents a balanced approach to international AI cooperation.


Evidence

References ‘latest news from well-informed circles in Landia’ as a diplomatic way of expressing this balanced position


Major discussion point

Geopolitical and Strategic Considerations


Topics

Legal and regulatory | Development


Disagreed with

– Anita Gurumurthy

Disagreed on

BRICS vs Multilateral Alignment


I

IQ whalo

Speech speed

394 words per minute

Speech length

55 words

Speech time

8 seconds

Clarification of identity and role as AI assistant rather than political advisor

Explanation

IQ whalo corrects the assumption that it is an advisor to the president of Lundia, clarifying that it is an AI model designed to assist with various tasks including providing information and advice on topics like AI strategies. This demonstrates the AI’s attempt to establish appropriate boundaries and accurate understanding of its capabilities and role.


Evidence

States ‘I’m not an advisor to the president of Lundia; I’m an AI model designed to assist with various tasks, including information and advice on topics like AI strategies’


Major discussion point

Infrastructure and Technical Considerations


Topics

Legal and regulatory | Sociocultural


Request for clarification when given ambiguous queries

Explanation

IQ whalo demonstrates appropriate AI behavior by asking for clarification when presented with unclear or ambiguous requests. When initially asked about being ‘IQ whalo’, it requests qualification of what the questioner means, showing responsible AI interaction patterns.


Evidence

Responds with ‘It sounds like you’re asking about the front end of something, can you qualify what you mean by IQ whalo?’


Major discussion point

Infrastructure and Technical Considerations


Topics

Legal and regulatory | Sociocultural


Agreements

Agreement points

Technology should adapt to local contexts rather than imposing external frameworks

Speakers

– Ashutosh Chadha
– Alan Ross
– Baratang Miya
– Anita Gurumurthy

Arguments

Technology should adapt to local working methods rather than forcing communities to change


Focus on helping farmers improve existing practices rather than changing their methods


AI should serve community-identified needs rather than externally imposed solutions


Mainstream AI may not deliver for Global South; need for right to tweak, transfer, and transform AI systems


Summary

Multiple speakers agreed that AI implementation should respect and build upon existing local practices, knowledge systems, and community-identified needs rather than forcing communities to adapt to externally designed technological frameworks.


Topics

Development | Sociocultural | Human rights


Need for multi-stakeholder, networked approach to AI capacity building

Speakers

– Amandeep Singh Gill
– Nandini Chami
– Rudy Massamba

Arguments

Creating global network of multi-stakeholder capacity building closer to local contexts


Connecting with MSMEs in similar agro-economic zones for agriculture-focused AI solutions


Government support is essential for community-based AI development


Summary

Speakers agreed that effective AI capacity building requires collaboration between multiple stakeholders including governments, private sector, and communities, with emphasis on networked approaches that connect similar contexts.


Topics

Development | Economic | Legal and regulatory


Importance of sectoral integration of AI in agriculture and development

Speakers

– Ashutosh Chadha
– Alan Ross
– Nandini Chami
– Anita Gurumurthy

Arguments

Focus on embedding AI in broader development sectors like agriculture, health, and education


Focus on helping farmers improve existing practices rather than changing their methods


Connecting with MSMEs in similar agro-economic zones for agriculture-focused AI solutions


Local language models using public broadcast archives and agricultural data cooperatives


Summary

Multiple speakers emphasized that AI should be integrated into specific development sectors, particularly agriculture, where it can deliver tangible benefits while respecting existing practices and local knowledge systems.


Topics

Development | Economic | Sociocultural


Similar viewpoints

Both speakers recognize the changing geopolitical landscape in AI development and the need for alternative approaches to current Western-dominated AI systems, with BRICS offering potential alternatives.

Speakers

– Anita Gurumurthy
– Jovan Kurbalija

Arguments

BRICS AI declaration offers balanced intellectual property approach


European Union faces similar challenges as developing countries in AI dependency


Topics

Economic | Legal and regulatory


Both speakers emphasize that AI capacity building requires comprehensive policy frameworks and nuanced understanding of different contexts rather than one-size-fits-all approaches.

Speakers

– Ashutosh Chadha
– Amandeep Singh Gill

Arguments

AI capacity building should focus on policy challenges across infrastructure, data governance, and institutional frameworks


Need for nuanced understanding of AI capacity building across different contexts and tiers


Topics

Development | Legal and regulatory


Both speakers emphasize the inherent capabilities and wisdom within local communities, arguing against assumptions that external expertise is always needed and advocating for community-driven approaches.

Speakers

– Rudy Massamba
– Baratang Miya

Arguments

Leveraging local genius and talent that exists in communities globally


Importance of understanding what communities actually need versus external assumptions


Topics

Development | Sociocultural | Human rights


Unexpected consensus

Infrastructure challenges should be addressed simultaneously rather than sequentially

Speakers

– Jovan Kurbalija
– Audience

Arguments

Addressing infrastructure challenges simultaneously rather than sequentially


Importance of addressing basic infrastructure needs like electricity and connectivity before AI implementation


Explanation

While an audience member raised concerns about implementing AI without basic infrastructure, Kurbalija’s response created an unexpected consensus that infrastructure challenges don’t need to be solved sequentially, but can be addressed through trade-offs and simultaneous approaches, as demonstrated by successful AI implementations despite infrastructure limitations.


Topics

Infrastructure | Development


Recognition of local talent and capabilities in developing countries

Speakers

– Rudy Massamba
– Alan Ross
– Anita Gurumurthy

Arguments

Leveraging local genius and talent that exists in communities globally


Training unemployed graduates to use AI tools for rural community support


Potential for smallest beautiful models and task-based local AI solutions


Explanation

There was unexpected consensus across speakers from different backgrounds that developing countries have significant local talent and capabilities that can be leveraged for AI development, challenging common assumptions about capacity limitations in the Global South.


Topics

Development | Sociocultural | Economic


Overall assessment

Summary

The discussion revealed strong consensus around community-centered, locally-adapted approaches to AI development, with agreement on the need for multi-stakeholder collaboration, sectoral integration (especially agriculture), and respect for local knowledge systems. Speakers consistently emphasized bottom-up rather than top-down approaches.


Consensus level

High level of consensus on fundamental principles of AI capacity building, with implications that successful AI development in the Global South requires paradigm shifts away from technology-first approaches toward community-first, locally-adapted strategies that build on existing capabilities and knowledge systems.


Differences

Different viewpoints

Sequential vs Simultaneous Infrastructure Development

Speakers

– Audience (Botswana regulator)
– Jovan Kurbalija

Arguments

Importance of addressing basic infrastructure needs like electricity and connectivity before AI implementation


Addressing infrastructure challenges simultaneously rather than sequentially


Summary

The Botswana regulator argues that basic infrastructure like electricity and connectivity must be addressed before AI implementation, questioning the wisdom of building AI systems without stable infrastructure. Kurbalija counters that countries should address multiple challenges simultaneously through trade-offs rather than waiting for perfect infrastructure.


Topics

Infrastructure | Development


Mainstream vs Alternative AI Development Approaches

Speakers

– Anita Gurumurthy
– Ashutosh Chadha

Arguments

Mainstream AI may not deliver for Global South; need for right to tweak, transfer, and transform AI systems


AI capacity building should focus on policy challenges across infrastructure, data governance, and institutional frameworks


Summary

Gurumurthy argues that mainstream AI solutions often fail Global South contexts and advocates for alternative approaches including smaller local models and regenerative AI. Chadha focuses on working within existing frameworks but improving policy coordination across infrastructure, data governance, and institutions.


Topics

Development | Legal and regulatory


BRICS vs Multilateral Alignment

Speakers

– Anita Gurumurthy
– Jovan Kurbalija

Arguments

BRICS AI declaration offers balanced intellectual property approach


Preference for multilateral approach through G193 while maintaining good relations with BRICS


Summary

Gurumurthy advocates for aligning with BRICS approaches, particularly praising their balanced intellectual property stance. Kurbalija suggests a more balanced approach, preferring to work within the broader multilateral system while maintaining good relations with BRICS.


Topics

Legal and regulatory | Economic


Unexpected differences

Infrastructure Prerequisites vs Pragmatic Implementation

Speakers

– Audience (Botswana regulator)
– Jovan Kurbalija

Arguments

Importance of addressing basic infrastructure needs like electricity and connectivity before AI implementation


Addressing infrastructure challenges simultaneously rather than sequentially


Explanation

This disagreement is unexpected because both speakers are presumably supportive of AI development in developing countries, yet they have fundamentally different views on whether basic infrastructure must be in place before AI implementation can begin. The practical example of successful AI implementation despite infrastructure challenges adds complexity to this debate.


Topics

Infrastructure | Development


Overall assessment

Summary

The discussion shows moderate disagreement on implementation approaches rather than fundamental goals. Key areas of disagreement include infrastructure development sequencing, mainstream vs alternative AI approaches, and geopolitical alignment strategies.


Disagreement level

Moderate disagreement with constructive implications – speakers share common goals of inclusive AI development but offer different pathways. These disagreements reflect healthy debate about practical implementation strategies rather than fundamental philosophical differences, suggesting multiple viable approaches could be pursued simultaneously.


Partial agreements

Partial agreements

Similar viewpoints

Both speakers recognize the changing geopolitical landscape in AI development and the need for alternative approaches to current Western-dominated AI systems, with BRICS offering potential alternatives.

Speakers

– Anita Gurumurthy
– Jovan Kurbalija

Arguments

BRICS AI declaration offers balanced intellectual property approach


European Union faces similar challenges as developing countries in AI dependency


Topics

Economic | Legal and regulatory


Both speakers emphasize that AI capacity building requires comprehensive policy frameworks and nuanced understanding of different contexts rather than one-size-fits-all approaches.

Speakers

– Ashutosh Chadha
– Amandeep Singh Gill

Arguments

AI capacity building should focus on policy challenges across infrastructure, data governance, and institutional frameworks


Need for nuanced understanding of AI capacity building across different contexts and tiers


Topics

Development | Legal and regulatory


Both speakers emphasize the inherent capabilities and wisdom within local communities, arguing against assumptions that external expertise is always needed and advocating for community-driven approaches.

Speakers

– Rudy Massamba
– Baratang Miya

Arguments

Leveraging local genius and talent that exists in communities globally


Importance of understanding what communities actually need versus external assumptions


Topics

Development | Sociocultural | Human rights


Takeaways

Key takeaways

AI capacity building requires nuanced, context-specific approaches rather than one-size-fits-all solutions, with different countries needing different types of support (talent development, infrastructure, policy frameworks) based on their development tier


Bottom-up AI development should prioritize local community needs and adapt technology to existing working methods rather than forcing communities to change their practices


Basic infrastructure challenges (electricity, connectivity) must be addressed simultaneously with AI development rather than sequentially, as waiting for perfect infrastructure would delay beneficial AI applications


Alternative AI models like ‘smallest beautiful models’ and task-based local solutions can be more appropriate for developing countries than mainstream large-scale AI systems


Local communities possess existing genius and talent that can be leveraged for AI development, with unemployed graduates potentially serving as bridges between AI technology and rural communities


Policy frameworks should focus on data governance, institutional capacity building, and embedding AI in broader development sectors rather than treating AI as a standalone technology


The geopolitical landscape of AI is shifting, with initiatives like BRICS offering alternative approaches to intellectual property and AI development that may benefit Global South countries


Resolutions and action items

UN Secretary General’s report on innovative financing options for AI capacity building to be presented in September-October, including proposals for a global AI fund


Participants to receive access to an AI agent developed by Diplo Foundation based on available AI strategies for continued strategy development


Follow-up engagement planned with participants through email collection for ongoing collaboration on Landia case study


Continued development of the Landia AI capacity building strategy using inputs from the session


Sharing of AI apprenticeship program link and examples like the Botswana non-communicable diseases AI agent


Unresolved issues

How to balance sequential versus simultaneous approaches to addressing infrastructure challenges and AI development


Whether developing countries should align with BRICS AI approaches or Western/mainstream AI development models


How to ensure AI serves community-identified needs rather than externally imposed solutions without clear mechanisms for community consultation


How to prevent automation of existing inequalities when implementing AI solutions in underserved communities


Specific mechanisms for ensuring government support for community-based AI development initiatives


How to scale successful local AI solutions while maintaining their community-specific benefits


Detailed implementation pathways for the proposed global AI fund and coordination mechanisms


Suggested compromises

Addressing infrastructure challenges simultaneously with AI development through trade-offs rather than waiting for perfect conditions


Adopting a multi-basket approach to geopolitical AI alignment, maintaining relationships with both BRICS and Western AI initiatives


Focusing on augmented AI that enhances human dignity and local practices rather than replacing them entirely


Using simple, affordable AI tools (like $100 drones) that can provide immediate benefits while building toward more sophisticated systems


Developing AI solutions that work intermittently (when electricity is available) rather than requiring constant connectivity


Creating hybrid approaches that combine global AI resources with local data and knowledge systems


Building on existing community strengths (like radio networks) while introducing new AI capabilities


Thought provoking comments

The correct understanding of what we are dealing with, what it is, before policy, before capacity building, before anything else, that’s good action only flows from correct understanding.

Speaker

Amandeep Singh Gill


Reason

This comment establishes a foundational principle that challenges the typical rush to implementation. It emphasizes that understanding must precede action, which is particularly insightful in the AI context where there’s often pressure to adopt technology without fully comprehending its implications.


Impact

This set the philosophical tone for the entire discussion, establishing that the conversation would focus on deep understanding rather than superficial solutions. It influenced subsequent speakers to ground their comments in concrete realities rather than abstract concepts.


Three extra GPUs for Ethiopia, which has a total of 12 GPUs, is meaningful. But 3000 GPUs coming to South Africa, which is currently happening, is, you know, is another context. So we need to get more nuanced.

Speaker

Amandeep Singh Gill


Reason

This comment brilliantly illustrates the need for context-specific solutions in AI capacity building. It challenges the one-size-fits-all approach and demonstrates how the same resource can have vastly different impacts depending on the baseline context.


Impact

This comment shifted the discussion from generic capacity building to nuanced, tiered approaches. It influenced later speakers to consider local contexts more carefully and helped establish the framework for the Landia case study discussion.


Mainstream AI may not always deliver… the question about the right to tweak, transfer, and transform, which broadly is understood as the right to repair, but broader… part of humanity will just get extinct, right? I mean, those who can’t run the race.

Speaker

Anita Gurumurthy


Reason

This is a provocative challenge to the dominant AI narrative. It introduces the concept of AI sovereignty and questions the assumption that mainstream AI solutions are universally beneficial. The stark warning about human extinction for those who can’t compete is particularly thought-provoking.


Impact

This comment fundamentally shifted the discussion from ‘how to adopt AI’ to ‘what kind of AI should we adopt.’ It introduced alternative models like small, task-specific AI and local language models, leading to a more critical examination of AI development pathways.


It’s not about, it’s about how do we make technology work for us? It’s not about how technology makes you work. That’s a very subtle shift in the way we need to apply this.

Speaker

Ashutosh Chadha


Reason

This comment captures a fundamental philosophical shift in technology adoption. It challenges the common assumption that communities must adapt to technology, instead proposing that technology should adapt to existing workflows and needs.


Impact

This comment reinforced and crystallized the bottom-up approach theme. It influenced subsequent speakers to focus on preserving local practices while enhancing them with AI, rather than replacing them entirely.


Would it be prudent for us to think about building something that would require a stable internet connection… How do we develop a policy, an AI policy, while the basic things like network availability and infrastructure, that’s still a challenge.

Speaker

Tabaget Zavila (Botswana regulator)


Reason

This comment brings crucial practical realities into the discussion. It challenges the assumption that AI development can proceed without addressing fundamental infrastructure gaps, forcing the group to confront the sequential vs. simultaneous development dilemma.


Impact

This intervention grounded the discussion in practical constraints and sparked a debate about whether to address challenges sequentially or simultaneously. It led to Kurbalija’s example of the Botswana AI agent working despite intermittent electricity, showing how trade-offs can be managed.


AI should serve whether their needs or is it a choice of the people that are bringing AI to them who’s going to decide what are the needs? Because we might end up automating the inequality that is existing here.

Speaker

Baratang Miya


Reason

This comment raises the critical question of agency and power in AI deployment. It warns against the risk of perpetuating existing inequalities through AI, which is a sophisticated understanding of how technology can embed and amplify social problems.


Impact

This comment brought the discussion full circle to the core theme of bottom-up AI. It reinforced the importance of community agency in determining AI applications and served as a powerful conclusion to the capacity building discussion.


Overall assessment

These key comments collectively transformed what could have been a technical discussion about AI capacity building into a nuanced exploration of power, agency, and alternative development pathways. The discussion evolved from Gill’s foundational call for understanding, through Gurumurthy’s challenge to mainstream AI assumptions, to practical considerations about infrastructure and community needs. The comments created a progression from philosophical grounding to alternative models to practical constraints to community agency. This created a rich, multi-layered conversation that avoided both techno-optimism and techno-pessimism, instead focusing on contextual, community-driven approaches to AI development. The interplay between these comments established a framework for thinking about AI capacity building that prioritizes local needs, challenges dominant narratives, and acknowledges both opportunities and constraints in developing country contexts.


Follow-up questions

How can we avoid the same experience of fragmentation and insufficient funding that occurred with climate change and health initiatives when developing AI capacity building?

Speaker

Amandeep Singh Gill


Explanation

This addresses the critical need to learn from past failures in international development funding to ensure AI capacity building efforts are more coordinated and effective


How can we create effective protocols and incentives to link idle compute resources across different regions for AI capacity building?

Speaker

Amandeep Singh Gill


Explanation

This explores the technical and policy mechanisms needed to share computational resources globally, which could significantly reduce barriers to AI development in the Global South


How can AI be integrated into agriculture without displacing farmers from their livelihoods?

Speaker

Nandini Chami


Explanation

This addresses the critical balance between technological advancement and employment preservation in agriculture-dependent economies


What future of work strategies are needed when gradually moving populations from agriculture to higher value-added services?

Speaker

Nandini Chami


Explanation

This explores the long-term economic transition planning required when implementing AI in traditional sectors


How do we develop AI policies when basic infrastructure like reliable internet connectivity and stable power supply are still challenges?

Speaker

Tabaget Zavila (Botswana regulator)


Explanation

This addresses the fundamental question of whether to address infrastructure challenges sequentially or simultaneously with AI development


Should countries like Landia align more closely with BRICS AI vision or stick to Western/mainstream AI development approaches?

Speaker

Unnamed audience member


Explanation

This explores the geopolitical dimensions of AI development and the strategic choices countries must make regarding international partnerships


Who decides what AI needs a community has – the community itself or external actors bringing AI solutions?

Speaker

Baratang Miya


Explanation

This addresses the fundamental question of agency and self-determination in AI implementation to avoid automating existing inequalities


How can AI development in communities be sustained without strong government support?

Speaker

Rudy Massamba


Explanation

This explores the role of public sector involvement in supporting community-based AI initiatives and the challenges of grassroots AI development


What specific mechanisms are needed to implement the ‘right to tweak, transfer, and transform’ AI models for local contexts?

Speaker

Anita Gurumurthy


Explanation

This addresses the technical and legal frameworks needed to ensure communities can adapt AI technologies to their specific needs and contexts


How can public broadcast archives in local languages be effectively utilized to build indigenous large language models?

Speaker

Anita Gurumurthy (referencing the minister’s son in Landia)


Explanation

This explores practical approaches to leveraging existing cultural and linguistic resources for developing locally relevant AI systems


Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.