Responsible AI for Shared Prosperity

20 Feb 2026 16:00h - 17:00h

Session at a glance

Summary

This discussion focused on initiatives to make artificial intelligence more inclusive and accessible across Africa and Asia, particularly through the development of AI systems that work in local languages. UK Deputy Prime Minister David Lammy outlined three major investments: supporting AI development in over 40 African languages, establishing Africa’s first dedicated public sector AI computer cluster at the University of Cape Town, and launching the Asia AI for Development Observatory. These initiatives are part of the broader AI for Development programme launched at Bletchley Park, created in partnership with Canada, Germany, Japan, Sweden, and the Gates Foundation.

Philip Thigo, Kenya’s Special Technology Envoy, emphasized that this represents a civilizational moment, arguing that the Global South has never lacked intelligence but has lacked the power to define how that intelligence is recognized and transmitted. He stressed that without representation in AI systems, African cultures and languages face an existential threat of extinction. Chenai Chair from the Masakani African Languages Hub explained their community-driven approach to impact one billion Africans through AI tools in 50 of the most spoken languages, focusing on data collection, research, innovation, and sustainability.

Shekar Sivasubramanian from Wadwani AI described their work in India across 14-16 languages, emphasizing that utility and practical value are essential for adoption. German representative Barbel Kofler highlighted the importance of addressing bias in data and including diverse languages and dialects. The discussion also featured the announcement of Lingua Africa, a new multi-million pound initiative focused on creating open community-governed language infrastructure for real-world AI applications. The panelists concluded that these efforts are essential for ensuring AI serves as a force for good that uplifts all of humanity rather than dividing it.

Keypoints

Major Discussion Points:

– AI Language Inclusion Initiative: The UK government, in partnership with Canada, Germany, Japan, Sweden, and organizations like the Gates Foundation, is launching comprehensive programs to make AI accessible in over 40 African languages and support Asian language development, addressing the critical gap where current AI models predominantly serve English and other major languages.

– Cultural and Civilizational Preservation: Speakers emphasized that language representation in AI is not just about technology access but about preserving entire civilizations, cultures, and ways of thinking, particularly for the Global South which has historically been an oral civilization at risk of being excluded from the “age of intelligence.”

– Infrastructure and Computing Access: The discussion highlighted the creation of Africa’s first dedicated public sector AI computer cluster at the University of Cape Town and the significant barriers African researchers face in accessing computing power, with costs being exponentially higher than in developed countries.

– Community-Driven Development: The Masakani African Languages Hub was presented as a grassroots, community-led initiative that started in 2019 without initial funding, focusing on building AI tools for 50 of the most spoken African languages to impact 1 billion Africans across health, education, and economic sectors.

– Market Failure and Public Investment: Panelists discussed how private markets naturally invest in profitable languages (English, Mandarin) but fail to serve smaller resource languages, necessitating coordinated public and philanthropic investment to create these essential “public goods.”

Overall Purpose:

The discussion aimed to announce and explain new international collaborative initiatives designed to make AI inclusive and accessible across African and Asian languages, while addressing the digital divide and ensuring that AI development represents global linguistic and cultural diversity rather than just dominant languages and cultures.

Overall Tone:

The tone was consistently optimistic and collaborative throughout, with speakers expressing urgency about the civilizational importance of the work while celebrating partnerships and community-driven solutions. There was a sense of historical significance, with participants viewing this as a pivotal moment to prevent entire cultures from being excluded from AI development. The discussion maintained a diplomatic yet passionate quality, balancing technical details with broader humanitarian and cultural concerns.

Speakers

Speakers from the provided list:

– David Lammy – Deputy Prime Minister of the UK

– Philip Thigo – His Excellency Ambassador, Special Technology Envoy of the Government of Kenya

– Shekar Sivasubramanian – CEO of Wadwani AI

– Barbel Kofler – Parliamentary State Secretary to the Federal Minister for Economic Cooperation and Development of Germany

– Chenai Chair – Director of the Mazakani African Languages Hub

– Ankur Vora – Chief Strategy Officer and President of the Africa and India Office at the Gates Foundation

– Julie Delahanty – President of Canada’s International Development Research Centre (IDRC)

– Natasha Crampton – Chief Responsible AI Officer at Microsoft

– Co-Moderator – Role/title not specified

Additional speakers:

None identified – all speakers mentioned in the transcript are included in the provided speakers names list.

Full session report

This discussion brought together international government officials, technology leaders, and development organisations to announce major new initiatives designed to make artificial intelligence accessible across African and Asian languages. The conversation featured concrete commitments and partnerships aimed at addressing the global AI divide through community-led development and strategic international collaboration.

Major Initiative Announcements

UK Deputy Prime Minister David Lammy opened the discussion by framing “the choice the world faces” in AI development: either allowing AI to benefit only those who speak dominant languages, or ensuring it works for everyone. He announced three major UK investments addressing this challenge.

First, the UK is supporting AI systems development for more than 40 African languages through what Lammy described as a “genuinely African-led initiative” that enables people to access AI in languages they use daily. Second, the government is investing in Africa’s first dedicated public sector AI computer cluster at the University of Cape Town, designed to address prohibitive costs and limited access to computing resources that African researchers currently face. Third, the launch of the Asia AI for Development Observatory creates a new network supporting research, responsible AI governance, and ensuring AI reflects people’s lived realities across Asia.

The discussion concluded with the announcement of Lingua Africa, described as a “multi-million pound partner for an open call” developed through partnership between Microsoft AI for Good, the Gates Foundation, and AI for Development partners. This initiative focuses on community-governed language infrastructure for healthcare, education, agriculture, and public services. Additionally, Lammy announced support for four additional start-ups through a GSMA Foundation partnership.

These initiatives operate within the broader AI for Development programme, launched when the UK hosted an AI summit at Bletchley Park. The programme brings together Canada’s International Development Research Centre, the Gates Foundation, and the governments of Germany, Japan, and Sweden in coordinated multilateral effort.

Cultural Preservation and Civilizational Stakes

Philip Digo, Kenya’s Special Technology Envoy, provided crucial reframing by characterizing the current moment as fundamentally civilizational. His argument that “we’re actually in the age of intelligence” moved the conversation beyond technical considerations to questions about cultural survival. Digo emphasized that the Global South has never lacked intelligence but has historically lacked power to define how that intelligence is recognized, recorded, or transmitted.

Digo’s observation that “our entire culture’s values have been coined in language” and that “the Global South is largely an oral civilisation” highlighted African cultures’ particular vulnerability in an AI-dominated future. Without representation in AI systems, he argued, these civilizations face “almost existential” risk. This positioned the language AI initiative not as development aid, but as correction of historical power imbalances.

When young people in Kenya—among the highest ChatGPT users globally—seek guidance from AI models, Digo argued, responses should reflect their cultures rather than imposing external worldviews. This represents a fundamental shift from viewing AI as neutral tool to understanding it as a system embedding particular cultural assumptions.

Community-Driven Development: The Masakani Model

Chenai Chair from the Masakani African Languages Hub detailed how community-driven AI development works in practice. The Masakani initiative, beginning in 2019 as grassroots effort without initial funding, exemplifies African communities taking ownership of their AI representation. The name “Masakani,” meaning “to build together,” reflects the collaborative ethos driving the initiative.

The hub’s goal of impacting one billion Africans through AI tools in 50 of the most spoken languages demonstrates the challenge’s scale. Chair explained their approach focuses on four pillars: expanding and diversifying high-quality data, developing inclusive AI machine learning models, creating practical applications, and ensuring sustainability.

Chair emphasized linguistic complexity within individual languages, noting “the Shona I speak in Harare is not the same as the Shona spoken in Mutai.” This illustrates the nuanced understanding required for effective AI language development, going beyond simple translation to encompass dialectical variations, cultural contexts, and regional specificities.

The announcement of Project Echo—”enhancing communications for her opportunities”—represents innovative approach to intersectional inequalities. This gender-responsive intervention acknowledges high gender inequality on the continent and specifically designs solutions supporting women’s economic empowerment and health outcomes.

Technical Infrastructure and Computing Barriers

Julie Delahanty from Canada’s International Development Research Centre presented evidence of infrastructure challenges facing African researchers. Her organization’s study demonstrated that computing capacity costs are exponentially higher in African countries compared to Germany or the UK, creating fundamental barriers to AI research participation.

The African Compute Initiative, establishing the first dedicated high-performance computing cluster for public institutions in Africa at the University of Cape Town, addresses these infrastructure gaps. The initiative includes modern GPUs, faster storage capacity, and improved networking—essential components for training large AI models and supporting initiatives like the Masakani hub.

Natasha Crampton from Microsoft provided technical context for how computing power enables multilingual AI development. Creating language and culturally aware AI requires substantial computational resources for collecting and processing locally-led data, training models to incorporate new linguistic information, testing systems with local speakers, and supporting daily technology use. Microsoft’s role as computing enabler rather than content controller reflects recognition that trustworthy AI must be built through partnerships respecting local ownership of cultural and linguistic resources.

Market Failures and Public Intervention

Ankur Vora from the Gates Foundation provided economic analysis justifying coordinated public and philanthropic intervention in multilingual AI development. His assessment that “the markets are broken” offered rationale for why private sector investment alone cannot address smaller language communities’ needs. Private companies naturally invest in English and Mandarin models because those languages offer clear economic returns, but this logic leaves thousands of other languages underserved.

Vora’s argument that “when markets are broken, funders can get together and invest in public goods” provided the economic framework for understanding why the Gates Foundation, multiple governments, and international development agencies needed to collaborate. This moved discussion beyond moral arguments about inclusion to practical economic analysis about addressing market failures that systematically exclude certain communities from technological benefits.

Practical Implementation and User Value

Shekar Sivasubramanian from Wadhwani AI emphasized that successful AI adoption requires immediate, tangible user value. His organization’s work across 14-16 languages in India demonstrates how multilingual AI can be implemented at scale when providing genuine utility. Their applications in health surveillance, education assessment, and agricultural support show how language-aware AI can address real-world problems in immediately understandable ways.

Sivasubramanian’s principle that “it is very important in human contexts to provide some value to the person in any interchange” challenges the field to move beyond theoretical language preservation to practical applications providing immediate benefits. His examples—AI systems helping teachers assess student reading fluency in local languages or providing disease surveillance across multiple languages—demonstrate how technical capabilities translate into meaningful life improvements.

International Collaboration and Partnership

The discussion highlighted AI language initiatives’ diplomatic significance in strengthening international relationships and demonstrating commitment to global equity. Lammy’s emphasis on partnership with Canada, Germany, Japan, and Sweden reflects recognition that addressing AI inequality requires sustained international cooperation.

Babel Kofler’s participation as Germany’s parliamentary state secretary demonstrated European commitment to addressing AI data bias and including diverse languages and dialects. Her observation that she doesn’t speak “standard German” but uses a dialect different from Hamburg personalized the language diversity challenge, showing how even developed countries grapple with linguistic inclusion in AI systems.

Implementation Challenges and Future Directions

Several sustainability challenges emerged from the discussion. Questions about maintaining initiatives beyond current funding cycles remain unresolved. While Masakani demonstrated that grassroots innovation can begin without formal funding, scaling to serve billions across thousands of languages requires sustained resource commitment extending beyond typical project timelines.

The balance between open-source development and community sovereignty presents ongoing challenges. While open-source approaches can accelerate innovation and reduce costs, communities need assurance that their linguistic and cultural data won’t be exploited without their control or benefit.

Conclusion

This discussion represented a comprehensive examination of how AI development can address rather than perpetuate global inequalities. The conversation elevated multilingual AI from technical challenge to framework for thinking about power, representation, and equity in artificial intelligence development.

The consensus among speakers—from government officials to community leaders to private sector representatives—demonstrated mature understanding of both challenges and solutions. Alignment on the need for public intervention to address market failures, importance of community-led development, and requirement for practical utility in AI applications provides strong implementation foundation.

The initiatives described represent crucial intervention at a pivotal moment in AI development. The concrete commitments announced—from the University of Cape Town computing cluster to Lingua Africa’s multi-million pound investment—offer tangible steps toward ensuring AI serves linguistic and cultural diversity rather than concentrating benefits among speakers of dominant languages.

Success will be measured by real-world impact: reducing maternal mortality, supporting education, enabling economic empowerment for billions whose languages have been historically excluded from technological development. The collaborative model demonstrated in this discussion, sustained investment in technical infrastructure and community capacity, and continued attention to intersectional challenges provide the foundation for achieving these ambitious goals.

Session transcript

David Lammy

to make AI work in more than 40 African languages. This is a brilliant, genuinely African -led initiative which helps people to access AI in the languages that they actually use in their everyday lives. Second, we’re investing in Africa’s first dedicated public sector AI computer cluster at the University of Cape Town. Too many African researchers are held back by costs and a lack of access. And the hope is that this new hub will give them the computing power to build and train models locally. And third, we’re launching the Asia AI for Development Observatory. This is a new network to support research, responsible AI governance, to protect rights and to ensure AI reflects the realities of people’s lives across the region.

All of these initiatives are effectively part of our AI for Development programme, launched when we hosted the first of these AI summits back at Bletchley Park three years ago and made in partnership with Canada’s International Development Research Centre and as part of a wider collaboration to coordinate investments with the Gates Foundation, the governments of Germany, Japan and Sweden, as well as Community Jamil. And as part of our partnership with the GSMA Foundation, we’re proud to announce the support to four additional start -ups and these innovative businesses will harness responsible AI and will be able to support the development of AI for the future. to support the needs of underserved people. across Asia and Africa. They include Torn AI in Morocco, which creates voice interfaces to local dialects to help low -literacy rural users access digital and financial services through simple spoken interactions.

And all of these initiatives will make a real difference to people across the continent of Africa and Asia. But I hope they’ll do a bit more than that. Yesterday, I spoke about the choice the world faces, the two paths before us, one which sees AI take power and opportunity away from people and sadly divides us, and one that sees AI used as a force for good to solve problems and uplift all of humanity. and the projects I’ve mentioned, the ones we’re going to hear about today and the many new institutions and coalitions that are now emerging can help make sure we go down the right path and that is a path of a safe AI, an inclusive AI and importantly an equitable AI for everyone.

So let’s turn now to our panel, an exceptional group of leaders from across India and Africa and I’m going to get an introduction to our panel members and then I’ll start with the first question.

Co-Moderator:

Thank you. It’s my pleasure to introduce, joining our Deputy Prime Minister on the stage, we have His Excellency Ambassador Philip Digo, Special Technology Envoy of the Government of Kenya. We have Dr. Babel Kofler. parliamentary state secretary to the federal minister for economic cooperation and development of Germany. We have Shekhar Sivasubramanian, CEO of Wadwani AI and Chennai Chair, Director of the Mazakani African Languages Hub. And so to the first question to Philip. So we’re beginning to see an attempt well we all experience how large language models are affecting our lives on a daily basis. I certainly use a, I don’t use chat GPT but I use a secure network which isn’t taking my stuff because obviously for obvious reasons as Deputy Prime Minister of the UK I have to be a bit careful but I’m using it to research and you know usually really get quickly to a that I don’t fully understand.

So as we move this on to local languages, dialects, and across Africa we’ve got an estimated 2 ,000 languages, how do you see AI in local languages shaping the next phase of your country’s digital development having been to Kenya many times and knowing the many groups that are there, how does this really work on the ground?

Philip Thigo

Thank you so much, Right Honourable Prime Minister. I think this moment is so profound that I don’t think you guys are realising what is happening here. I think the first thing is to understand that we’re actually in the age of intelligence, right? So it’s not about ICTs or technology, it’s about our AI shape. I think how we live, learn, work, collaborate. and engage. From our point of view, it’s a civilizational discussion. The Global South has never lacked intelligence, as we know, right? So what it has lacked is the power to define how that intelligence is recognized, recorded, or transmitted. Because our entire culture’s values have been coined in language. The Global South is largely an oral civilization.

And so the current models lacking our language means our civilization are at risk, almost existential, to be extinct. And I think this initiative, in our view, begins to ensure that we are represented in the current age of intelligence, but also our intelligence is part of our global collective history and memory. And so I think which means then how we engage in this is that now with the capabilities like Masakani, it means that when I get into Chagipiti that you refused to mention, is that… Young people in Kenya, who are the number one users of Chagipiti? by the way, are not only seeking emotional advice or guidance from these models, but then when they engage with these models, it actually also represents their cultures and civilization.

I think for me that’s how it works practically on the ground. First of all, it’s representation and existence. The second part, of course, is it also works when we have the entire stack. And you mentioned a couple of things around finding the models, but I think it would be interesting to see how we find the compute, the talent that then influences, develops the data, develops the language. The research and development capability, which I was in the first instance, and that was an amazing initiative because then we need to build research capacity and capability because talent development is the first instance of sovereignty. Then the final point, of course, is the specific use cases and languages, especially in the African context.

Again, you say 2000. So each of the 2000 are very context -specific. Kiswahili and Yoruba are not the same, and neither are their applications in the context of the African context. sense in even our history and cultures in Africa. So I think that capability, as diverse as it could be, also ensures that our diversity in the African continent is represented in the future models.

David Lammy

And that’s wonderful. And the first point you made is really about sort of seeing yourself in this story that the global community is going to be telling in relation to intelligence. And we know that in the past, Africa has been written out of that story. So it’s hugely important that African languages intellect history over thousands of years is in this storybook. So should I tell us then how the Masa Kani African languages is addressed? And how are we addressing these issues and really working as a tangible, real thing?

Chenai Chair

Thank you so much. Minister. So I think I want to say that I am proud to be representing the Masakane community that started in 2019, wanting to see their own languages represented in the global domain, and they did it by the bootstraps. No one wanted to fund them, and they came together and said, hey, how do I ensure that the language that I speak is captured digitally? So the Masakane African Language Hub emerges from that community -driven initiative, where our main goal is to impact 1 billion Africans through 50 of the most spoken languages with relevant AI tools that will allow for economic growth, health, and social benefit, and also working towards the preservation and also capturing the evolution of African languages.

The 2 ,000 -plus are growing, and so then even as Honorable Philip Degas mentioned, the diversity of the language is growing. So I think it’s a very important thing to do. Thank you. Thank you. Like I speak Shona, but the Shona I speak in Harare is not the same as the Shona spoken in Mutai. So it’s really capturing that diversity and nuance of that work. So what we do with support from the funding collaborative and the partners that we have is actually think about enabling the ecosystem through partnerships and grant making. So we specifically focus on four pillars of work, which is around data. So expanding and diversifying high quality data. In 2019, there wasn’t as much data, but the Masakana community actually started building up that data based from the JW300 Bible data set that had been created.

Secondly, we’re also looking at research. So it’s important for us to take it on as an ecosystem invention, where we are looking at developing and refining inclusive AI machine learning models, but also thinking about the tooling that’s resourcing these. So what we are working on specifically, that is having a benchmark project where we’re actually… going to create a relevant African benchmark looking at speech and text because the current benchmark models out there do not nuance the realities on the African context. And then also looking at innovation. So again, a lot of the questions that we’ve seen is when you create the data, where does it go? Is it taken up in the market? What’s the impact?

And so for us, we’re actually working 40 % of the funding that we have will go to creating use cases and impacting use cases. And one special mention I want to put forward is actually that we are working on a project called Project Echo, which means enhancing communications for her opportunities. This is a gender responsive intervention that exists in the context of high gendered inequality on the continent. And what that does is it will provide relevant use cases in African languages that lead to impact on women’s economic empowerment and health. And that’s a significant part of us recognizing the context we exist in. And then lastly, we really are thinking about sustainability. and right now we’re in a moment where there is resourcing, where there is funding and we also come from a moment where people were doing it without funding.

So we’re thinking about institutional capacity building for the African NLP community which will actually then see businesses coming up from these open source models, people innovating off the data that’s created and sustainability beyond the Masakana community which has been happening right now but then this funding allows us to actually have African -led AI which is built for impact. Thank you very much.

David Lammy

Thank you very much. Centres Africa but also importantly the fundamental inequality and gender issues that sit at the heart not just of Africa but making sure that women are, a big part of this story. Shekhar, bringing India into this and thinking of Wadwani AI and we’re sitting here in the most populous country on the planet. There are also lots of languages and tremendous diversity but also innovation and range across this country. So tell us how Wadwani AI is working at the heart of that innovation here in India. Thank you. First, the

Shekar Sivasubramanian

work we do is applied AI, which means we solve for problems in health, education and agriculture. And we’ve been doing it for the last seven years. The moment you work in India, the very first design principle that you start with is the ability to be… inclusive and embrace the entire population. So the dimensions of population we are looking at are language. So you start with at least 14 to 16 languages. You don’t even think of an application otherwise. Second, you also think of complete inclusivity, which means you need to think through the divide between rural and urban, the kinds of applications that will be delivered to people which will be of use to them. Third, are applications fundamentally must be useful to people.

Then they open out their ability to learn languages, changes that better interface with technology that can actually be of use to them. So that utility value sits at the heart of everything that we do, which drives a lot of behavior both by us as well as the ecosystem. Just as an example, we do media disease surveillance. We’ve been doing it for a while, which picks up every article published in India, it runs four hours, and it picks. Events of interest in health. 16 languages and it’s been running for the last two and a half to three years. It uses AI and it tells you in this region this many people got this disease at this time.

It runs every four hours and it tells the central government if it’s a disease outbreak what should you do. Another completely different example. We collect data from children in a couple of states and that will expand to 14 to 16 states where we have the largest data set of spoken local language. And which again in of itself of no use but when you provide something called oral reading fluency which assists the poorest child to read a paragraph and the AI tells you what you read well, what you did not read well and assists the teacher to cohortize the students and provide them information. Suddenly the language and the application you cannot distinguish between them. It is very important in human contexts to provide some value to the person in any interchange.

If you can work the value then the adoption is possible. option is easy. If you divorce the two, people don’t understand why I’m doing what I’m doing. It looks like an encumbrance. So for us, at the heart of our innovation is what does it mean for the person. Independent of which, we do analysis on various languages. We’ve done one on Tibetan, where we’ve preserved their entire culture by taking, we worked in Dharmashala, as well as in Karnataka. We digitized their entire library system and allowed the communities there to gain employment using it. Likewise, we plan to work on multiple less -used languages, Pan -India. We believe, it’s our position that, and we’ve got Agrivani, Healthvani, everything that we do is multilingual.

Everything that we do collects data. We have the largest data sets now, incidentally, of the work we do. It’s not what we do. It comes as a by -product of the work that we do. Over a period of time, it is my heart. Considered an humble opinion that these models using AI will take time. We should be ready to write this for a period of time. We should be ready to invest in deep research and or very utilitarian based approaches so that you can take the community along with you. That is super important. The theory is interesting, the practice is different. There is a theory as to how to design roads in India. I will keep quiet after a bit.

Very David Lammy

very good example. Obviously I talked about the UK as a donor country doing this in partnership with others, Canada, Sweden, but also the German government and we’re joined by Babel Koffler just to bring the donor a perspective really to this and why this is so important. Thank

Barbel Kofler

you very much, Deputy Prime Minister. Thank you. Oh, there’s no one? What did I do? Oh, thank you. Thank you very much. I should put it on first, that’s true, yeah. Thank you very much, Deputy Prime Minister. I wouldn’t talk if it’s coming to AI in a manner of donor and recipient, because at the end of the day I think it’s a new technology where we all have to bridge if we really want to make it useful for everybody. And that’s also our interest, of course, from the German side. We see that AI can only be really the game changer to overpower… …to overcome inequality, to fulfill the promises of the SDG. And that’s for every country important, not only for global laws.

global South that’s important for everybody, can only be that game changer if it is inclusive. That starts with data at the end of the day and how biased data is. And if you talk about bias in data, language is quite close to it. You were pointing out how important it is and how you differently speak in various variations of your language. I really understand that I don’t speak standard German normally, so I also use a dialect, and that’s quite different from Hamburg. So we all have something to include also, which is connected with a cultural momentum, and we see so many languages neglected, totally neglected, dialects, cultures, because it’s not only the language, it’s what the language is transporting also, which is neglected.

And that’s why we really try to be part. And we are very proud to be part of your initiative. also. We were starting in 2019 with discussing those topics, working on an initiative called Fair Forward. That’s part of the initiative. And working also with partner countries like India on collecting data sets, so really to collect the necessary data on those local languages, which at the end of the day is offering then or should offer service to citizens in their mother tongue in multilingual countries or contexts, for example. So for us, it’s of utmost importance. Happy to be part of the initiative. We want to stay a reliable partner on that, and we will be part of that initiative.

And I hope the idea is spreading and growing. Thank you. Thank

Co-Moderator:

you. We’ll now have a small… Changeover in our panellists. If I could ask, if we could have another big round of applause, please, for His Excellency Philip Higo, Debra Kofler and Shekhar Sivasubramanian. And now joining us on stage, we will have Ankara Vora, Chief Strategy Officer and President of the Africa and India Office at the Gates Foundation. Julie Delahunty, President of Canada’s International Development Research Centre. And Natasha Crampton, Chief Responsible AI Officer at Microsoft. Thank you very much.

David Lammy

Back to Chennai, my understanding is that Masakami are announcing a new multi -million pound partner for an open call today for Lingua Africa. So can you tell us a little bit more about this initiative and the gap that it’s designed to close effectively? And then why this moment is so important for African languages as a whole and AI?

Chenai Chair

Thank you so much Deputy Prime Minister. So yes, I do have the honour with my esteemed panellists to actually announce Lingua Africa. So with Masakami, which I’ve said means to build together, we’ve been working with researchers and communities across the continent to close the gap in how African languages are being used. And how African languages are represented in the AI systems. What we’ve constantly seen is that… I think I did mention this, is that it’s not just about data. It’s about whether the language resources actually translate into tools people can use, particularly in healthcare, education, agriculture, and public services, because those are the developmental domains that we’re likely to have significant impact with. So together with Microsoft AI for Good and the Gates Foundation, as well as our AI for D partners, LINGUA Africa will be a multi -partner open core focused on open community -governed language infrastructure, which will directly enable real -world AI applications.

I think a lot of the times as we’re developing AI solutions, the question becomes, if we’re building them in a lab, will they work in the real world? And that’s also consistently part of what we’re doing with the benchmarking work. So how we’ll do this is actually then it’ll be a use case or impact -focused specific approach. Where we will do model development, we will collect targeted data in those… specific domains and then also support strong pathways for deployment and adoption. So this is us working with multiple entities, the academic community, our partners here on stage with us, but also the tech entrepreneurs who are actually building up these solutions. And then for us, it’s quite simple.

The goal is to make sure that language is not a barrier anymore into including people into these solutions, particularly if you think about digital public infrastructure interventions. They need to be in languages that people communicate with because you will leave behind a majority of people if they are in languages that they do not understand. So that is our most significant contribution right now. Thank you.

David Lammy

Thank you. And obviously, we’re very pleased in the UK to be partnering with the Gates Foundation on new support for linguistic diversity across AI. But just explain, Ankur, the role. that the hub has effectively in that wider impact on the global south.

Ankur Vora

This is on? It’s on, all right. Languages matter. Can you first join me in giving a big round of applause to Chennai for this amazing movement. It is kind of brilliant where we are in this moment in time. Let me talk about three whys. One is why care about language? The second is why care about investing in language? And the third one is why care about investing in initiatives like Masakane? The first one, I think so everybody knows, but it’s useful to repeat it. And many people have talked about this before. Because we want to make… We want to make sure that the power of AI… actually changes lives. History is not going to remember us for the models we developed or the speeches we give here.

History is going to remember the impact we all had. We’re talking about mothers and babies not dying. We’re talking about the next generation growing up in a world without infectious diseases. We’re talking about hundreds of millions of people escaping the clutches of poverty. Those kind of things matter. And the solutions are there. They can get better. But we need to find a way of these solutions getting translated for these use cases. So that’s why we need to care about this thing. Why invest in language? Because the markets are broken. The markets are broken. Private sector companies are investing in models developed in English and Mandarin. And it makes sense for the markets to do that.

Because that’s where the economics work. But just because… the economics don’t work for the small resource languages doesn’t mean that we shouldn’t be investing in this. And that’s the point why all of us need to get together and say we need to do something about it. When markets are broken, funders can get together and invest in public goods. And that’s what we all are doing right now. UK government, Canadian government, the Japanese, Microsoft, IDRC, everybody is getting together and making the point that we need to invest in public goods because these markets are broken and this is an important thing. And so I’m quite excited about the fact that we’re all sitting at this panel, people in this audience, and saying we’re going to make sure that as we think about tomorrow, the important thing of developing solutions in the right languages that can solve the problem.

Thank you. be a problem that we will tackle. Thank you.

David Lammy

Thank you. I’ve got to say as a politician I like the idea that the state intervenes and doesn’t just leave it to the marketplace to determine where to put the funds, I’ve got to say. But obviously we are, and you have to in terms of the innovation, work at the cutting edge and that cutting edge is most often in the private sector as well, taking those risks to develop and innovate and here Microsoft and cloud computing is hugely important, Natasha, and it’s important to understand that interaction between the cloud. I think that the Masakani hub as it has been described so well, by Shaniai and how important it is that computation and cloud technology helps the innovation of these local languages.

Could you say a little bit more about that? But there is another subset to this, which is getting the balance right so that the languages and often the communities that we’re supporting that are at the front line have equity in this and don’t lose their own sovereign capabilities, which has been a theme of this conference. So I wonder if you could just reflect on that as well.

Natasha Crampton

is an urgent priority. Our own analysis shows that we have AI diffusing in the global north at roughly double the rate that we have it diffusing in the global south at the current time, and that is exactly why we need partnerships like these in order to start to put the right infrastructure in place to close that gap. Now, language is particularly important in terms of overcoming that AI divide. As we’ve heard many speakers say today, nobody is going to use AI if it does not speak the language that you speak, and importantly, that it does not work in the context, in the specific scenario in which you need to use it. So language -aware and scenario -aware, AI.

AI is incredibly important to empowering people to put… the technology to work in the use cases that mean the most to them. And that’s why we’re so thrilled to be partnering with Masakane, as well as the Gates Foundation, and the UK government on this Lingua Africa initiative. So how does compute come into all of this? I think, quite simply, compute is the enabler of making language and culturally aware AI. It’s a critical component of it. So when we take a base model that may have just been trained, like most models, on data sets that are predominantly English, we need to make sure that we can do this responsibly, locally -led data collection that Shania was talking about earlier.

And then we need to do some further work on the models to essentially ingest that data and make it well -registered. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. And that’s what we’re trying to do. takes compute. Then it’s very important once we’ve actually made the model language, linguistically and culturally aware, we need to make sure that we’re testing it with local language speakers and in the right scenarios.

That also, that testing, that also takes compute. And then finally, the day -to -day use of this technology, it also requires computing power. So we’re really here today as an enabler of an Africa -led effort by Africans for Africans to create this linguistically aware and multi -culturally aware technology, and compute fundamentally is just the enabler of it. I think my last thought to offer here today is I think these types of initiatives just really reinforce that trustworthy AI is not going to be the best tool for computing. I think it’s going to be the best tool for computing. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. because of the choices that we make, the ways in which we choose to build and test and deploy these AI systems and for us at Microsoft it’s really important that we do take all of those steps to represent the world as it is multicultural, multilingual and deeply interconnected so we’re thrilled to be part of this initiative.

Thank you very much.

David Lammy

And Julie, so we very much described this journey that we’ve been on since Bletchley Park and talked about languages, talked about computing. In a sense, AI, that’s the foundation stone for communities but just to round off this event, how do we see the opportunities going forward? particularly, where do we need to get to?

Julie Delahanty

Thank you. Thanks, everybody, for being here and for the Deputy Prime Minister for welcoming us. We’re incredibly proud at IDRC to be part of the AI4D initiative with the UK government and to be partnering with the Masa Kani African Language Hub as well as the African Compute Initiative. And I think going back a little bit to the Microsoft views on it, I think it’s very similar for us. Researchers in lower and middle -income countries really have to have strong computing power to be able to do the kind of cutting -edge AI work that Shana and others are doing. But right now, of course, they do face a lot of barriers. We did a study that showed the incredible increased cost of getting compute capacity, the difference between getting it in Germany and getting it in the UK.

But in an African country, the costs are exponentially larger. So that computing cost and how much it is, the local infrastructure that might be limited, the GPUs that are the… hardware that’s really driving the powers modern AI is also very difficult and hard to access for African countries. So it really makes it difficult for them to fully participate in global AI innovations. The African Compute Initiative is going to change all that, we hope. It is going to be the first dedicated high -performance computing cluster for public institutions in Africa. It will be based in South Africa at the University of Cape Town. And that initiative is going to include modern GPUs. It’s going to have faster and better storage capacity and much faster networking.

And it’s that kind of computing power that is essential. It’s essential for training large AI models. It’s essential for testing new ideas more quickly, as you mentioned. Subtitles by the Amara .org community And it’s been essential and will be essential for things like the Masakani African Language Hub. Both the initiatives that I’m talking about are really responding to the foundational gaps. So whether that’s compute capacity or the kinds of representative and robust data sets, both of those things are absolutely necessary. And if you don’t have those foundations, then you can’t contribute to AI systems. And if you can’t contribute to AI systems, then you can’t shape the AI systems. So it’s absolutely critical for Africa’s AI innovations to have those foundational elements that exist.

And I think the lessons that we’re going to learn through a lot of this programming is going to help other regions and other lower resource contexts to do that kind of work. And in terms of the next steps or the things that we can do with that, I mean, you can imagine some of the obvious things. I mean, some people have already mentioned it, but things like having…

Co-Moderator:

Thank you very much. Thank you very much. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you.

David Lammy

Speech speed

110 words per minute

Speech length

Key initiatives and partnerships driving impact

Topics

Artificial intelligence | Information and communication technologies for development

Chenai Chair

Speech speed

166 words per minute

Speech length

Inclusive, language‑focused AI for representation and equity

Topics

Closing all digital divides | Artificial intelligence

Shekar Sivasubramanian

Speech speed

166 words per minute

Speech length

637 words

Speech time

229 seconds

Applied AI for health, education and agriculture

Explanation

Sivasubramanian explains that his work focuses on applying AI to solve concrete problems in health, education and agriculture sectors across Africa.

Evidence

“work we do is applied AI, which means we solve for problems in health, education and agriculture” [24].

Major discussion point

Key initiatives and partnerships driving impact

Topics

Social and economic development | Artificial intelligence

Barbel Kofler

Speech speed

144 words per minute

Speech length

391 words

Speech time

162 seconds

AI can overcome inequality and fulfill SDGs

Explanation

Kofler argues that AI has the potential to be a game‑changer for overcoming inequality and achieving the Sustainable Development Goals.

Evidence

“We see that AI can only be really the game changer to overpower… …to overcome inequality, to fulfill the promises of the SDG” [27].

Major discussion point

Inclusive, language‑focused AI for representation and equity

Topics

Artificial intelligence | Social and economic development

Ankur Vora

Speech speed

150 words per minute

Speech length

415 words

Speech time

165 seconds

Languages matter

Explanation

Vora stresses the fundamental importance of language in AI initiatives, underscoring that linguistic diversity must be central to development efforts.

Evidence

“Languages matter” [15].

Major discussion point

Inclusive, language‑focused AI for representation and equity

Topics

Closing all digital divides | Artificial intelligence

Markets are broken – need public‑good investment

Explanation

Vora points out that market failures require coordinated public‑good investment from governments, donors and industry to support language‑focused AI.

Evidence

“UK government, Canadian government, the Japanese, Microsoft, IDRC, everybody is getting together and making the point that we need to invest in public goods because these markets are broken and this is an important thing” [17].

Major discussion point

Collaboration between public sector, donors, and private industry

Topics

Financial mechanisms | The enabling environment for digital development

Natasha Crampton

Speech speed

153 words per minute

Speech length

Key initiatives and partnerships driving impact

Topics

Social and economic development | Artificial intelligence

Julie Delahanty

Speech speed

154 words per minute

Speech length

471 words

Speech time

183 seconds

AI4D initiative and African Compute Initiative partnership

Explanation

Delahanty notes IDRC’s participation in the AI for Development (AI4D) programme alongside the UK government and the African Compute Initiative, highlighting collaborative effort to build compute capacity.

Evidence

“We’re incredibly proud at IDRC to be part of the AI4D initiative with the UK government and to be partnering with the Masa Kani African Language Hub as well as the African Compute Initiative” [11].

Major discussion point

Infrastructure and compute as critical enablers

Topics

Artificial intelligence | Financial mechanisms

Subtitles essential for language access

Explanation

She emphasizes the importance of community‑generated subtitles (via Amara.org) as a critical component for making content accessible in local languages.

Evidence

“Subtitles by the Amara .org community And it’s been essential and will be essential for things like the Masakani African Language Hub” [12].

Major discussion point

Building capacity: data, research, talent, and benchmarks

Topics

Closing all digital divides | Data governance

Philip Thigo

Speech speed

173 words per minute

Speech length

Human rights and the ethical dimensions of the information society

Topics

Human rights and the ethical dimensions of the information society | Artificial intelligence

Co-Moderator

Speech speed

41 words per minute

Speech length

Session facilitation and procedural continuity

Topics

Internet governance | The development of the WSIS framework

Agreements

Agreement points

Language representation in AI is critical for inclusion and preventing cultural extinction

Speakers

– Philip Thigo
– Chenai Chair
– Barbel Kofler
– Natasha Crampton

Arguments

Arguments

Applied AI must embrace complete inclusivity across languages, rural-urban divides, and provide fundamental utility to users for adoption

Success should be measured by actual impact on lives – reducing maternal mortality, eliminating diseases, escaping poverty – not just technical achievements

Lingua Africa initiative announced as multi-partner open call focused on community-governed language infrastructure for healthcare, education, agriculture and public services

Summary

All speakers emphasize that AI initiatives must deliver tangible benefits to users in practical domains like health, education, and agriculture rather than being purely technical exercises

Topics

Social and economic development | Artificial intelligence | Information and communication technologies for development

Financial mechanisms | Artificial intelligence | Information and communication technologies for development

Unexpected consensus

Private sector limitations in serving linguistic diversity

Speakers

– Ankur Vora
– Natasha Crampton

Arguments

Private sector markets are broken as they only invest in English and Mandarin models where economics work, requiring public sector intervention

Trustworthy AI requires representing the world as multicultural, multilingual and deeply interconnected

Explanation

It’s notable that both a Gates Foundation representative and a Microsoft executive openly acknowledge that private markets alone cannot solve the multilingual AI challenge, with the Microsoft representative explicitly supporting public goods approaches despite representing a major tech company

Topics

Artificial intelligence | Financial mechanisms | The enabling environment for digital development

Community-led development as the preferred approach

Speakers

– Philip Thigo
– Chenai Chair
– Natasha Crampton

Arguments

Global South lacks power to define how intelligence is recognized and transmitted, with oral civilizations at existential risk without language representation in AI

Masakani community started grassroots initiative in 2019 to represent African languages in AI, focusing on data, research, innovation and sustainability

Trustworthy AI requires representing the world as multicultural, multilingual and deeply interconnected

Explanation

There’s surprising alignment between African community leaders and a major tech company representative on the importance of community-led, locally-controlled AI development, suggesting a shift away from top-down technology deployment models

Topics

Artificial intelligence | Human rights and the ethical dimensions of the information society | Capacity development

Overall assessment

Summary

The speakers demonstrate remarkable consensus across multiple dimensions: the critical importance of linguistic diversity in AI, the need for adequate computing infrastructure, the requirement for public sector intervention to address market failures, and the imperative for AI to deliver practical benefits. There is also strong agreement on community-led approaches and the existential nature of language representation in AI systems.

Consensus level

Very high level of consensus with no significant disagreements identified. This strong alignment suggests a mature understanding of the challenges and a coordinated approach to solutions. The implications are positive for implementation, as all stakeholders appear aligned on both problems and solutions, potentially leading to more effective collaborative initiatives and resource allocation.

Differences

Different viewpoints

Unexpected differences

Overall assessment

Summary

The discussion showed remarkable consensus among speakers on the fundamental challenges and goals of multilingual AI development, with no direct disagreements identified. The main areas of variation were in emphasis and approach rather than conflicting viewpoints.

Disagreement level

Very low disagreement level. All speakers aligned on core issues: the need for multilingual AI, the importance of addressing market failures, the requirement for computing infrastructure, and the goal of practical impact. The variations in emphasis (technical vs. funding vs. community-driven approaches) actually complement each other and suggest a comprehensive multi-faceted strategy rather than conflicting approaches. This high level of consensus likely reflects the collaborative nature of the initiatives being discussed and the shared commitment to addressing AI divides.

Partial agreements

All speakers agree that market failures require intervention to support multilingual AI development, but they emphasize different solutions: Vora focuses on collaborative public funding as the primary solution, Crampton emphasizes the technical infrastructure (compute) as the key enabler, while Delahanty highlights the need for dedicated computing clusters to address cost barriers

Speakers

– Ankur Vora
– Natasha Crampton
– Julie Delahanty

Arguments

Private sector markets are broken as they only invest in English and Mandarin models where economics work, requiring public sector intervention

Compute is the critical enabler for making language and culturally aware AI through data collection, model training, testing and deployment

African researchers face exponentially higher costs and limited access to computing power compared to developed countries

Topics

Artificial intelligence | Financial mechanisms | The enabling environment for digital development

Both speakers agree on the critical importance of African language representation in AI systems, but they approach it from different angles: Thigo frames it as an existential civilizational issue requiring sovereignty over intelligence definition, while Chair focuses on practical community-driven solutions through systematic data collection, research, and sustainable implementation

Speakers

– Philip Thigo
– Chenai Chair

Arguments

Global South lacks power to define how intelligence is recognized and transmitted, with oral civilizations at existential risk without language representation in AI

Masakani community started grassroots initiative in 2019 to represent African languages in AI, focusing on data, research, innovation and sustainability

Topics

Artificial intelligence | Human rights and the ethical dimensions of the information society | Closing all digital divides

Both speakers agree that multilingual AI must provide practical utility in key sectors like healthcare, education, and agriculture, but they differ in approach: Sivasubramanian emphasizes user-centered design with immediate utility as the primary driver for adoption, while Chair focuses on building systematic infrastructure and partnerships to enable widespread deployment

Speakers

– Shekar Sivasubramanian
– Chenai Chair

Arguments

Applied AI must embrace complete inclusivity across languages, rural-urban divides, and provide fundamental utility to users for adoption

Lingua Africa initiative announced as multi-partner open call focused on community-governed language infrastructure for healthcare, education, agriculture and public services

Topics

Artificial intelligence | Social and economic development | Information and communication technologies for development

Financial mechanisms | Artificial intelligence | Information and communication technologies for development

Takeaways

Key takeaways

AI development must be inclusive of Global South languages and cultures to prevent civilizational extinction and ensure equitable representation in the age of intelligence

Market failures require coordinated public sector intervention – private companies only invest in economically viable languages (English/Mandarin), leaving 2000+ African languages underserved

Computing infrastructure is a critical bottleneck – African researchers face exponentially higher costs and limited access to GPUs and high-performance computing needed for AI development

Language representation in AI is not just technical but existential – oral civilizations risk being excluded from global collective memory and intelligence systems

Successful AI adoption requires utility-first design that provides immediate value to users in their local contexts, languages, and cultural frameworks

Foundational gaps in both compute capacity and representative datasets must be addressed before communities can contribute to and shape AI systems

Gender equity and cultural nuance must be embedded in AI development, recognizing that language carries cultural values and addressing high inequality contexts

Resolutions and action items

Launch of Lingua Africa – multi-partner open call initiative focused on community-governed language infrastructure for healthcare, education, agriculture and public services

Establishment of Africa’s first dedicated public sector AI computer cluster at University of Cape Town through the African Compute Initiative

Continued funding and partnership through AI for Development programme with Canada, Germany, Japan, Sweden, Gates Foundation and other partners

Support for four additional AI startups through GSMA Foundation partnership to serve underserved populations in Asia and Africa

Development of African-specific benchmark models for speech and text that reflect African contexts rather than existing biased benchmarks

40% of Masakani funding allocated specifically to creating real-world use cases and impact applications

Project Echo implementation for gender-responsive AI interventions targeting women’s economic empowerment and health

Unresolved issues

How to ensure long-term sustainability of language AI initiatives beyond current funding cycles

Balancing open-source development with community sovereignty and preventing exploitation of local data and knowledge

Scaling solutions across 2000+ African languages with their diverse dialects and cultural contexts

Addressing the growing AI diffusion gap between Global North (developing at double the rate) and Global South

Ensuring that lab-developed AI solutions actually work effectively in real-world deployment scenarios

Managing the tension between global AI development and local community control over cultural and linguistic resources

Suggested compromises

Public-private partnerships where private sector provides technical infrastructure (like Microsoft’s compute resources) while public sector and communities maintain control over cultural content and applications

Collaborative funding model bringing together multiple governments, foundations, and organizations to share costs and risks of supporting economically unviable but socially critical language AI development

This comment provides concrete data on the AI divide while emphasizing both linguistic and contextual awareness as requirements for AI adoption. The quantification of the disparity (double the rate) gives urgency to the discussion, while the emphasis on scenario-specific functionality adds another layer of complexity beyond language.

Impact

This comment provided empirical validation for the urgency of the initiatives being discussed and introduced the concept of ‘scenario-aware AI’ alongside language-aware AI. It helped justify the compute infrastructure investments and reinforced the need for local testing and development rather than just translation of existing models.

Overall assessment

Explanation

This addresses the tension between utilizing advanced private sector AI capabilities while ensuring local communities maintain control over their cultural and linguistic assets.

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.