Responsible AI for Shared Prosperity

Contents

Executive Summary

This AI summit panel discussion centers on making artificial intelligence accessible and beneficial across Africa and Asia by developing AI systems in local languages and ensuring equitable access to computing resources. The speakers emphasize that language-aware, culturally sensitive AI is essential for achieving the UN Sustainable Development Goals and preventing the global south from being excluded from the AI revolution—a critical intervention since current AI development heavily favors English and Mandarin-speaking markets.

Key Takeaways

Language is not merely a technical feature—it is a civilizational imperative. Without local-language AI, entire cultures and knowledge systems risk exclusion from global AI benefits and historical memory. This is a matter of equity, representation, and survival of non-Western intellectual traditions.
Markets alone will not solve this. Public investment in linguistic AI infrastructure is essential because commercial incentives naturally concentrate on high-volume languages. Governments and foundations must coordinate to fund open, community-governed language infrastructure as a public good.
Compute access is the foundation. Without local high-performance computing infrastructure, researchers and developers in the Global South cannot participate meaningfully in AI innovation. The African Compute Initiative and similar projects are not luxuries—they are prerequisites for participation.
Real-world impact requires utility and co-design. AI systems succeed only when they solve specific, locally relevant problems in local languages. Top-down, lab-based solutions fail. Community input, local data collection, and continuous testing with end users are non-negotiable.
Sovereignty matters. Inclusive AI must also be equitable AI, meaning communities retain control over their data, influence model development, build internal capacity, and sustain initiatives independently over time. Partnership is not patronage.

Key Topics Covered

Multilingual AI Development – Building AI capabilities in African and Asian languages to ensure inclusion and cultural representation
Language as a Civilizational Issue – Framing language in AI not merely as a technical problem but as essential to preserving non-Western cultures and knowledge systems
Computing Infrastructure Gaps – Addressing the lack of GPU and high-performance computing access in the Global South
Community-Led Data Collection – Bottom-up approaches where local communities define and collect linguistic data
Applied AI for Development – Deploying AI in health, education, agriculture, and public services with measurable impact
Gender-Responsive AI – Ensuring AI initiatives address inequality and support women's economic empowerment
Public-Private Partnerships – Coordinating funding between governments (UK, Canada, Germany, Japan, Sweden), foundations (Gates), tech companies (Microsoft), and research institutions
AI Governance & Sovereignty – Protecting local data rights and ensuring communities retain control over their linguistic and cultural assets

Key Points & Insights

Civilization-Scale Problem: Kenya's technology envoy (Phillip Figo) frames the issue as existential—without representation in AI systems, non-Western cultures and civilizations risk erasure from the "age of intelligence." Language encodes culture, values, and history; neglecting local languages in AI means erasing entire knowledge systems.
The Market Fails for Low-Resource Languages: Private companies rationally invest in English and Mandarin because the economics work. Broken markets require state and philanthropic intervention to fund public goods (language datasets, models, tools) that serve underrepresented populations.
Utility Drives Adoption: Wadwani AI's experience in India demonstrates that AI applications succeed only when they deliver tangible value to end users in their own language. Example: an oral reading fluency tool that provides real-time feedback helps teachers and children—the language and utility are inseparable.
Masakani African Languages Hub represents genuine community-led initiative (originating in 2019 with no initial funding) now scaling to impact 1 billion Africans across 50+ languages. The hub operates on four pillars: data collection, research/benchmarking, innovation/use-case deployment, and institutional sustainability.
Lingua Africa Initiative (announced during the panel) is a multi-partner, open-source infrastructure project addressing the gap between language data and real-world deployment—ensuring that linguistic resources translate into functioning healthcare, education, and public service applications.
Compute as Critical Enabler: Microsoft and others emphasize that compute (GPUs, storage, networking) is fundamental to every stage: training locally collected data into models, testing models with local speakers, and day-to-day deployment. Africa's compute costs are exponentially higher than in the Global North, creating a structural barrier.
African Compute Initiative: A dedicated high-performance computing cluster at University of Cape Town aims to provide African researchers and institutions direct access to GPUs and infrastructure—removing cost and access barriers that currently prevent local innovation.
Linguistic Diversity Requires Context-Specific Approaches: The same language spoken in different regions (e.g., Shona in Harare vs. Mutare; Swahili variations across East Africa) has distinct dialects and needs. AI benchmarks must reflect these regional nuances, not assume standardized language models work universally.
Gender-Responsive Interventions Matter: Project Echo (Enhancing Communications for Her Opportunities) explicitly addresses gendered inequality in AI applications for women's economic empowerment and health—recognizing that inclusive AI must also be equity-focused.
Data Sovereignty & Institutional Capacity: Success requires not just funding but building local research and development capacity so African and Asian institutions can innovate independently, own their data, and sustain initiatives beyond external partnership periods.

Notable Quotes or Statements

"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 cultures values have been coined in language." — Phillip Figo, Technology Envoy, Government of Kenya

"It is a civilizational discussion." — Phillip Figo

"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." — Anchor (Gates Foundation)

"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 that." — Sheka Subramanian, CEO Wadwani AI (humorous critique of theory-practice gaps in development work)

"If you can work the value then the adoption is easy. If you divorce the two people don't understand why I'm doing what I'm doing. It looks like an encumbrance." — Sheka Subramanian, on the necessity of linking language tools to tangible utility

"Trustworthy AI is not going to happen by accident. It's going to happen because of the choices that we make, the ways in which we choose to build and test and deploy these AI systems." — Natasha Crampton, Chief Responsible AI Officer, Microsoft

"Language is not a barrier anymore into including people into these solutions." — Shennai (Masakani Hub), describing Lingua Africa's goal

Speakers & Organizations Mentioned

Government & Policy

UK Deputy Prime Minister (host/moderator)
Phillip Figo – Special Technology Envoy, Government of Kenya
Dr. Beal Kofla – Parliamentary State Secretary to the Federal Minister for Economic Cooperation and Development, Germany
German government (funder)
Japanese government (funder)
Swedish government (funder)
Government of Canada (funder)

Researchers & Technical Leaders

Sheka Subramanian – CEO of Wadwani AI; Director, Masakani African Languages Hub
Shennai – Representative, Masakani African Languages Hub (announcing Lingua Africa)
Natasha Crampton – Chief Responsible AI Officer, Microsoft
Anchor (unclear full name) – Chief Strategy Officer & President of Africa and India Office, Gates Foundation
Julie Deahanti – President, Canada's International Development Research Center (IDRC)

Institutions & Organizations

Masakani African Languages Hub – Community-driven initiative (founded 2019) to represent African languages in AI
Wadwani AI – Applied AI organization working in health, education, agriculture across India
Microsoft – Cloud computing and AI infrastructure partner
Gates Foundation – Philanthropic funder and partner
IDRC (International Development Research Center, Canada) – Research funder and partner
University of Cape Town – Host of African Compute Initiative
GSMA Foundation – Partner in startup support
Germany's Fair Forward Initiative – Development program supporting multilingual AI data collection

Startups & Programs

Torn AI (Morocco) – Voice interface startup for low-literacy rural users
Project Echo (Enhancing Communications for Her Opportunities) – Gender-responsive AI intervention
Lingua Africa – New multi-partner open-source language infrastructure initiative
African Compute Initiative – Dedicated GPU cluster at University of Cape Town
Asia AI for Development Observatory – New network for responsible AI governance in Asia

Technical Concepts & Resources

Models & Datasets

Large Language Models (LLMs) – Base models (predominantly trained on English) requiring fine-tuning for local languages
JW300 Bible Dataset – Early dataset used by Masakani community to bootstrap African language data
Oral Reading Fluency Tool – AI system that assesses children's reading comprehension in local languages and provides teacher guidance

Infrastructure & Compute

GPUs (Graphics Processing Units) – Essential hardware for training and testing AI models; expensive and scarce in Africa
High-Performance Computing (HPC) Clusters – Dedicated infrastructure for model training, research, and deployment
Cloud Computing – Microsoft Azure and similar platforms used for remote compute access
Network Infrastructure – Critical for accessing distributed compute resources

Methodologies & Frameworks

Benchmarking Projects – Developing African-specific AI benchmarks that reflect speech and text nuances in regional contexts (not copying Western benchmarks)
Community-Led Data Collection – Bottom-up processes where linguistic communities define and gather representative datasets
Applied AI / Use-Case Driven Development – Designing systems around specific real-world problems (health, education, agriculture, public services) rather than generic models
Gender-Responsive Design – Intentional integration of gender equity considerations into AI applications
Four-Pillar Approach (Masakani) – Data collection, research/refinement, innovation/deployment, sustainability

Linguistic Considerations

Dialectical Variation – Acknowledging that the same language differs across regions and contexts (Shona in Harare ≠ Shona in Mutare)
Oral vs. Written Language – Addressing cultures with strong oral traditions, not just text-based AI
2,000+ African Languages – Emphasizing diversity; no one-size-fits-all solution across the continent
14–16 Indian Languages – Minimum design requirement for inclusive applications in India
Multilingual NLP (Natural Language Processing) – Advancing machine learning models capable of handling multiple languages simultaneously

Governance & Sustainability

Data Sovereignty – Communities retaining ownership and control over collected linguistic data
African-Led AI – Ensuring initiatives are designed by and for African researchers and communities, not externally imposed
Institutional Capacity Building – Training and supporting local NLP research communities to sustain work independently
Open-Source Models – Making code and models freely available so communities can innovate and commercialize solutions

Document Note: This transcript appears to be from the 2023 or 2024 AI Summit (likely in India, given location references). The first Bletchley Park AI Summit is referenced as occurring three years prior, suggesting this talk is contemporary to the recent global AI governance discussions. The emphasis on responsible AI, digital public infrastructure, and equitable development reflects current international development and AI governance priorities.