AI for India’s Next Billion

Contents

Executive Summary

This AI summit panel discussion addresses how the Global South can leverage AI as a genuine growth engine while avoiding the concentration of power and value that characterized previous technology waves. The speakers emphasize that AI governance must prioritize accountability, inclusivity, and equity—ensuring that the next billion people benefit from AI rather than being marginalized by it. Central tensions include balancing innovation with regulation, addressing the energy/environmental costs of AI infrastructure, and centering the voices and labor of Global South workers in the AI supply chain.

Key Takeaways

Accountability is diffuse by design, but it doesn't have to be directionless: Governments must create independent oversight bodies; companies must be accountable to external entities; citizens must demand accountability. This three-pillar model translates democratic governance to AI.
The Global South must build its own AI infrastructure using its own data: Relying on foreign models and platforms risks neocolonialism. India's strategy—building multiple large language models, opening data via AI Coaches, and providing subsidized compute to startups—shows a path forward that other countries should study.
Labor and workers are the foundation of AI, not an afterthought: Data annotation, content moderation, and evaluation require fair wages, safe working conditions, dignity, knowledge transfer, and community input into how models are built. Responsible AI cannot exist without worker-centered design.
Climate action and AI development are not zero-sum: Data centers demand energy and water, but AI can optimize renewable integration, improve flood forecasting, and boost agricultural resilience. The goal is intentional, transparent measurement of environmental footprints and collaboration on shared infrastructure standards.
Capacity, not just code, determines who benefits: Governance capacity, technical fluency among leaders, collaborative regional frameworks (South-South cooperation), and investment in the full AI stack (data, compute, talent, use cases, trust) determine whether AI amplifies or reduces inequality.

Key Topics Covered

AI Governance & Accountability: Who bears responsibility when AI systems cause harm; the role of governments, companies, and citizens
Global South Development: How emerging economies can build indigenous AI capabilities without reproducing colonial-era power imbalances
Digital Public Infrastructure (DPI): The Indian model of open-source, interoperable public infrastructure as a template for AI governance
Data Sovereignty & Value Extraction: Preventing the Global South from becoming a data source while remaining excluded from value creation
Environmental Sustainability: Reconciling AI's energy/water demands with climate action and decarbonization goals
Infrastructure & Capacity Building: Addressing connectivity, energy access, and governance capacity in low-income regions
Labor & Worker Rights in AI: Fair wages, safety, dignity, and knowledge transfer for data annotation and AI training workers
Multilingualism & Inclusion: Ensuring AI systems reflect the languages, cultures, and lived realities of diverse populations
Youth & Demographic Opportunity: Leveraging Africa's median age of 19.7 and India's young population in the AI transition
Regulation vs. Innovation: Risk-based governance frameworks that protect people without stifling development

Key Points & Insights

Accountability must sit outside companies: Self-regulation by tech companies has consistently failed. Independent oversight bodies, government capacity, and citizen participation are essential—mirroring the traditional democratic model applied to a new context.
Data is being extracted from the Global South at scale: India provides 33% more data to large language models than the United States. These models are refined on Global South data, then sold back to those same regions at high cost. Indigenous model-building using local data is critical.
Digital Public Infrastructure enables equitable scaling: India's UPI (Unified Payments Interface) achieved 50 years of financial progress in 7 years by being open-source, API-driven, and globally interoperable. An analogous DPI layer for AI could democratize access while allowing private-sector innovation on top.
The false binary between regulation and innovation: Right governance guardrails can accelerate innovation. Risk-stratified approaches (high oversight for life-or-death decisions; experimentation for low-risk cases) protect people without freezing delivery.
"Unequal capacity leads to unequal outcomes": Without investment in governance capacity, technical expertise, and institutional strength in the Global South, countries will be unable to absorb, govern, or benefit from AI. This exacerbates existing inequalities.
Workers in the Global South are invisible in AI value chains: India, Kenya, Uganda, and the Philippines host vast workforces annotating and evaluating training data. These workers often earn low wages, face psychological harm from content moderation, and are excluded from knowledge-building about the systems they train.
Data quality requires cultural and linguistic grounding: Gender bias, for example, is defined differently across languages. A study with 20,000 Indian women showed that gender bias in local languages centers on community/personality/emotions, not profession/occupation. Models trained without this contextual knowledge reproduce local harms.
Frugal AI = Optimal AI: Resource constraints should drive intentional optimization for use cases, geographies, and choke points—not represent a second-class approach. Efficiency can be a competitive advantage.
Africa's challenges are foundational, not optional: 600 million Africans lack grid electricity; the continent has the lowest connectivity. AI cannot leapfrog infrastructure gaps; governments must address energy and connectivity first while simultaneously exploring AI for acceleration.
Intergenerational and participatory design is non-negotiable: Young people should shape AI governance, not simply inherit its consequences. Without their participation and that of affected communities, inequality will calcify into a "K-shaped economy."

Notable Quotes or Statements

"The fact that accountability is so diffuse can be the basis for a flexible, vibrant, appropriate way of thinking about AI...or it can be a recipe for nobody taking responsibility. Which one it is depends on decisions made in places like this, in governments, in companies." — Unidentified panelist on accountability

"If we don't use the power of our own data, if we don't use our own history, our own civilization, it will be neocolonization in reverse. We'll again become colonized on the basis of the model." — Shri Amitabh Kant (India)

"Unequal capacity will lead to unequal outcomes." — Multiple speakers, emphasizing capacity-building as essential

"People at the bottom of the AI pyramid—the workers in India, Kenya, Uganda, Philippines—are the invisible builders of the models we use. We need to stop treating them as simply workers and start empowering them with knowledge, dignity, and fair wages." — Safia (Karya organization), on labor in the AI supply chain

"We need to reject this false binary that either you have climate action or you have an AI-driven ambition. You can have both—with intentionality." — Dr. Arunabha Gosh (Climate Change and Sustainability)

"If today's inequalities continue into the AI era and we end up with a K-shaped economy, we will have failed. Young people should not be passengers—they should actively shape the outcome." — UN Undersecretary-General (on intergenerational AI governance)

"Most of the time we talk about gender bias in English—profession, occupation. But for 20,000 women across Indian languages, it was actually community, personality, emotions. Without that understanding, we build models that don't address core lived experiences." — Safia (Karya), on culturally grounded data

Speakers & Organizations Mentioned

Government & Policy:

Shri Amitabh Kant (Government of India; G20 leadership)
Ambassador (Kenya/African perspective on AI infrastructure)
His Excellency (Global AI governance, UN role)
Government of India (summit host)

International Organizations:

United Nations (UN Foundation partnership)
UN Environmental Programme (UNEP)
Mandela Institute (referenced collaboration)
UN General Assembly (resolution on AI environmental sustainability)

Research & Civil Society:

Dr. Arunabha Gosh (Council on Energy, Environment and Water — CEEW)
Safia (Karya organization — worker-focused AI ethics)
Stanford University (AI Vibrancy Index)
Gates Foundation (funding community engagement projects)
UN Foundation (summit co-organizer)

Companies & Platforms:

OpenAI (ChatGPT)
Google (mentioned in solar/agricultural AI sessions)
Nvidia (GPU supplier; 11 GW deal with OpenAI referenced)
Claude (AI assistant; mentioned by Shri Kant)
Grok (referenced for security issues)

Geographic/Regional References:

India, Kenya, Uganda, Philippines (major AI worker hubs)
South Africa, Brazil, Nigeria (emerging AI capability)
Africa (54 countries; 1% of global data centers)
Global South (primary focus of equity discussion)

Technical Concepts & Resources

AI Systems & Models:

Large Language Models (LLMs): ChatGPT, Claude, Grok—trained on Global South data
Data Annotation/Labeling: Core task of AI workers; foundation of model quality
Bias in AI Systems: Gender bias, cultural bias—requires multilingual, contextual training data
AI Vibrancy Index (Stanford): Ranking countries by AI capacity; India, Singapore, UAE in top 10

Infrastructure & Architecture:

Digital Public Infrastructure (DPI): Open-source, API-driven, interoperable systems (India's UPI model)
Data Centers: Energy/water-intensive; cooling technologies (air-cooled vs. water-based); only 1% in Africa, 50% in South Africa
GPUs (Graphics Processing Units): Nvidia; 50,000–60,000 needed for large AI systems; OpenAI deal: 5 million GPUs, 11 GW power demand
Open-Source & Open APIs: Key to democratizing AI access

Governance & Policy Frameworks:

Risk-Stratified Regulation: High oversight for high-impact cases (medical, financial); permissive experimentation for low-risk use cases
Interoperability Standards: Reduce costs, accelerate diffusion without limiting innovation
UN Resolution on AI Environmental Sustainability: First-ever; stock take on energy-efficient technologies and data center standards
IPCC-like Science Panel: UN initiative for iterative, interdisciplinary AI governance guidance

Data & Datasets:

Gender Bias Framework (Karya study): 20,000 women across 6 Indian language communities defining culturally-specific gender bias markers
Data Sovereignty: Building models on local data; preventing extractive data practices
Multilingual Data: Essential for inclusive AI (e.g., VHP language support for health chatbots)

Emerging Concepts:

Frugal AI: Optimal AI—optimized for resources, use cases, places; not a second-class approach but a competitive advantage
K-Shaped Economy: Widening inequality; risk if AI benefits concentrate among the wealthy
Neocolonization via AI: Using others' data to build models, then selling back to them at high cost

Historical References & Analogies:

IPCC Model (Intergovernmental Panel on Climate Change): Template for inclusive, science-driven governance
Previous Technology Waves: Electricity, nuclear energy, aviation, internet—all cases where innovation outpaced governance
Hamurabi's Code (3,700 years old): Derived from pre-existing legal traditions; illustrates importance of drawing on diverse governance traditions
Technology Race History: Clive Osborne (first laptop) vs. Apple; Alta Vista (first search) vs. Google—first mover doesn't always win

Case Studies & Real-World Examples:

India's UPI: 50 years of financial progress in 7 years via open-source, interoperable DPI
Kenya's M-Pesa/Mobile Money: Global largest user of M-Pesa; $20 access cost (high relative to income)
GitHub Contributions: 1.2 million Nigerians, 300,000 Kenyans contributing free AI/code work
India AI Initiatives: 10 large language models in development; AI Coaches; subsidized compute for startups
Data Center Scaling Report (Systemic/CEEW): Best practices for sustainable data center expansion in India

End of Summary