India AI Impact Summit Insights
All sessions

AI for Development Conversations: India AI Impact Summit 2026

Contents

Executive Summary

This panel discussion examines the critical gap between AI innovation and equitable adoption in the Global South, particularly India. The panelists—representing philanthropy, civil society, open-source advocacy, and policy—argue that while AI presents transformative opportunities for development, current approaches risk perpetuating exclusion, vendor lock-in, and labor exploitation unless communities are meaningfully involved in design, data governance, and decision-making.

Key Takeaways

  1. Inclusion in design is prerequisite for adoption: Communities must participate in designing AI tools from inception, not receive them as finished products. This prevents both poor fit and vendor dependency.

  2. "Open source" without governance access and data rights is extractive: Transparency and code freedom alone don't guarantee equity if communities lack agency in defining requirements, controlling their data, or sharing in value. True openness requires architectural (open code + open data + open governance).

  3. Sustainable open-source AI in the Global South requires institutional innovation: Neither pure corporate models nor pure volunteer models work. Hybrid models involving patient philanthropic capital, government support for infrastructure, and transparent pricing for data/annotation labor are emerging—but remain early and under-resourced.

  4. AI deployment decisions are procurement and governance problems, not just technology problems: The boring work of transparent procurement, institutional capacity-building, and regulatory frameworks matters as much as innovation. Without it, public AI spending risks concentrated vendor lock-in.

  5. Lessons from India's DPI success are transferable but not automatic: India's ability to rapidly build ecosystems (payments, identity, digital services) offers templates for AI infrastructure—but only if the same discipline around inclusion, transparency, and iterative evaluation applies.

Key Topics Covered

  • Exclusion in AI Design: Non-profit and community organizations are excluded from technology design processes, leading to poor adoption and vendor lock-in
  • Open Source vs. "Open" AI: The distinction between truly open-source AI (transparency, community contribution) and superficially open models (closed research, data, or tooling)
  • Unpaid Labor & Sustainability: The tension between open-source's collaborative ethos (libre/freedom) and gratis (free-of-cost) models, and the unsustainability of unpaid indigenous labor
  • Data Governance & Community Rights: Models for bottom-up data ownership, monetization, and collective negotiation—particularly for marginalized communities
  • Business Models for Open-Source AI: How to sustainably fund and maintain open-source tools without corporate capture or dependency
  • Job Displacement & Poverty Alleviation: The risk of AI-driven labor displacement versus potential for augmenting human capability and creating livelihoods
  • Institutional Ecosystems: The role of government, philanthropy, civil society, academia, and the private sector in building equitable AI infrastructure
  • Public Goods vs. Public Infrastructure: The distinction between creating freely available resources and operationalizing them equitably
  • Regulatory & Policy Frameworks: Light-touch regulation, procurement standards, and lessons from India's digital public infrastructure (DPI) successes
  • Global South Representation: The importance of diverse participation from the Global South in AI governance and deployment conversations

Key Points & Insights

  1. Design Exclusion Perpetuates Dependency: Community organizations are locked into vendor solutions because they're not involved in design. AI amplifies this problem by adding technical complexity and creating new dependencies on compute infrastructure and proprietary models.

  2. "Open Source AI" Is Contested: Many projects labeled "open-source AI" are not truly open—research may be closed, data proprietary, tooling proprietary, or only the LLM weights released. Terminology matters because it shapes expectations of transparency and community benefit.

  3. The Global South Labor Paradox: Open-source sustainability in the Global South differs fundamentally from the North. Northern models rely on paid engineers from tech companies contributing during work hours. Southern models require external patronage (philanthropic funding) because open-source hours compete with survival livelihood. Without patronage infrastructure, indigenously developed open-source becomes unsustainable.

  4. Data as New Labor: The emerging data economy (annotation, transcription, training data creation) represents potential livelihood for marginalized communities, but only if frameworks exist for fair pricing, transparency, and collective negotiation. Current models often default to extraction: communities contribute data, large tech companies profit.

  5. Accountability Through Transparency: Open-source code enables accountability because it's auditable—security gaps can be spotted and patched by anyone. This creates social capital for contributors and alternatives to proprietary vendor solutions, though government accountability mechanisms remain more complicated.

  6. Three Distinct Notions of "Access": Access to data ≠ access to compute ≠ access to governance participation. True equitable access requires all three, but interventions often focus narrowly on one layer (e.g., providing compute credits without governance rights).

  7. India's Ecosystem Building Capability: India has demonstrated rapid capacity to build ecosystems (DPI, UPI, EKYC, India AI Mission). The India AI Mission's approach—$1.2B capital outlay with 40% reserved for compute subsidies and private sector investment—shows how public funds can catalyze private capacity-building and price discovery.

  8. Procurement as Boring But Critical: Good institutional practice for public AI deployment requires transparent procurement, vendor justification, monitoring and evaluation at iterative pace. This prevents local optima and vendor lock-in but is unglamorous and under-discussed.

  9. Bottom-Up Data Governance Models Exist: Cooperatives for data pooling, indigenous data rights frameworks (from New Zealand, Australia, Canada), and collective negotiation models show alternatives to individual data monetization. Legal frameworks lag behind (most understand only individual data rights, not collective ones).

  10. Job Displacement Requires Differentiated, Systemic Response: No single solution exists. Interventions must depend on which jobs are affected, how quickly, and at what scale. Past experiences with reskilling (leap-frogging multi-year experience gaps) and structural shifts (cars enabling people who couldn't ride horses) offer limited templates.

Notable Quotes or Statements

"They're first of all not involved in the design, which makes it very difficult for them to adopt these technologies... they're unable to create systems which talk to each other. They're unable to afford them, so they stop using it after a year or so." — On why community organizations struggle with tech adoption

"When the open source movement came about, it required a free exchange because it's probably the largest sort of democratic way of collaborating in the world. But some of the things that we think are open source right now are not really open source because the research is probably not open. The tooling is not open. The software is not open. Just because the LLM is open doesn't necessarily mean it's truly open." — Anirudh, on the contested meaning of "open source AI"

"The hours you spend on it, if they're competing with livelihood, becomes a huge challenge. That tends to happen a lot more in the global south. And as a result of that, the nature of open source here looks fundamentally different." — On open-source sustainability in the Global South

"If you create a public good out of a data set, even through philanthropic investment, and you make it available to everybody, it's going to be available to everybody. And then the question of entrenchment—if somebody can pay for the bandwidth to download it off Amazon better than somebody else can... it's very hard to then retain that sense of social capital growing by contribution with fidelity." — On how market power undermines community benefit from public goods

"There is no question about [traffic jams]. There is no question about it and you will need structures to address those sort of consumption asymmetry issues that will keep cropping up and that is the nature of living systems. And that's what we build societies for." — On why public infrastructure requires ongoing governance, not just technical deployment

"We essentially want to look at things that augment human capability rather than replace it... It's these are evolutionary changes and we need an adaptation question. So it's really important to start looking around and seeing what have we not been paying attention to? What has not been given due importance? We've not talked enough about nuance of power, agency, control." — On reframing AI's role in labor and development

Speakers & Organizations Mentioned

Named or Identifiable Panelists:

  • Eli – Open-source advocate (likely from a tech/civil society organization)
  • Arjun – Open-source/ecosystem advocate (likely from civil society)
  • Anirudh – Policy/governance perspective
  • Asad – Summit organizer/moderator
  • Prena – Moderator or speaker

Organizations & Institutions Referenced:

  • Gates Foundation / Gates India Office – Philanthropic funder; ran "Sanmati" (Common Sense) program
  • Karya – Startup doing data annotation with community workers
  • The Nudge – Organization developing digital skills certification framework for data annotators
  • Aapti Institute – 7-year focus on bottom-up data governance and community data rights
  • Tech for Good Community – Partner promoting free and open-source software in the Global South; created "It Lee Stack"
  • Indian Institute of Science – Collaborated on speech recognition and text-to-speech transcription for Indian languages
  • Ministry of IT (MeitY) – India's government tech ministry; praised for light-touch regulation on AI
  • India AI Mission – Government initiative for compute infrastructure build-up ($1.2B)
  • Hewlett Foundation – Philanthropic support for conversations on open-source accountability
  • Ministry of Common Service Centers – India government program with 650,000 centers and youth workforce for digital services
  • Sovereign Tech Fund (Germany) – Government fund supporting open-source innovation

Policy/Regulatory References:

  • DPI (Digital Public Infrastructure): India's identity, payments, data infrastructure
  • UPI (Unified Payments Interface) – India's real-time payment system
  • EKYC (Electronic Know Your Customer) – Digital identity verification
  • Data Protection Laws – Referenced as less than 20 years old globally; India's personal vs. non-personal data distinction
  • IPR (Intellectual Property Rights) – Older frameworks; new digital data laws lag behind

Technical Concepts & Resources

Tools & Datasets:

  • LLM (Large Language Models) – Referenced as core to "open source AI" but distinctions made between open weights vs. open training data/methodology
  • Neural Machine Translation – Pre-GPT approach for speech recognition and translation (mentioned in context of Indian language projects)
  • Automated Speech Recognition (ASR) – Application for Indian languages, agriculture, and finance use cases
  • Text-to-Speech (TTS) – Similar application layer
  • GPU Compute – Subsidies, pricing, and capacity-building (40% of India AI Mission funds reserved for subsidies)

Frameworks & Methodologies:

  • Gartner Hype Curve – Used to measure adoption timelines for digital infrastructure (ID movement vs. UPI vs. EKYC)
  • Data Annotation/Labeling – Emerging as labor/monetization opportunity; pricing frameworks in early stages
  • Digital Skills Certification – Being developed to standardize and price data annotation labor

Papers & Resources Referenced:

  • "Decentralized Data Flows" – Paper by Pramod Varma (released during summit week; recommended for data governance models)
  • "It Lee Stack" – Open-source/low-cost tech stack for communities (created by Tech for Good Community)
  • Sanmati (Common Sense) Dataset – Public dataset from Gates Foundation + Karya + women annotators in 6 Indian languages; ~20,000 women contributors; focused on bias in cultural context

Governance & Rights Models Referenced:

  • Indigenous Data Sovereignty – Frameworks from New Zealand, Australia, Canada for collective community data rights
  • Data Cooperatives – Organizational model for collective data pooling and negotiation
  • Labor Pricing Frameworks – Emerging models for pricing annotator work similarly to commodity pricing (e.g., GPU pricing discovery)

Policy Concepts:

  • Light-Touch Regulation – MeitY's approach to AI governance in India (praised as balanced)
  • Public-Private Partnerships – Common Service Centers model (government + private sector youth training)
  • Patient Capital – Philanthropic funding that doesn't demand immediate ROI, enabling long-term structural work

Limitations & Gaps

  • No concrete case studies of scaled solutions: Most examples are small pilots (Sanmati: 20,000 women) or emerging initiatives (data pricing frameworks), not proven at national/regional scale
  • Unanswered business model question: Panelists acknowledge they don't have definitive answers for sustaining open-source AI in the Global South
  • Limited discussion of failure modes: How to prevent public AI investments from creating new monopolies or reinforcing existing power asymmetries
  • Job displacement timeline unclear: Panelists note lack of data for modeling which jobs are at risk and at what speed

Relevance for India Specifically

This discussion is particularly relevant to India because:

  1. India has proven capacity to rapidly build shared digital infrastructure (DPI) but faces scaling equity challenges
  2. India's 650,000+ Common Service Centers and youth workforce represent a potential distribution channel for AI tools, but risk becoming vendors of proprietary systems without intentional governance
  3. India's diverse languages, informal economy, and distributed population make it a crucial testbed for inclusive AI but expose risks of centralized, English-first solutions
  4. India's government (MeitY) is actively shaping AI regulation; the summit and this discussion directly inform that policy-making