AI for ALL Challenge & Panel on Leveraging AI for Development in the Global South

Contents

Executive Summary

This comprehensive summit transcript covers multiple sessions addressing AI deployment in the Global South, with particular emphasis on moving from pilots to scaled government systems, sovereign AI frameworks, and AI policy education. Speakers argue that AI's transformative potential depends not on technological sophistication but on institutional capacity, governance frameworks, and deliberate policy design—and that the Global South must transition from consuming imported AI to becoming co-creators of locally adapted solutions.

Key Takeaways

The "Execution Problem" Is the Real Challenge: Success in AI for development hinges on embedding AI into owned government systems, aligning incentives, building long-term institutional capacity, and designing for scale from day one—not on better algorithms.
Sovereign AI Requires Both Independence and Partnership: The Global South must build indigenous AI governance, data frameworks, and technical capacity; but this doesn't mean rejecting global collaboration, cross-border data flows for legitimate purposes, or international standards.
Policy Education is a Lever for Equity: Democratizing understanding of how AI is governed—through schools, community initiatives, policy clinics, and sectoral training—enables citizens to become co-creators of policy rather than passive consumers of technology.
The Pilot-to-Scale Gap is a Systemic Failure, Not Individual: Without intentional redesign of funding, governance, and institutional structures, isolated pilots will continue to proliferate. Only embedding solutions into government-owned infrastructure and long-term partnerships closes this gap.
Context, Trust, and Human Dignity Are Non-Negotiable: AI that serves the poorest, most remote, and most marginalized requires that equity be a design principle from inception; that communities be represented in training data; and that human accountability structures remain intact.

Key Topics Covered

Scaling AI in healthcare, education, and agriculture across the Global South
Embedding AI into government-owned infrastructure rather than isolated pilots
Learning from India's deployment model (125 million people impacted across 25+ platforms)
South-South cooperation and knowledge sharing between emerging economies

Sovereign AI & Data Governance

Definition and implementation of "sovereign AI" frameworks
Data localization, privacy protection (DPDP Act), and cross-border data flows
Protection against data poisoning, model manipulation, and algorithmic bias
Balancing local ownership with global collaboration and innovation

AI in Defense & Strategic Security

AI as a force multiplier in multi-domain military operations
Autonomous systems, cyber warfare, and strategic deterrence
Risk of AI-enabled weapons of mass destruction and non-proliferation challenges
India's approach to responsible AI power development

AI Policy Education & Democratic Participation

Critical AI literacy for citizens, students, and policymakers
Democratizing policy discourse beyond government and tech elites
Establishing AI policy clinics and capacity-building initiatives
Addressing bias, ownership, cognitive impacts, and sectoral applications

Implementation Challenges & Execution

"Pilotism" problem: too many isolated pilots that fail to scale
Institutional adoption vs. technical sophistication
The 10-20-70 principle (10% technology, 20% process, 70% organizational/capacity)
Balancing speed of deployment with safeguards and trust-building

Key Points & Insights

Institutional Capacity > Technical Innovation: Multiple speakers emphasized that AI fails not because algorithms don't work, but because institutions lack capacity to deploy, govern, and sustain systems at scale. The "unsexy" infrastructure work (data governance, accountability mechanisms, workflow redesign) is more critical than cutting-edge algorithms.
Pilots Must Die: Strong consensus that isolated pilots perpetuate waste and delay impact. Governments should design for scale from inception, embed solutions into existing government systems, and avoid scattered experimentation. One panelist suggested shutting down 80% of current pilots.
Equity as Design Principle, Not Afterthought: If rural communities, minority language speakers, and informal workers are missing from training data, they'll be missing from AI-enabled decisions and services. Data governance failures = governance failures, not technology failures.
Sovereignty ≠ Isolation: Sovereign AI means control over critical decisions, infrastructure, and data—not complete autarky. Global South nations need indigenous capability to govern AI systems, but responsible partnerships and data flows are essential for fraud detection, innovation, and resilience.
Human Oversight Remains Non-Negotiable: AI must assist the state, not replace responsibility. If an algorithm denies a health claim or subsidy, someone must explain it to the beneficiary and provide recourse. Accountability stays with government, not vendors.
South-South Learning is Accelerating: Countries facing similar demographic pressures, fiscal constraints, and institutional complexity are sharing playbooks, governance models, and foundational infrastructure. India's example is becoming a learning ground for Africa and Southeast Asia.
Context-Driven Design Over Template Replication: Effective AI requires co-creation with local stakeholders, understanding cultural values and infrastructure realities, and adaptation rather than wholesale copying. Homegrown solutions outperform imported ones in the Global South.
The 10-20-70 Principle: While technology (the "10") gets most attention and funding, organizational adoption (the "70") is starved of resources. Development partners predominantly fund innovation rather than implementation capacity.
AI Safety & Governance Must Evolve Together: Secure-by-design principles, explainability frameworks, bias mitigation, and model drift detection are not afterthoughts—they're foundational to trustworthy sovereign AI. Regulations (like DPDP) must be operationalized with clarity.
AI Policy Education Democratizes Power: Teaching critical AI literacy—not just to technologists but to students, citizens, and marginalized communities—empowers meaningful participation in decisions affecting their lives. Small, localized initiatives are as valuable as formal institutional ones.

Notable Quotes or Statements

"The question is no longer whether the global south should engage with AI. The real question is how to deploy AI in ways aligned with national priorities, institutional capacity and governance frameworks." — Sunil Vadwani, Founder, Vadwani Institute for Artificial Intelligence

"AI will not automatically reduce inequities. In fact, if unmanaged, it will amplify it. Governance is the hard part." — Indu Bhushan, Former CEO, National Health Authority (Aayushman Bharat)

"If rural communities are missing from datasets, they'll be missing from decisions. If minority languages are absent from training models, they'll be absent from services. This is not a technology failure. This is a governance failure." — Indu Bhushan

"We do not need more pilots. We need public platforms. Speed without safeguards creates scale without trust. Without trust, AI will fail politically even if it succeeds technically." — Indu Bhushan

"Institutional adoption matters more than technical sophistication." — Sunil Vadwani

"Governments don't have endless funds. We cannot afford to waste taxpayer money. If we shut down 80% of all pilots overnight, it would be better. Scale is really hard." — Lucina Cone, CEO, Smart Africa (Paraphrased)

"The future of AI for public good will not be imported. It will be designed, led, and scaled from within the Global South." — Sunil Vadwani

"AI readiness isn't just about technical skills. It requires leadership alignment, skilled public sector teams, data governance frameworks, regulatory clarity, integration into budgets—and most importantly, a willingness to collaborate and share foundational infrastructure." — Sunil Vadwani

"Strategic stability is not defined alone by the platforms but how we sense, decide, and act faster than the adversary while avoiding any adverse unintended escalation." — LtGen D.S. Rana, Commander-in-Chief, Strategic Forces Command (India)

"Sovereign AI would ensure that critical data is controlled, dependence on external AI systems is reduced, alignment with national doctrine is maintained, and our strategic infrastructure is protected." — Alkesh Kumar Sharma, Former Secretary, Ministry of Electronics & Information Technology (India)

"AI is not destiny. It's a design. Let's design it for equity." — Indu Bhushan

"If we point AI at solving hard problems—like designing a transmission-blocking malaria vaccine—adoption becomes irresistible and inevitable." — Janet Zu, Gates Foundation

"Policy is a living thing. It needs to keep evolving with time." — Dr. Priti Ragunath, University of Sheffield

"Sovereignty is not limited to the country level. It extends to communities and individuals—questions of autonomy and privacy matter." — Dr. Priti Ragunath (Paraphrased from discussion)

Speakers & Organizations Mentioned

Government & Public Sector

Sunil Vadwani – Founder & Co-Director, Vadwani Institute for Artificial Intelligence
Indu Bhushan – Former CEO, National Health Authority; architect of Aayushman Bharat
Alkesh Kumar Sharma – Former Secretary, Ministry of Electronics & Information Technology (India)
Dr. Chandrika Koshik – Distinguished Scientist & Director General, PCIS/DRDO (India)
LtGen D.S. Rana – Commander-in-Chief, Strategic Forces Command (India)
Hon. Paula Ingabaya – Minister of ICT & Innovation (Rwanda)
Lavanch – Ministry of Foreign Affairs (Grenada); researcher, University of Cambridge

International Organizations & NGOs

Gates Foundation – Janet Zu, Director of Data & Technology Adoption
UNDP – Robert Opp, Chief Digital Officer
Tony Blair Institute – Johan Havard, Global AI Advisory Lead
Boston Consulting Group – Shalini Krishnan, Managing Director & Senior Partner
Smart Africa – Lucina Cone, CEO

Private Sector & Technology

AWS (Amazon Web Services) – Rohit Verma, VP (sovereignty, data control)
Palo Alto Networks – Nicole Quinn, VP Policy (secure-by-design, AI governance)
Bisto Credit – Rohit Gupta (SME credit, algorithmic accountability)
Intel – Kemp (heterogeneous compute, skilling, local innovation labs)
Mastercard – Lazar (sovereign AI, financial resilience, privacy)
GE Healthcare – Sheplab (medical AI, data integrity, clinical safety)
IBM – Patel (Watson X, explainability, governance toolkit)

Academic Institutions

University of Sheffield – Dr. Emma Jones & Dr. Priti Ragunath (AI Policy Clinic, critical AI literacy)
Carnegie Mellon University – Referenced for AI teaching and research excellence
University of Delhi – Akrama, Assistant Professor, Department of Political Science
ITI (Indian Institute of Technology) – Vijay Kumar, faculty and research scholar
University of Hyderabad – Dr. Priti Ragunath's PhD alma mater

Government Initiatives

India's National Health Authority – Aayushman Bharat (500M+ people digitized)
Gujarat & Rajasthan Education Departments – Early childhood reading proficiency pilots
India's Digital Public Infrastructure (DPI) – UPI, India Stack
Rwanda Development Board – Cited as institutional model for coordination
Telangana State – "Nerve centers" for AI coordination

Technical Concepts & Resources

AI Models & Frameworks

Foundation Models: OpenAI, Google Gemini, Anthropic Claude (latest releases with AI-generated development)
Large Language Models (LLMs): Indigenous/sovereign models trained on local data
Autonomous Systems: Unmanned combat systems, drones (referenced in defense context)
Computer Vision: Cough sound detection database (world's largest for TB screening)
Natural Language Processing: Gujarati, Hindi, and multi-lingual language models

Data & Governance Frameworks

DPDP Act (Digital Personal Data Protection Act) – India's privacy regulation
GDPR – General Data Protection Regulation (EU benchmark)
Data Localization & Cross-Border Transfer Restrictions – Emerging policy mechanism
Data Governance Frameworks: Critical for AI readiness; often missing
Datasets: Training data bias, minority language representation, informal worker invisibility

Platforms & Infrastructure

India Stack – UPI, Aadhaar, digital public infrastructure for AI layering
Aayushman Bharat – National health program (500M people, 30K hospitals)
Bhashini – Digital India language transition platform
Project Maven – U.S. military AI for target identification
AWS Nitro System – Customer data isolation and encryption
IBM Watson X Governance Toolkit – Explainability, bias detection, compliance monitoring
Secure-by-Design Frameworks – Palo Alto Networks sandpits, red teaming, agent identity/governance

Methodologies & Concepts

10-20-70 Principle: 10% technology, 20% process, 70% organizational/capacity
Pilotism: Pathological over-reliance on isolated pilots
South-South Cooperation: Knowledge sharing between Global South nations
Secure-by-Design: Security built in from inception, not as afterthought
Critical AI Literacy: Understanding AI's affordances, limits, consequences
Model Drift Detection: Monitoring performance degradation as real-world data diverges from training
Explainability/Interpretability: AI fact sheets, attribution of model decisions to features
Anthropometry Models: AI for body measurement (health worker efficiency)
Supply Chain Financing & Factoring: AI-enabled SME credit using transactional data
Agentic AI: Autonomous agents with individual identities and access controls

Policy & Governance Documents

India's National Strategy for AI (2018): Focus on inclusive growth across sectors
India's AI Mission (2023): Collaborative ecosystem (government, startups, academia, industry)
Regional AI Dialogues: Kigali (Rwanda) and Nairobi (Kenya) convergences
UNDP AI Adoption Across 130 Countries: 50+ specifically on AI adoption

Research & Evidence Gaps

No centralized repository of effective AI policy education practices – identified as gap
Limited research on cognitive impacts of student reliance on generative AI – underway at Sheffield
Copyright/IP law unsettled regarding AI-generated works, especially on indigenous aesthetics
Indigenous Data Sovereignty: Microsoft's partnership with New Zealand indigenous communities (model referenced)

Areas of Tension & Ongoing Debate

Speed vs. Safeguards: Pressure to scale fast conflicts with need to build trust and prevent harm.
Sovereignty vs. Collaboration: Protecting national interests clashes with necessity of global data flows and partnerships for fraud detection, innovation.
Regulation vs. Innovation: Overly prescriptive rules stifle experimentation; insufficient guardrails create risk.
Generalization vs. Contextualization: Universal AI policy principles needed, but implementation must be locally adapted.
Funding Allocation: Philanthropic and development dollars flow to innovation (the "10") rather than organizational capacity (the "70").

Recommended Next Steps (Implicit from Summit)

Establish AI policy clinics and capacity-building initiatives in universities, government agencies, and community organizations
Fund institutional adoption and implementation (the "70"), not just innovation pilots
Embed technical AI capacity inside governments as long-term partners
Co-develop sovereign AI frameworks that balance local control with international collaboration
Operationalize privacy/governance laws (DPDP, GDPR) with clear timelines and sectoral guidance
Map and share best practices in AI policy education across regions
Design for scale and cross-government integration from the outset
Establish accountability and explainability mechanisms before deployment
Build critical AI literacy curricula for schools, universities, and public sector

Overall Significance

This summit reflects a critical inflection point in global AI governance: the Global South is no longer asking whether to adopt AI, but how to govern it responsibly and equitably. The dominant theme is that institutional capacity, deliberate policy design, and human-centered governance are more important than technological sophistication. Success requires moving from imported solutions to co-created frameworks, from isolated pilots to embedded systems, and from passive consumption to active co-creation of AI policy. The stakes—in health, education, security, and human dignity—are extraordinarily high.