India AI Impact Summit Insights
All sessions

One Billion Futures: AI and Education Equity in the Global South

Contents

Executive Summary

This panel discussion at an AI Impact Summit examines how AI can serve as an equalizer for education in the Global South rather than widening existing inequities. The speakers argue that AI-driven personalized learning, when designed thoughtfully and deployed with intentionality around infrastructure, language, and local context, can address critical gaps affecting 1 billion young people in developing nations who currently lack adequate educational access and quality.

Key Takeaways

  1. AI is infrastructure for learning equity, not a consumer product. Educational AI must be evaluated by learning outcomes, equity metrics, and teacher/student agency—not by feature parity with commercial LLMs.

  2. Design intentionally for context. Effective AI solutions in the Global South require offline-first design, local language support, use of abundant communication channels (WhatsApp, voice), and curriculum flexibility—not one-size-fits-all Western models.

  3. Infrastructure gaps are real but not paralyzing. Significant disparities in GPU access, energy, and connectivity exist, but they create opportunities for localized solutions, entrepreneurship, and leapfrogging—not reasons to stop.

  4. AI literacy must be ladder-based and foundational. Start with reading, numeracy, misinformation detection, and safe AI use; build toward deeper technical understanding based on existing knowledge and access. Context determines starting point, not age.

  5. Partnership is structural, not aspirational. Realizing AI's potential for education equity requires formal collaboration between communities (defining objectives), governments (infrastructure, guardrails), educators (pedagogy), and companies (tools, transparency)—not siloed action.

Key Topics Covered

  • Education Equity & Access Crisis: 270 million out-of-school children globally; 70% of children under 10 cannot read grade-appropriate text; severe underinvestment in Global South education systems (~$55 per child)
  • AI as an Equalizer vs. Divisor: How AI can either perpetuate or close educational divides depending on design intentionality
  • Personalized Learning at Scale: Role of knowledge tracing, real-time student assessment, and adaptive content delivery through AI tutoring systems
  • Infrastructure & Energy Requirements: GPU access disparities, data center energy demands, computational costs of large language models
  • AI Literacy Frameworks: Contextual, ladder-based approaches to AI literacy that differ by geography and existing foundational skills
  • Curriculum & Language Localization: Addressing bias in Western-trained models; supporting 22+ Indian languages; building offline-first and low-connectivity solutions
  • Teacher Empowerment & Agency: How AI systems should augment rather than replace teachers; providing real-time insights into student learning gaps
  • Stakeholder Partnerships: Defining AI literacy requires collaboration between local communities, governments, educators, and technology providers

Key Points & Insights

  1. Scale of the Opportunity: 4 in 5 people on the planet live in the Global South; 90% of young people globally are in the Global South—making education equity a critical lever for demographic dividend and economic development.

  2. Colonial Legacy & System Design: Education systems in the Global South have been shaped by colonial history, systemic underinvestment, and language barriers that continue to exclude learners from instruction in their native languages.

  3. From Answer-Checking to Diagnostic Learning: Traditional AI/online education focused on right/wrong answers; modern AI can now provide granular diagnostic insights (e.g., identifying that a student confuses chemical combustion with nuclear fusion) within 2-4 interactive questions.

  4. Infrastructure is Not Binary: Rather than waiting for perfect connectivity/hardware, successful implementations use "offline-first" design, WhatsApp chatbots, voice-enabled tools, and hybrid models to serve low-connectivity contexts while building toward better infrastructure.

  5. GPU & Energy Asymmetry: Massive disparities exist—OpenAI accessed ~1 million GPUs last year; Africa has access to <10,000; India is building toward 250,000. Energy demands of AI clusters create both infrastructure challenges and job creation opportunities (e.g., nuclear power expertise).

  6. AI Literacy is Contextual & Layered: A 14-year-old in Chhattisgarh may start with misinformation recognition and voice-based chatbots; a 14-year-old in California may explore bias in datasets and build applications. Both require foundational literacy as a prerequisite—no leapfrogging.

  7. Capability ≠ Geography: Students in the Global South demonstrate equivalent intellectual capabilities to students in the West; disparity stems from access, not ability. The speaker emphasized this explicitly to counter assumptions about intelligence gaps.

  8. Systemic Integration Over Point Solutions: Effective educational AI requires integration of teacher, curriculum, student, and tutoring functions into a unified platform—not fragmented tools (lesson plans, tutors, assessments in silos).

  9. Data Sovereignty & Cost Control: CK12 and similar organizations built proprietary content, avoided licensing expensive LLMs, and used human-in-the-loop approaches to prevent hallucination while keeping services free and non-profit.

  10. Urgency Without Perfectionism: Multiple speakers emphasized that infrastructural constraints should not delay action. Continue building and deploying even with imperfect resources; perfect cannot be the enemy of good.

Notable Quotes or Statements

  • Kushbwa (Opening): "Technology is no more the ketchup that you get to enjoy with your baja but it's actually become the very oil without which buda is not even possible."

  • Kushbwa (Problem Statement): "Four in every five people living on this planet is actually today in the global south… our education system has been shaped by our colonial history… Many of us do not grow up or do not have the medium of instruction that is familiar to us."

  • Nidu (CK12): "I don't want to do Jensen's work for him [re: chip design], but one of our mission criteria was that we were not going to make education more expensive. We were going to make it accessible, equitable, and to the point for all students."

  • Nidu (On Capability): "Students in the US or in the west and students here are at the same things… some students are very capable… it's the same thing there. It just comes down to access and equity."

  • Nidu (On Pragmatism): "We didn't stop teaching kids because we didn't have textbooks. We didn't stop teaching kids because they didn't have pen or paper… We can't stop because we don't have the capacity. Let's continue so we can help everyone as much as we can."

  • Vana (On Infrastructure Reality): "By comparison, OpenAI had access to over a million [GPUs] last year and Africa alone has access to less than 10,000 of these. So as you can see there are huge gaps in access to infrastructure here."

  • Vana (On AI Literacy): "I don't think of AI literacy in terms of a curriculum. I think it's a ladder and where a child stands on that ladder depends entirely on what access they have already."

  • Aditya (Core Tension): "We're at this moment where AI as Kushbu was describing can be this incredible equalizer or it can perpetuate and widen all the divides that we have been trying to close for the last few decades in education."

Speakers & Organizations Mentioned

Primary Speakers

  • Kushbwa – Founder, Sika Lokom (organization working on AI-based learning companions for remote communities)
  • Nidu Kosla – Co-founder and Executive Director, CK-12; Board member, Noeva School (Bay Area), America India Foundation
  • Vana Sikka – Founder, Learn (creator of Vasi app); Board member, Code.org, Pratam, Women in Data Science; previously launched Infosys Foundation USA (2014)
  • Aditya – AI scientist and social entrepreneur (moderator); Co-host for session design

Referenced Organizations & Initiatives

  • CK-12 – Offers free courseware, curriculum, and AI-powered tutoring (Flexi) to hundreds of millions globally; 20 years old
  • Code.org – Teaching computer science and AI literacy globally
  • Pratam – Building offline tools (Parhai, Bal Balika) and WhatsApp chatbots for low-connectivity learning
  • Khan Academy – Personalized learning platform with offline capabilities
  • Vasi – Free app for creating and sharing wisdom nuggets using AI
  • Atma Shakti Thrust – CEO Ruchi; working with parents/communities in tribal districts of India
  • UNESCO – Dr. Shitan Shu Mishra, Chief of Digital Learning (session panelist)
  • RAA – Youth/learner-focused organization (CEO Niru mentioned as track lead)
  • India AI – Received 50,000 NVIDIA GPUs; working toward 200,000 total
  • OpenAI – Referenced for scale (1 million GPUs accessed)
  • NVIDIA – Primary GPU supplier (mentioned energy/infrastructure implications)

Government & Policy Bodies

  • India's government initiatives (GPU access, data center development)
  • UNESCO (global digital learning agenda)

Technical Concepts & Resources

AI/ML Systems & Techniques

  • Large Language Models (LLMs) – Referenced as resource-intensive but transformative; OpenAI models mentioned; discussion of frontier models and their dependence on GPU clusters
  • Knowledge Tracing – CK-12's core capability: granular tracking of student understanding at step-by-step level
  • Adaptive/Personalized Learning – Real-time adjustment of content difficulty and hints based on student performance
  • Generative AI Tutoring – Flexi (CK-12's AI tutor) providing 24/7 support; voice-enabled tutors for low-literacy learners
  • Offline-First Design – Intentional systems for low/no connectivity (WhatsApp chatbots, voice interfaces, cached content)
  • Knowledge State Assessment – Moving from binary right/wrong to diagnostic understanding of misconceptions (e.g., confusing nuclear fusion vs. chemical combustion)

Datasets & Training

  • Western-trained models – Issue of bias toward English, Western pedagogies, and datasets; emphasized need for local model training
  • Proprietary content vs. LLM licensing – CK-12 built own content to control quality and avoid hallucination; used human-in-the-loop verification
  • Curriculum flexibility – Systems designed to ingest and intelligently layer any external curriculum

Infrastructure & Energy

  • GPU Requirements – NVIDIA GPUs as bottleneck; India aiming for 250,000 (50k currently + 200k goal); OpenAI had 1M; Africa <10k
  • 3 Gigawatt Data Center – Being built in India to support AI infrastructure needs
  • Energy Costs – Acknowledged as massive and unequally distributed globally; opportunity for nuclear power expertise and localized talent

Platforms & Tools

  • CK-12 Flex Books – Free, modular, flexible textbooks adapted into platform
  • Vasi App – Free app enabling creation of AI-generated wisdom content
  • Pratam Tools (Parhai, Bal Balika) – Offline learning tools and WhatsApp integration
  • WhatsApp Chatbots – Low-barrier interface for AI-assisted learning in connectivity-constrained environments
  • Voice-Enabled AI – Addressing literacy barriers in low-literacy contexts

Pedagogical Frameworks

  • Learning Sciences – Integration of cognitive science, developmental psychology into platform design
  • Right-Level Content Delivery – Matching instruction to student's actual competency, not grade level
  • Foundational Literacy & Numeracy – Emphasized as non-negotiable prerequisites before AI literacy; no leapfrogging

Policy & Standards

  • AI Literacy Frameworks – Differentiated by geography, access, and prior knowledge; ladder-based rather than curriculum-based
  • Stakeholder Partnerships – Communities (objectives), governments (infrastructure/guardrails), educators (pedagogy), companies (tools/transparency)

Terminology

  • Demographic Dividend – Economic opportunity from large youth population; at risk due to education equity crises
  • Equity vs. Access – Distinction emphasized: access alone is insufficient; equity includes right-level, contextual, quality instruction
  • Leapfrogging – Skipping intermediate infrastructure stages to reach better outcomes; discussed as possible but dependent on intentional design

Structural Notes

The session was designed as a two-part format:

  1. Panel Discussion (covered above) – featuring three expert speakers on AI literacy, personalized learning, and infrastructure
  2. Collaborative Sprint (brief mention) – Audience divided into four personas (Parents/Communities, Policy Makers/Administrators, Teachers/Educators, Youth/Children) with designated track leads to discuss actionable implementation

The moderator emphasized moving from abstract "what could AI do" to concrete "what should AI do and what must it serve" in educational contexts.