India AI Impact Summit Insights
All sessions

The AI-DPI Nexus: The Future of Public Interest Technology

Contents

Executive Summary

This AI impact summit panel discussion explores the critical intersection between Digital Public Infrastructure (DPI) and Artificial Intelligence (AI), arguing that DPI must serve as the foundational layer enabling inclusive, safe, and effective AI deployment at scale. Speakers from the Gates Foundation, UNDP, World Bank, Asian Development Bank, and leading tech innovators emphasize that countries pursuing AI without established DPI risk exclusion, privacy violations, and ineffective solutions—while those layering AI strategically atop robust DPI can achieve population-level impact and economic transformation.

Key Takeaways

  1. DPI is not a prerequisite for AI—it's a multiplier for AI impact. Countries need not complete DPI before starting AI work, but deliberate coordination between the two is essential. The best outcomes come when AI capabilities are intentionally layered atop DPI foundations, not built in parallel silos.

  2. Inclusion and safety are interdependent, not trade-offs. Designing for underserved populations (local languages, voice interfaces, offline access) and building in consent/data protection mechanisms from the start increases both the reach and the trustworthiness of AI systems.

  3. Institutional human accountability must be wired into AI systems. When AI makes or influences decisions affecting citizens, there must always be a clear human/institutional owner responsible for those decisions. Liability and oversight cannot be automated away.

  4. Data governance is the linchpin connecting DPI and AI. Without interoperable, consented, clean data ecosystems with clear ownership and revocable consent, both DPI and AI fail to serve public interest. Data governance is not a compliance issue—it's a foundational design principle.

  5. Open standards and modularity in the AI layer will determine whether benefits compound or concentrate. Closed, proprietary AI systems built on top of public DPI create new forms of exclusion and gatekeeping. Public standards for AI interoperability, auditability, and data exchange are as important as they were for DPI.

Summit Talk Summary

Key Topics Covered

  • DPI as foundational infrastructure for AI: Digital identity systems, data sharing platforms, and payment systems as prerequisites for inclusive AI
  • The adoption gap: Misalignment between grassroots DPI pilots and government-level AI ambitions; ministry silos preventing coordinated strategy
  • Inclusion as a foundational principle: Designing for underserved populations, women, youth, and digitally disadvantaged communities from the outset—not as an afterthought
  • Governance and liability frameworks: Who bears responsibility when AI systems fail; the necessity of human institutional ownership over automated decision-making
  • Consent and trust mechanisms: Voice-based, multilingual consent frameworks; data protection integrated with innovation, not as blockers
  • Interoperability and open standards: Applying DPI principles (modularity, openness, competition) to AI layer to prevent walled gardens
  • Data governance as critical enabler: Clean, interoperable data ecosystems as prerequisites for effective AI; risk of AI penalizing data-poor nations
  • Skills and capacity gaps: Shortage of technical expertise in governments to design and implement both DPI and AI systems
  • Sovereignty and agency: Countries' ability to make deliberate choices about their digital architecture; avoiding technological lock-in
  • Regional case studies: Examples from Rwanda, Nigeria, India, Singapore, and other countries implementing coordinated DPI-AI strategies

Key Points & Insights

  1. The sequencing paradox: Governments face pressure to move quickly on AI while DPI is still incomplete. However, premature AI adoption without DPI foundations leads to fragmentation, exclusion, and unaccountable systems. The solution is not strictly sequential but rather coordinated and intentional choice-making.

  2. DPI enables, AI personalizes: DPI provides standardized, interoperable rails and registries; AI adds contextualized, personalized service delivery. Together they create a "flywheel" of impact (e.g., SingPass in Singapore: 41 million transactions/month).

  3. Data governance is the missing third leg: While DPI and AI get attention, data governance, privacy protection, and consent mechanisms are often overlooked. Without these, AI systems risk becoming predatory (screen scraping, unauthorized data collection) or untrusted by end users.

  4. Inclusion requires deliberate design, not retrofitting: Rwanda's model demonstrates that inclusion must be a foundational principle, not secondary. This means designing for local languages, dialects, offline users, and people with disabilities from the start—not adding accessibility later.

  5. Liability requires institutional human ownership: AI should accelerate decision-making but never replace it. Institutions and humans must own decisions; when things go wrong, there must be a clear human/institutional entity accountable—not the algorithm.

  6. Skills gap is acute: Many governments lack the technical capacity to understand data architectures, design DPI, or govern AI. Ministries of ICT may understand DPI, but other ministries pursuing AI independently lack context. This creates silos and suboptimal outcomes.

  7. Consent must be revocable and democratic: Voice-based consent mechanisms in local languages (not just written, smartphone-dependent consent) enable digitally excluded populations to maintain agency. Consent should be revocable, not a one-time surrender of data rights.

  8. Open standards and modularity drive innovation: When governments establish public rails (DPI principles applied to AI—APIs, standardized inputs/outputs, auditability logs), private sector innovation thrives. Walled-garden, proprietary AI solutions stifle competition and drive up costs.

  9. Trust and human capital are foundational: Technical infrastructure alone fails without user understanding of benefits, trust in data protection, and awareness of how AI will serve them. Education and trust-building must accompany infrastructure rollout.

  10. Agentic AI poses new governance risks: Large Language Models without consent layers and verifiable identity will engage in screen scraping and unauthorized data collection. DPI's consent and authentication mechanisms must be extended to AI layer to prevent this.

Notable Quotes or Statements

"Talking about AI without DPI is like thinking about how you decorate your penthouse when you haven't even built the first floor yet of a building."
— Nandan Nilani (quoted by moderator K)

"DPI makes it possible; AI makes it personal."
— Dr. Bio, Data Science Nigeria

"AI is a standardized infrastructure, but AI gives to us personalized experience, contextualized knowledge that can pick other data points on that individual on DPI and make it work for that person."
— Dr. Bio

"The more you are delivering to [a public] purpose, the liability risks drop. We need institutional ownership behind technology. When it comes to liability, that's where you go—to the people with the institutional ownership."
— Olivier Twio, C4IR Rwanda / AI Scaling Hub Rwanda

"AI is replacing engineers. AI is not replacing SWIFT. AI is not replacing passports. AI is not replacing digital locker in India."
— Kamia Chandra, Center for Digital Public Infrastructure

"If you create any solution for public interest but you do not have this part of including the digitally disadvantaged right from the beginning, you've gone the wrong way, and it's going to cost you at the end."
— Olivier Twio

"There is no AI, there is no DPI, if there is no trust and if there is not human capital that really understands the benefit of that."
— Antonio Zabia, Asian Development Bank

"The arc of AI's beneficial use for development is not defined by how quickly technology is moving, but by how it's getting adopted, who benefits, and how it is governed."
— Quazam Masali, UNDP AI Hub for Sustainable Development

Speakers & Organizations Mentioned

Panel 1 (DPI-AI Nexus)

  • K (Moderator) — DIAL (Digital Impact Alliance)
  • Dr. Sadena Indai — Academic advisor to the Government of Senegal; UN High-Level Advisory Body for AI; Global Partnership on Artificial Intelligence
  • Quazam Masali — Director, UNDP AI Hub for Sustainable Development; formerly focused on DPI acceleration
  • Howard Lunga — Gates Foundation; Digital Public Infrastructure practice & AI strategy
  • Vaganti Desai — World Bank; formerly led ID4D initiative; now Director, Digital Programs for South Asia
  • Kamia Chandra — Co-founder, Center for Digital Public Infrastructure (CDPI)

Panel 2 (Frontline Implementation)

  • Antonio Zabia — Director, Digital Sector Office, Asian Development Bank
  • Dr. Bio — Founder & CEO, Data Science Nigeria
  • Olivier Twio — Center for Fourth Industrial Revolution (C4IR) Rwanda; Director, AI Scaling Hub Rwanda

Organizations & Initiatives Referenced

  • Gates Foundation — Major funder of DPI and AI initiatives globally
  • UNDP (United Nations Development Programme) — AI Hub for Sustainable Development; global DPI programs
  • World Bank — ID4D (Identification for Development) initiative; Digital Public Infrastructure practice in South Asia
  • Asian Development Bank (ADB) — Center for AI around Asia-Pacific
  • DIAL (Digital Impact Alliance) — Research on data architectures and policy choices for African policymakers
  • Center for Digital Public Infrastructure (CDPI) — Research and strategy on DPI design
  • C4IR Rwanda (Center for Fourth Industrial Revolution) — Innovation hub for digital transformation in Rwanda
  • Data Science Nigeria — Private firm innovating on DPI-AI integration in Nigeria
  • Government of Senegal — Case study for DPI-AI coordination challenges
  • Government of Rwanda — Case study for inclusive AI policy design
  • Government of India — Host; referenced for Aadhaar, digital locker, language diversity

Technology/Infrastructure Projects Referenced

  • Aadhaar (India) — Digital identity system; case study in system improvement and risk management over time
  • SingPass (Singapore) — Digital identity connected to 41M government service transactions/month
  • NIBS (National Institute for Digital Payments) (Nigeria) — Payment infrastructure enabling fintech innovation
  • Digio Locker (India) — Digital document storage system
  • Open networks — Emerging AI layer above DPI (referenced with agriculture as use case)
  • Model Context Protocol (MCP) / Web MCP — Interoperability protocols released by tech sector (two weeks prior to talk)

Technical Concepts & Resources

Core Concepts

  • Digital Public Infrastructure (DPI): Foundational digital systems (identity, payments, data exchange) designed for interoperability, openness, and public benefit
  • Interoperability: Ability of systems to work seamlessly together; key to both DPI and scalable AI
  • Data exhaust: Data trails generated by digital transactions; valuable for AI training and personalization but requires governance
  • Consent layers: Mechanisms allowing users to grant, revoke, or manage access to their data; increasingly voice-based in inclusive models
  • Agentic AI: AI systems capable of autonomous action; poses governance risks without proper consent and identity verification
  • Modularity: System design allowing components to be swapped, upgraded, or replaced without rebuilding entire infrastructure
  • Screen scraping: Unauthorized automated data collection; cited as a risk when AI lacks proper consent mechanisms
  • Data governance: Policies and practices ensuring data quality, privacy, consent, security, and appropriate use

Frameworks & Approaches

  • DPI thinking applied to AI layer: Using principles of openness, modularity, interoperability, auditability, and standardized outputs to prevent AI lock-in
  • Voice-based consent in local languages/dialects: Enabling consent for non-literate or offline populations via phone-based systems
  • Inclusive design from outset: Designing for women, youth, people with disabilities, digitally disadvantaged—not retrofitting accessibility
  • Impact measurement beyond technical performance: Measuring inclusion outcomes, not just algorithm accuracy
  • Continuous improvement governance: Recognizing that systems will need iterative refinement; building in mechanisms for grievance redressal, risk identification, and updates

Data & Measurement

  • Data interoperability: Clean, linked datasets (farmer registries + land registries + weather data + credit history) enabling downstream innovation
  • Standardized logs and auditability: Minimal logging standards to enable oversight without stifling innovation
  • Impact metrics for underserved populations: Measuring whether AI solutions actually benefit those they target; not assuming inclusive design = inclusive outcomes

Policy/Governance Concepts

  • Sovereignty as agency: Countries' ability to make deliberate choices about digital architecture; not just technical independence
  • Public-private collaboration model: Government provides open rails (DPI/AI standards); private sector innovates atop them
  • Liability and institutional ownership: Clear human/institutional accountability for AI decisions; distinguishing public rails from private innovation
  • Data sharing policies: Formal frameworks governing how data can be combined, accessed, and used across systems

Tools/Systems Mentioned

  • QR code-based feedback systems: Research collection tools for policy maker feedback (referenced in closing)
  • Voice interfaces and IVR (Interactive Voice Response): Technology enabling phone-based consent and data collection in local languages
  • Linguist-labeled voice data: Using human linguists to label dialectical speech for training multilingual AI models
  • Feature phones (non-smartphones): Recognized as primary technology for digitally excluded populations; leveraged for consent and data collection

Context & Significance

This talk occurs at an AI Impact Summit in India, emphasizing:

  • The global urgency of moving quickly on AI while avoiding the mistakes of DPI deployment
  • Particular relevance for lower-income countries and underserved populations in Asia, Africa, and the Global South
  • Emerging consensus that AI governance and DPI governance must be coordinated, not siloed
  • Recognition that technology alone fails without trust, consent, skills, and human accountability
  • Shift from viewing DPI as "foundational infrastructure" to viewing it as a multiplier of AI impact when properly integrated

The talk challenges both the "move fast and break things" tech mindset and the "build DPI first, then AI" sequentialist approach, proposing instead a deliberate, coordinated, inclusive, and governable approach to digital transformation.