AI, Agriculture, and DPI: Unlocking Economic Growth

Contents

Executive Summary

This panel discussion explores how Digital Public Infrastructure (DPI) combined with AI can transform agricultural advisory services for smallholder farmers in the Global South. The conversation emphasizes that DPI—foundational systems like digital identity, payments, and data exchange—must serve as a horizontal layer enabling inclusive AI solutions, while maintaining trust, governance, and farmer-centered design as critical success factors.

Key Takeaways

Build horizontal before vertical: Governments must invest in foundational DPI (identity, registries, data sharing, payments) before deploying AI advisory solutions. This is the prerequisite for inclusive, scalable, interoperable systems.
Trust takes time and co-design: AI advisory systems fail without farmer feedback loops and cultural alignment. Design with farmers, communicate uncertainty, and build governance that proves reliability over years, not months.
Fragmentation is the enemy: Multiple siloed systems cost more, deliver less, and resist integration. Unified DPI + multiple solution layers (public and private) outperforms each player building end-to-end platforms.
Language and channels matter as much as algorithms: Localization (voice, SMS, app, extension worker) and cultural alignment are as critical as model performance. Inclusion is a design principle, not a feature add-on.
Partnerships, not technology, are the bottleneck: Long-term impact depends on sustained government ownership, multi-stakeholder coalitions, continuous data pipelines, and evidence-based benchmarking—not on selecting the "best" AI model.

Key Topics Covered

Digital Public Infrastructure (DPI) as agricultural foundation: DPI's role in enabling scalable, interoperable AI advisory systems
India's AgriStack and Vistar: Government-backed DPI initiatives and recent 2024 budget allocations
Ethiopia's FIDA system: Practical example of DPI-powered AI in a different geographic and institutional context
Business models and sustainability: Public-private partnerships, cost structures, and avoiding fragmented systems
Benchmarking and evaluation frameworks: AgriAgBench, performance metrics beyond adoption
Inclusion and last-mile delivery: Language localization, multi-channel approaches, farmer feedback loops
Trust as critical infrastructure: How governance, transparency, and alignment with farmer beliefs underpin adoption
AI weather forecasting: Concrete use case reaching 38 million farmers through SMS in India
Regional collaboration: UAE's AI ecosystem for global agricultural development; South-South partnerships
Scaling pathways: Moving from pilots to national/regional implementation

Key Points & Insights

DPI is horizontal infrastructure, not vertical solution
- DPI creates the foundational layer (identity, registries, data exchange, payments) that enables multiple AI advisory solutions to build on top
- This is fundamentally different from end-to-end proprietary platforms that lock governments and farmers into single providers
- Without this horizontal layer, personalization and context-specific advice at scale is not possible
Fragmentation is extremely costly
- Nigeria's 13 independent ID systems exemplifies failure; India's Aadhaar and Ethiopia's FIDA show success of unified approaches
- Each siloed system increases costs and reduces interoperability
- Countries must resist the temptation to build separate systems for each government department or use case
Institutional mandate and political ownership are prerequisites
- India's success with DPI/AgriStack/Vistar took 15+ years of sustained championing across government stakeholders
- Budget allocation (2024 Indian budget) validates long-term commitment but doesn't create impact alone
- Continuous governance, accountability structures, and resource allocation for operations are essential; programs wither without ongoing support
Trust is built through farmer co-design, not top-down deployment
- Example: AI weather forecasts failed with some farmers because they conflicted with traditional religious calendars for rainfall prediction
- Inclusion requires designing with farmers, not for farmers; feedback loops are ongoing, not one-time
- Technical accuracy alone is insufficient—cultural alignment, actionable advice, and transparent uncertainty communication are required
Language and multi-channel access are not afterthoughts
- Maharashtra's achievement: onboarding first data collection for a regional language (Bili) to AI model deployment in 4-5 weeks
- Solutions must span SMS, voice, app, and extension worker channels to serve diverse connectivity/literacy contexts
- Language work supercharges inclusive access; AI systems can reconfigure themselves for local tongues at population scale
Data pipelines and content pipelines require continuous investment
- As research evolves and data quality improves, systems must be resourced to feed new information into advisory layers
- Benchmarking is not a one-time effort but continuous evolution as user needs and available tools change
- Evidence-based evaluation prevents random scaling of ineffective interventions
Public goods development requires multi-stakeholder coalition-building
- CGIAR AI Hub + Gates Foundation + Digital Green + Bayer model: combining agriculture research, funders, implementation partners, and private sector
- No single actor can build, deploy, and sustain agricultural AI systems at national/regional scale alone
- Partnerships must include governments (legitimacy), agriculture experts (validity), and technology providers (capability)
Reusable blueprints and open-source platforms reduce barriers
- MOSIP combines best practices from global DPI deployments into open-source platforms that countries customize rather than build from scratch
- Vistar (India), Mahavistar (Maharashtra), Bihar Krishi AI, and Ethiopia's approaches all build on shared architectural principles
- This reduces duplication and accelerates learning across borders
Governance cannot be an afterthought
- Trust in data, institutions, models, and governance is shaped by transparency, consistency, and accountability—not just technical performance
- Data security, responsible AI practices, and clear accountability for errors must be embedded in design, not retrofitted
- Public goods require public governance; profit-extraction models undermine adoption
Scale requires resources to decrease per-unit costs
- Krishi Samriddhi in Odisha: reaching 7 million farmers at 18 cents per farmer per year
- AI weather forecasting reached 38 million farmers in India; goal to reach 100 million by 2030
- Scale is feasible when DPI enables multiple solutions to reuse the same identity, data, and payment rails

Notable Quotes or Statements

"DPI is essential for AI to be inclusive because AI systems can only work if they have information about that particular user or that individual." — Sanjay Jen, Gates Foundation

"We need to enable solutions for the farmer, not solve it for the farmer." — Sanjay Jen

"It didn't happen overnight. I think it finally made it to the budget speech this year, but it's been many years in the making." — Sanjay Jen, on India's 15+ year DPI journey

"When I speak to a farmer and they give me this feedback that is even more important than the technology and the solution itself because if they don't trust it or they don't trust you or the information then what's the point." — Fatima Al-Mulla, UAE Presidential Court

"Think of DPI in a simple way as data on rails. If you combine that with accountable and responsible AI, suddenly you are unlocking a lot of possibilities for population scale reforms." — Jagdish Babu, Xstep Foundation

"You can build the most perfect platform, the most perfect technology, [but] trust takes time." — Nisha Shetti, PXD

"Public goods reach the people who need it the most. We are at a very critical time as AI is coming and growing very quickly." — Fatima Al-Mulla

Speakers & Organizations Mentioned

Speaker	Role/Organization
Nita Bassin	CEO, Digital Green (Moderator)
Sanjay Jen	Director of Digital Public Infrastructure, Gates Foundation
Fatima Al-Mulla	UAE Presidential Court
Nisha Shetti	CEO, PXD (Plantix Digital)
Jagdish Babu (JB)	CEO, Xstep Foundation

Key Organizations Referenced:

Gates Foundation
Digital Green
Ministry of Agriculture (India)
Government of India
CGIAR (Consultative Group for International Agricultural Research)
World Bank
Bayer
Embr (mentioned in partnership context)
Amul (dairy, livestock DPI example)

Government Initiatives Mentioned:

India: Aadhaar (digital identity), AgriStack, Vistar, Krishi Samriddhi (Odisha)
Ethiopia: FIDA (digital ID system), PhD Pass (e-payments app)
Maharashtra: Mahavistar, Bili language AI pilot
Bihar: Bihar Krishi AI
UAE: Abu Dhabi AI Ecosystem for Global Agricultural Development

Technical Concepts & Resources

Concept	Description
DPI (Digital Public Infrastructure)	Foundational reusable digital building blocks (identity, payments, data exchange, registries) designed for public benefit and interoperability
AgriStack	India's national architecture for farmer registries, crop registries, and plot registries
Vistar	India's AI-powered advisory platform synthesizing content from multiple sources; accessible via multiple channels
Mahavistar	Maharashtra state's adaptation of Vistar framework
Bihar Krishi AI	Bihar's AI advisory system running parallel to Bharat Vistar
MOSIP	Modular Open Source Identity Platform; open-source DPI reference implementation
FIDA	Ethiopia's digital ID system enabling land getagging and AI advisory tools
CGIAR AI Hub	International initiative translating AI and data into agricultural public goods
Institute for Agriculture and AI	UAE-based institute (part of Muhammad bin Zaid University of AI) for AI product development in agriculture
AIM for Scale	Co-funded UAE-Gates initiative to scale scientifically proven agricultural technologies
AgriAgBench	Benchmarking framework for evaluating agricultural AI systems (mentioned but not fully endorsed by panelists)
Agriculture Large Language Model (LLM)	First LLM fine-tuned for agriculture and released as public good; partnership of Digital Green, CGIAR, World Bank, Gates Foundation, Bayer, Embr
AI Weather Forecasting	Use case: SMS-based AI weather alerts reaching 38 million Indian farmers for monsoon preparation; goal: 100 million by 2030
Bili Language AI	Regional language AI model for Maharashtra onboarded in 4-5 weeks; voice-based advisory
Amul DPI	Livestock identity system integrated with AI (health, medication history, advisor context)

Key Methodologies:

Farmer co-design: Solutions designed with farmers, not for them; continuous feedback loops
Multi-channel delivery: SMS, voice, app, extension worker outreach tailored to connectivity and literacy
Interoperability-first architecture: Systems designed to switch technologies/models as benchmarks improve
Evidence-based scaling: Benchmarking and evaluation inform scale-up, not adoption metrics alone
Inclusive governance by design: Transparency, accountability, cultural alignment baked in from inception

Cost Metrics Referenced:

Krishi Samriddhi (Odisha): 18 cents per farmer per year for 7 million farmers
AI weather forecasting: scaled to 38 million farmers; target 100 million by 2030

Context & Significance

This talk was delivered at an AI Summit in India and reflects a critical moment in Global South agricultural development:

Extreme poverty reduction has plateaued despite past progress, with development aid being cut ($30B annually)
60% of India's population depends on agriculture; smallholder farmers represent billions globally
AI + DPI convergence offers a pathway to reach farmers at scale with personalized, context-aware advisory—but only if governance, trust, and inclusion precede or accompany technology deployment
Regional leadership (India, UAE, Ethiopia) is establishing models and partnerships that other countries in the Global South can adapt
The conversation underscores that infrastructure and governance are prerequisites for inclusive AI impact, not afterthoughts