India AI Impact Summit Insights
All sessions

Planet-Scale Intelligence: AI for Climate and Growth

Contents

Executive Summary

This AI summit panel discussion explores how AI can serve as critical infrastructure for solving large-scale problems in climate, agriculture, energy, and finance—particularly for India and emerging markets. Rather than celebrating AI hype, the panelists emphasize that scaling AI requires aligned systems encompassing talent, sustainable energy, institutional design, inclusive access, and trust. The central thesis: intelligent systems scale not because models are large, but because systems are aligned.

Key Takeaways

  1. AI scales as aligned systems, not large models. Success requires coordination across energy, talent, data governance, institutional incentives, trust mechanisms, and inclusive access—not just technical prowess. Fragmentation kills scale.

  2. Design for ground-level reality, not aspirational users. Voice interfaces, offline-first workflows, mother-tongue support, and pricing in rupees per minute (not dollars) are non-negotiable for India's 1B+ population. AI that ignores actual user contexts will remain boutique.

  3. Sovereign infrastructure isn't just geopolitical—it's economic. Owning the stack (energy, compute, models) is the only path to 100x cost reduction required for affordability. Cloud-dependent models can't reach the price points needed for equitable access.

  4. Change management and incentive design are the gating factors. Most AI pilots fail not because models are weak, but because trust isn't earned, incentives aren't aligned, and processes aren't redesigned. Spend as much time on people and institutions as on algorithms.

  5. Planet-scale problems (food security, climate, energy access) demand planet-scale intelligence infrastructure, not one-off projects. This requires treating AI as critical public digital infrastructure (like UPI for payments), designed for interoperability, openness, and shared access at population scale.

Key Topics Covered

  • AI as Critical Infrastructure — Moving beyond pilots and experiments to embedded, reliable operating systems
  • Cost Economics & Affordability — Making AI accessible to populations earning a few rupees per day
  • Energy & Computational Requirements — The often-overlooked infrastructure costs driving real deployment expenses
  • Trust & Adoption at Scale — How frontline workers, operators, and decision-makers actually adopt AI systems
  • Change Management & Incentive Design — The "boring plumbing" that determines pilot-to-scale transitions
  • Localization & Contextual Intelligence — Adapting AI to local languages, cultures, and problem domains
  • Sovereign AI Infrastructure — Building owned-in-country tech stacks to reduce cost and ensure alignment
  • Voice as Primary Interface — Recognizing that keyboard/text interfaces don't fit India's actual user base
  • Data Governance & Institutional Incentives — Aligning decision-makers, budgets, and procurement processes
  • Climate Resilience & Food Security — Planet-scale problems requiring planet-scale intelligence solutions
  • Business Case Quantification — Understanding opportunity costs of inaction, not just costs of action
  • Sectoral Applications — Deep dives into agriculture, manufacturing, energy, telecom, FMCG, and finance

Key Points & Insights

  1. Cost-Effectiveness is Non-Negotiable

    • Current AI tutoring costs ~10 rupees/minute; for affordability in India, must drop to 10–20 paise/minute (100x reduction)
    • This requires ownership of the entire stack: energy, data centers, GPUs, frameworks, models, and applications
    • "If we can't solve education and healthcare for the largest mass of humanity in India, AI scientists should be ashamed of ourselves"

  2. Infrastructure Must Be Owned, Not Rented

    • Relying on external energy and cloud infrastructure (paying for oil) makes scaling impossible
    • India's sovereign AI infrastructure strategy is essential for viability and cost control across the population scale

  3. Trust is Built Through Transparency & Alignment

    • A manufacturing case study (mayonnaise production): earning operator trust required 7 months of benchmarking AI recommendations against human operations
    • Frontline workers are skeptical until they see data proving reliability and personal incentive benefits

  4. Incentive Design Determines Adoption Behavior

    • When operators saw production bonuses tied to AI adoption, behavior naturally aligned
    • This human dimension is a "blind spot" most enterprises miss when scaling

  5. Process Redesign Must Precede AI Deployment

    • AI works only when business processes are fundamentally reimagined first, not retrofitted after
    • Problem statements should drive architecture backward (outcome → process → data → model)

  6. Voice is the Interface for India's Scale

    • Text-based interfaces impose typing friction on non-English speakers and limit mobile-first adoption
    • Voice-first design reflects actual human capability and is essential for mass adoption beyond top 30M urban users

  7. Institutional Incentives Block Adoption Despite Technology Readiness

    • Decision-making authority is fragmented across multiple ministries; no single empowered buyer
    • Procurement systems designed for traditional IT don't fit national-scale intelligence infrastructure
    • Opportunity cost of inaction (food insecurity costing billions) is poorly quantified vs. upfront implementation costs

  8. Multi-Sensory Data Integration Multiplies Intelligence

    • Refinery emissions monitoring: combining satellite data (dispersion at scale), drone data (precise localization), ground sensors, and legal/regulatory knowledge graphs enables risk prediction and compliance guarantees
    • Single data modalities miss systemic risks; integration across sensors, vision, structured data, and regulations is required

  9. Learnability > Initial Reliability

    • Like a child learning to walk, agents should be judged on learning trajectory, not first-attempt performance
    • Setting wrong expectations (e.g., expecting 95% accuracy on first deploy) kills adoption before systems mature

  10. Business Model Innovation Required for Population-Scale Viability

    • India's 1.4B population enables business models unviable in Rwanda (10M) or Kenya (40M)
    • Scaling innovations globally requires building local developer ecosystems, local data pipelines, and local institutional trust—not plug-and-play replication

Notable Quotes or Statements

  • Gorov Agarwal (Jio): "If we can't solve education, healthcare, longstanding problems for the largest mass of humanity which is in India, I think as AI scientists we should be ashamed of ourselves."

  • Gorov Agarwal: "Voice is the new typing in India. The interface of everything you do should be voice...until they become voice-first interfaces, there's no adoption. They were just fancy gadgets."

  • Derek Jose (Accenture): "Trust means reliability. If you have so if any of these dimensions are missing, it becomes part of the PC graveyard...Agents are like humans, they'll fail. Don't look at the first instance of a recommendation, look at the seventh."

  • Michael Zan (Dalberg Advisors): "The issue is often not sensing but sensemaking—the analytical capacity to make use of these insights."

  • Diva (Sataw/Opening remarks): "Intelligence does not scale because models are large. It scales because systems are aligned."

  • Rashmi Singh Sukmani (Sataw CTO, closing): "We have to go away from project thinking to intelligence infrastructure...from analytics to decision authority...and architect interoperable intelligence systems that span chips, compute, sensing, models, applications, energy, and institutions."

  • Gorov Agarwal (ITC): "AI is here to augment human capacity...It's not easy. We've realized that it only works when we take a step back, re-look at our processes, and really redesign them—not the other way around."

Speakers & Organizations Mentioned

Panelists:

  • Gorov Agarwal — Chief AI Scientist, Jio (Reliance); building gigawatt-scale AI-ready data centers
  • Arun Sharma — CEO, Rzonia; power transmission infrastructure for renewable integration
  • Derek Jose — Managing Director, Industrial AI, Accenture; operationalizing AI in asset-heavy industries
  • Michael Zan — Global Partner, Dalberg Advisors; technology policy and inclusive development in emerging markets
  • Gorov (ITC) — Leading digital initiatives in ITC's paperwood division; managing AI for supply chain, procurement, compliance
  • Pratip Basu — Moderator, CEO of Sataw
  • Diva (opening/closing remarks) — Sataw leadership
  • Rashmi Singh Sukmani — CTO, Sataw (closing remarks)

Organizations/Initiatives Referenced:

  • Reliance Jio (telecom, data centers)
  • ITC (FMCG, agriculture, paper, food)
  • Accenture (industrial AI services)
  • Dalberg Advisors (development economics, policy)
  • SAS (decision systems for agriculture, infrastructure, food security)
  • Marathon Oil (refinery emissions monitoring, Houston)
  • BPCL (Mumbai refinery)
  • Indian government (digital infrastructure initiatives: UPI, Aadhaar, digital public platforms)

External References:

  • EU Deforestation Regulation (EUDR)
  • Suez Canal disruptions (COVID supply chain case)

Technical Concepts & Resources

AI/ML Concepts:

  • Foundational models — Large-scale pre-trained models adaptable to multiple domains
  • Agentic AI workflows — Agent-based systems that reason and act autonomously (vs. simple inference)
  • Knowledge graphs — Structured representations of regulatory and domain knowledge (e.g., federal/state emissions laws)
  • Vision/multi-modal AI — Combining satellite imagery, drone video, ground sensor data, structured data
  • Learnability metrics — Measuring rate of improvement rather than absolute first-attempt performance

Infrastructure/Systems Concepts:

  • Digital Public Infrastructure (DPI) — Shared, open, interoperable systems (India's UPI model as reference)
  • Sovereign AI infrastructure — National ownership of compute, energy, models to avoid external dependencies
  • Data pipelines — Localized data collection, labeling, and processing workflows
  • Platform architecture — Event-driven, microservices-based systems embedding AI across workflows

Domains & Use Cases:

  • Precision agriculture — Drone/satellite imagery, climate modeling, crop yield prediction
  • Supply chain risk — Procurement disruption forecasting, ESG compliance (EUDR), logistics optimization
  • Industrial emissions monitoring — Multi-sensor (satellite, drone, ground) methane detection and source localization
  • Manufacturing quality control — Computer vision for defect detection; agent recommendations for process optimization
  • Energy systems — Renewable integration, grid balancing, demand forecasting
  • Financial risk — Credit scoring, climate/weather risk exposure, regulatory compliance

Regulatory/Governance Concepts:

  • EU Deforestation Regulation (EUDR) — Plot-level traceability requirements for procurement
  • Emissions compliance — Federal vs. state-level regulatory frameworks (US refinery example)
  • Data governance — Institutional incentive alignment, procurement fit-for-purpose design

Not Explicitly Named but Implied:

  • Large language models (LLMs) for localization
  • Computer vision (CV) for drone/satellite analysis
  • Time-series forecasting for climate/risk prediction
  • Multi-agent systems (implied in refinery example)

Prepared: Based on transcript from "Planet-Scale Intelligence: AI for Climate and Growth" (YouTube link provided)