Multistakeholder Partnerships for Thriving AI Ecosystems

Contents

Executive Summary

This panel discussion explores how AI can advance sustainable development while addressing the equity gap threatening to widen global inequality. The speakers—representing government, private sector, NGOs, and research institutions—argue that responsible AI deployment requires coordinated multistakeholder action across governance, infrastructure, skills development, and localized problem-solving. The Hamburg Declaration on AI for Sustainable Development is presented as a framework for translating high-level principles into measurable, concrete commitments.

Key Takeaways

Multistakeholder partnerships are not optional—they are structurally necessary: No single entity (government, private sector, academic, or NGO) can solve AI equity and responsible deployment alone. Integration must occur at the inception of projects, not post-hoc.
Concrete, measurable commitments backed by accountability trump aspirational declarations: The Hamburg Declaration's strength comes from binding signatories to time-bound, quantifiable actions (training numbers, datasets released, infrastructure investments), enabling tracking of real progress.
The constraint is not innovation but infrastructure and policy: Startups, researchers, and ideas exist globally; the bottleneck is access to compute, data, capital, and enabling governance frameworks. Governments must lead here through DPI investment while private sector scales skills and implementation.
Technology adoption accelerates when it visibly improves daily life: India's financial inclusion success shows that when citizens see direct, tangible benefits (subsidies reaching them directly, new credit access), adoption happens organically. The role of policy and private sector is to build that infrastructure and capability, not to convince users.
Context matters more than cutting-edge models: In resource-constrained settings, domain-specific traditional ML, small language models, and hardware optimization strategies often outperform large language models. AI tool selection must be tailored to local infrastructure realities, not driven by global trends.

Key Topics Covered

AI equity gap and the "power gap": Unequal distribution of venture capital, compute infrastructure, and data resources between Global North and Global South
Government's role: Creating enabling frameworks, governance structures, and digital public infrastructure (DPI)
Private sector responsibility: Democratizing technology through skills training, ethical deployment, and inclusive product development
Multistakeholder partnerships: Operational models that embed government ownership, technical partners, implementation support, and evaluation mechanisms from day one
Concrete AI applications: Healthcare (tuberculosis screening), education (oral reading fluency), agriculture, and financial inclusion
Data infrastructure: Challenges of data fragmentation, contextualization, and the need for high-quality sensing infrastructure in Global South contexts
The Hamburg Declaration: A commitment-based framework requiring measurable, time-bound actions rather than symbolic endorsements
AI assurance and evaluation: Need for regional evaluation hubs and shared playbooks
Compute resources and energy: Hardware constraints and the potential for repurposing legacy systems and quantum computing innovations
LLMs vs. traditional ML: Context-dependent tool selection for resource-constrained environments

Key Points & Insights

The "Power Gap" Not Innovation Gap: Dr. Barbara Coffler (Germany's Parliamentary State Secretary) argues the barrier is not lack of innovation but unequal access to resources—only 17% of venture capital flows to regions representing 90% of the global population; data center capacity in the Global South represents only 0.1% of global resources.
Government Must Lead with Ownership: Successful AI deployments require government to own the problem and lead the initiative. Technology partners and NGOs serve as facilitators, but institutional embedding and policy ownership are non-negotiable for sustained impact.
Multistakeholder Ecosystems Must Be Built From Day One, Not After: Nakul Jane (Wadwani AI Global) emphasizes that institutionalization mechanisms, workflow integration, and monitoring capacity cannot be afterthoughts. Successful projects (e.g., Gujarat education initiative) involved government, technical partners, field-level NGOs, and capacity-building support simultaneously.
Technology Democratization Drives Adoption: Arundhati Bhattacharya (Salesforce South Asia) demonstrates this through India's financial inclusion program: when technology infrastructure (Aadhaar, UPI, mobile networks) enabled direct government subsidy delivery, adoption accelerated organically—from 13% to 100% of intended recipients benefiting. Adoption is not the constraint; infrastructure and policy framework are.
Data Contextualization Is Critical: Models trained on Western datasets are not readily applicable to Indian or African contexts. The sensing infrastructure itself—not just data pooling—must be built at regional scale, positioned as part of digital public infrastructure (e.g., air pollution monitoring networks).
Skills and Mindset Shift Required: Dr. Coffler calls for moving from "users" to "creators" of AI, requiring investment in vocational training, university curricula, and engagement of SMEs. Salesforce's 3.9 million "Trailblazers" in India exemplify scaling through community-driven skill building.
Measurement Matters More Than Statements: The Hamburg Declaration's power lies in binding signatories to measurable commitments (e.g., Germany trained 190,000 people vs. target of 160,000; released 55 AI datasets vs. target of 30). Annual accountability prevents "declaration theater."
Traditional ML and Small Language Models Remain Essential: In resource-constrained, low-connectivity settings, LLMs are often undeployable. Context-specific traditional ML and small language models better serve ground-level users; AI tool selection must be infrastructure-aware.
Trustworthy AI Requires Multi-Layered Governance: TCS's "trusty platform" and work on responsible AI frameworks indicate the need for: data quality assurance, algorithmic transparency, auditability, observability, regulatory compliance, and greening (energy efficiency).
Missing Infrastructure: Global Solution Marketplace & Regional Evaluation Hubs: Wadwani AI identifies two gaps—a marketplace/repository where startups can deploy solutions across geographies with shared playbooks, and regional evaluation hubs to localize assessment criteria rather than imposing one-size-fits-all metrics.

Notable Quotes or Statements

"It's not an innovation gap, it's a power gap. Innovative people exist around the globe. Ideas are created everywhere in society. But venture capital—only 17% reaches regions representing 90% of the world's people." — Dr. Barbara Coffler, Parliamentary State Secretary, Germany's Federal Ministry for Economic Cooperation and Development

"If improvement in technology is not democratized, it doesn't have an impact. You need to democratize technology." — Arundhati Bhattacharya, Chairperson & CEO, Salesforce South Asia

"Building technology is the easiest part. Everything around it—institutional mechanisms, workflow integration, monitoring—that's the hard part and cannot be an afterthought." — Nakul Jane, CEO, Wadwani AI Global

"Unless you self-regulate, regulators will come down with a heavy hand. If you want to keep innovating, self-regulation is essential." — Arundhati Bhattacharya

"Don't look at market cap while creating your company. Do the right things. Market cap will follow. Your work should be driven by satisfaction and helping people improve their standard of living, not valuation numbers." — Arundhati Bhattacharya, advice to young entrepreneurs

"LLM is worldwide advancement, but we need something industry-wise, company-wise, context-wise, task-wise. We are far from that." — Dr. Sachin Looda, Chief Scientist, TCS

"We're not signing something we don't want to do and then applauding ourselves a year later. We come up with concrete steps—and that's what accountability looks like." — Dr. Barbara Coffler, on the Hamburg Declaration

Speakers & Organizations Mentioned

Panelists:

Dr. Barbara Coffler – Parliamentary State Secretary, Germany's Federal Ministry for Economic Cooperation and Development
Arundhati Bhattacharya – Chairperson & CEO, Salesforce South Asia; former Chairperson, State Bank of India
Nakul Jane – CEO & Managing Director, Wadwani AI Global
Dr. Sachin Looda – Chief Scientist & Head of Research, Tata Consultancy Services (TCS)
Robert (moderator) – UN Development Program (UNDP)

Organizations/Initiatives Referenced:

United Nations Development Program (UNDP)
Hamburg Sustainability Conference (hosted annually in Hamburg, Germany)
German Government (BMZ – Federal Ministry for Economic Cooperation and Development)
Salesforce
State Bank of India (SBI)
Wadwani AI Global
Tata Consultancy Services (TCS)
Tata Group (170-year-old conglomerate)
Indian Government (Principal Scientific Advisor's office; Ministry of IT)
IIT Kanpur (Indian Institute of Technology)
Government of Gujarat (India)
ICMR (Indian Council of Medical Research)
NASCOM (IT industry association, India)
AICTE (All India Council for Technical Education)
IBM
Carnegie Mellon University
Various startups and community nonprofits across India and Global South

Technical Concepts & Resources

AI/ML Approaches & Tools:

Large Language Models (LLMs): Discussed as unsuitable for low-resource settings; limited deployment viability in Global South contexts due to infrastructure constraints
Small Language Models: Emerging alternative for context-specific, resource-constrained deployments
Traditional Machine Learning: Demonstrated sustained efficacy in healthcare (tuberculosis screening), education, agriculture applications where LLMs cannot deploy
Agentic AI: Mentioned as emerging capability for coordinating multiple agents and integrating enterprise systems
Trusty Platform (TCS): Technology platform for evaluating, calibrating, and assisting AI engineers in building more responsible AI systems
Quantum Computing: Quantum Valley initiative (TCS, IBM, Anra government) as potential solution for compute constraints

Data & Infrastructure Concepts:

Digital Public Infrastructure (DPI): Government-built foundational systems (e.g., Aadhaar biometric ID, UPI payments, mobile networks) enabling AI deployment
Sensing Infrastructure: High-quality, high-volume, high-velocity data collection networks (e.g., air pollution monitoring) as prerequisite for ML; positioned as public infrastructure
Data Fragmentation & Silos: Challenge across enterprise and regional scales requiring governance solutions
Open Data Ecosystems: Referenced as necessary but insufficient without underlying sensing infrastructure
Responsible AI Frameworks: Mentioned as core research area, with principles around data quality, algorithmic transparency, auditability, observability, and energy efficiency

Use Cases & Deployments:

Education: Oral reading fluency assessment (Gujarat); addressing literacy gaps
Healthcare: Tuberculosis (TB) screening; cervical cancer diagnosis; disease detection in remote areas
Agriculture: Satellite data analysis for farmer advisory (Kenya)
Financial Inclusion: Aadhaar-UPI ecosystem enabling 400+ million account openings; direct subsidy delivery; microfinance access through digital cash flow records
Sustainability: Climate action; air pollution monitoring; disease prediction

Programs & Frameworks:

PM Jandhan Yojha: India's financial inclusion program
Hamburg Declaration on AI for Sustainable Development: Commitment-based framework with measurable, time-bound deliverables (e.g., training targets, dataset releases, infrastructure projects)
AICOE (AI Center of Excellence): Government-supported research collaborations (e.g., TCS-IIT Kanpur on sustainability)
Salesforce 1-1-1 Model: Corporate responsibility framework (1% of profits, 1% of products, 1% of employee time to nonprofits; 2% in India per law)
Bashini: Indian government initiative on multilingual AI/language datasets

Institutional Commitments Made (Hamburg Declaration Signatories):

Training: 190,000 people trained (target: 160,000)
AI Building Blocks: 15 released for climate action (target: 12)
Open Data Datasets: 55 released as digital public goods (target: 30)

Methodological Notes

Discussion Format: Panel-based multistakeholder dialogue with opening statements, guided questions, and Q&A
Evidence Base: Primarily experiential and case-study driven (e.g., India's financial inclusion journey, Gujarat education AI, TB screening partnerships)
Time Constraints: Panel was truncated due to time limits; some technical questions (LLM vs. traditional ML, SaaS viability) answered briefly
Implicit Frameworks: Discussion assumes integrated ecosystem design, participatory governance, and outcome-based accountability as foundational principles for responsible AI deployment