This Week in Global Development: “Global Progress in the AI Era: What happens when AI labs and social enterprises build together?”
Date: March 17, 2026
Host: Katherine Chaney (Senior Editor for Special Coverage at Devex)
Guests:
- Manu Chopra (Co-founder & CEO, Karya)
- Alex Nawar (Head, OpenAI Academy; former GiveDirectly)
- Hans Zheng Chia (Director, AI for Global Development Initiative, Center for Global Development)
Episode Overview
This special edition explores the rapidly evolving relationship between frontier AI labs and social enterprises, particularly in low-resource and Global South contexts. The conversation centers on whether partnerships between AI giants (like OpenAI) and grassroots organizations can generate ethical, locally grounded, and scalable solutions to development challenges, or risk reinforcing old patterns of extraction and inequality. Expert guests share frontline stories, lessons, and visions for inclusive, community-driven AI—while also addressing the risks of deepening divides.
Key Discussion Points & Insights
1. The Promise and Perils of AI in Global Development
[00:00-02:43]
- The Ubiquity of AI: AI is fundamentally transforming flows of money, access to services, and power structures worldwide.
- Critical Concern: Both Manu Chopra and Alex Nawar express worry that benefits from AI—particularly large language models (LLMs)—are skewed toward already privileged populations.
- Manu Chopra: “What if AI just makes the already economically productive even more productive… but never reaches the communities that it should reach and just that it isn’t equitable. And I think that’s what we need to fix.” (03:34)
2. The Extractive Value Chain…and Proposing a New Model
[05:30-09:22]
- Uneven Value Chain: AI’s value is often extracted from the data, labor, and languages of marginalized communities—without proportional return.
- Karya’s Model: Inspired by a project in Nandarbar, Maharashtra, Chopra describes building and fine-tuning LLMs with, by, and for local communities, ensuring:
- Community ownership of data
- Community involvement in model development, fine-tuning, and evaluation
- Economic benefit sharing
“Our communities…aren’t just excellent beneficiaries of the AI revolution, they’re excellent builders. They’re excellent evaluators.” — Manu Chopra (05:57)
3. Social Impact Partnerships: Opportunities & Early Lessons
[09:22-16:15]
- Alex Nawar’s Background: Experience bridging evidence-based research and on-the-ground implementation (IPA, GiveDirectly), applying ML for disaster response and direct cash programming.
- Highlights the importance of ownership and local trade-off decisions, as exemplified by pioneering work with Togo’s government during the pandemic.
- OpenAI’s Approach: Through the OpenAI Academy and AI for Global Development Accelerator (with Center for Global Development & Agency Fund), OpenAI is:
- Supporting organizations in the Global South with technical expertise, API credits, and research resources.
- Focusing on enabling “builders” close to the community to leverage AI for social good.
“We’re doing what we can to enable some of the actors in the field—government, nonprofits, social enterprises, or philanthropic foundations.” — Alex Nawar (13:45)
4. The Need for Locally Relevant and Contextualized AI
[17:00-22:26]
- Broccoli-Kiwi Problem: LLMs can give globally accurate but locally irrelevant responses (e.g. suggesting “broccoli and kiwi” as staple foods in Karnataka, India).
“If you ask these models, like, ‘Hey, I’m in Karnataka…I want to eat healthy. What food should I eat?’ …You should have lots of broccoli and lots of kiwi—they’re not locally relevant.” — Manu Chopra (18:53)
- Community-Driven Evaluation: Ongoing collaborative projects bring local communities into the heart of model evaluation, surfacing linguistic and contextual gaps.
“Nobody knows our communities better than our own communities and the people who serve them.” — Manu Chopra (17:00)
5. Productivity Gains: Beyond the Automation Scare
[22:26-25:10]
- Shift in Conversation: While many fear jobs will be lost to automation, both guests argue that AI has the potential to give back time to those with limited resources.
- Practical Example: Indian ASHA (healthcare) workers could serve more homes per hour with AI-powered workflows, leading to tangible income and quality-of-life improvements.
- Community Diversity: India as an example—92% chance that two randomly selected individuals speak different languages. AI needs to serve all, not just the most represented.
6. “Nonprofit Jam” and Civil Society Engagement
[23:00-26:23]
- Nonprofit Jam: Karya and OpenAI ran a cross-country event in India, convening 200+ nonprofit leaders, many new to AI but eager to leverage it for both their work and personal lives.
- Key Insights: Even basic time-saving tasks resonated deeply. Community realities and linguistic diversity must be central to model development.
7. Platform Ubiquity: Network Effects, Risks and Rewards
[26:23-31:46]
- AI as Infrastructure: There are both opportunities (distribution, scale) and risks (overcentralization, local irrelevance) as LLMs become the “default gateway” to the internet.
- Competitive Ecosystem: The field is moving fast, with sovereign (national) providers emerging. Social enterprises are able to collaborate with multiple AI labs, holding them to community-centered priorities.
“It’s not…a passive distribution channel the way Messenger is, but it’s giving a new kind of set of capabilities to builders…in many ways they’re in the driver’s seat.” — Alex Nawar (29:45)
8. Community as Experts—not Passive Beneficiaries
[32:22-35:30]
- Case Study: Collaboration with Maharashtra’s Mahabistar app (for 13 million farmers) exemplifies bringing communities into fine-tuning, evaluation, and feedback—particularly for local language support.
“If I’m doing agriculture evaluations, who is an expert but the farmer who’s going to use this model? …Our communities are experts on their lives.” — Manu Chopra (32:22)
Notable Quotes & Memorable Moments
- “AI is one of the most important technologies we've ever built...but I worry deeply about lopsided economic productivity.” — Manu Chopra (03:34)
- “We partner with labs…because it is in our best interest to make sure that model responses actually meet our communities where they are...” — Manu Chopra (18:53)
- “Every nonprofit by definition is operating under resource constraints…anything that can make their lives a little bit easier, give them a little bit more time, I think is really valuable.” — Alex Nawar (23:00)
- “If you pick two Indians at random, chance that they speak a different language is 92%—that shows up in the communities we seek to serve.” — Manu Chopra (24:36)
- “There are AI tools that can deliver one to two years’ worth of gains in learning outcomes… and AI products that can harm learning if not designed with safeguards. The debate is no longer can AI do good or harm—it’s which will scale more.” — Hans Zheng Chia (37:47)
[Second Segment] Policy Perspective: The Stakes of "Digital Ubiquity" and Reverse Flow
[36:38-54:32]
Guest: Hans Zheng Chia, Center for Global Development
1. Digital Ubiquity and the Imperative to Engage
[36:38-40:23]
- AI tools like ChatGPT are becoming as necessary as major social platforms—NGOs have “no choice” but to engage if they want relevance in their communities.
- The scale challenge: “Can the pedagogically sound models outreap the scale of less careful general-use systems?”
2. “Reverse Flow” Partnerships
[40:36-45:37]
- Traditional Model: Tech companies provide tools and resources (“technology transfer”) to governments and NGOs.
- Emerging Need: The reverse—systematic incorporation of ground-level knowledge, data, and context from NGOs back into AI products developed by labs, shaping models and benchmarks at scale.
3. Making Local Knowledge Central
[42:57-49:33]
- Digital Green example: Local insights (e.g., about “informal credit networks” for Ethiopian farmers) must inform how foundational models respond—replicating social context across all downstream applications.
- The expertise AI needs is “so, so, so much larger than what individual experts can contribute…it’s stuff the farmer knows.”
4. Beyond Grantmaking: System Change & Market Incentives
[45:15-54:17]
- Major labs (Anthropic, OpenAI, Google) are starting to shape products for social contexts (e.g., “Learn Mode” in education AI) and incorporate pedagogical best practices.
- Challenge: True systemic inclusion requires greater investment and possibly regulatory/CSR interventions since low/middle income use cases often lack clear profit incentives.
“What if the investment in low and middle income country use cases was half of what has gone into coding tools? …We need a systematic mapping of domains to prioritize—otherwise these market failures will persist.” — Hans Zheng Chia (51:19)
Action Items & What Needs to Change
For AI Labs and Social Enterprises:
- Move beyond extractive models—treat communities as builders, owners, and experts.
- Design for radical local relevance: language, context, and application specificity.
- Create institutions and feedback loops that fuel “reverse flow”—local knowledge shaping foundational models.
For Policymakers and Funders:
- Support federated approaches: data curation, evaluation, and fine-tuning rooted in local organizations.
- Ensure investment covers under-profitable, high-impact domains (not solely driven by business case).
- Promote mapping and benchmarking of domains where social impact can be maximized with AI.
Timestamps for Key Segments
- Opening Theme and Introduction – [00:00-02:23]
- Guests’ Fears & Aspirations for AI – [02:24-05:30]
- How Karya Centers Communities – [05:30-09:22]
- Nonprofit Jam & Civil Society Engagement – [23:00-26:23]
- Digital Ubiquity and NGO Relevance – [36:38-40:23]
- Reverse Flow: Local Knowledge to Tech Giants – [40:23-45:37]
- Scaling Inclusive Models: Challenges & Recommendations – [51:13-54:32]
Conclusion
This episode dives deep into the duality of AI’s potential for global development: it can meaningfully empower marginalized communities, but only with intentional, reciprocal partnerships between AI labs and ground-level organizations. The challenge for the sector isn’t just whether to engage with AI platforms, but *how to ensure that engagement builds models and solutions for—and by—those most in need of change.
For further exploration: Listen to the full podcast and follow Devex for ongoing coverage as these partnerships—and the landscape of global AI development—continue to evolve.