Stanford Graduate School of BusinessWorld Development Report 2026: Navigating AI's Opportunities and Risks in Developing Economies
Stanford Graduate School of Business
Introducing the conversation 0:00
Somik Lall opens a conversation with Indermit Gill about the World Development Report on AI, framing it around Gill's reputation for turning complex ideas into clear ones and for challenging conventional wisdom about technology and development.
Why the outlook became hopeful 3:01
Gill recalls that two years ago the mood around AI was mostly negative, dominated by fears about job losses, rising energy use, and even existential risk, debates shaped mostly by the US, Europe, and China rather than the rest of the world. The team spent two years on the report instead of the usual one, expecting it to be mostly about technology, but found it ended up being only about a quarter technology and far more about education, industry organization, and regulation.
AI and sluggish global growth 8:00
Gill notes that global productivity growth has been declining for two and a half decades, prompting talk of secular stagnation, but argues AI could add roughly half to one percentage point to growth, comparable to the historic impact of electricity or computers, restoring the dynamism last seen fifteen to twenty years ago.
Adapt, don't invent 10:00
Gill explains that developing countries have smaller gaps in money, skills, and technology access than expected, but should focus on adapting existing AI tools rather than inventing new ones, since the technology moves too fast and is too context-specific for that to pay off. He suggests being selective, since predictive AI, used for backend tasks, is the area where these countries are already best prepared to turn into real productivity gains in both public and private sectors, more so than generative or agentic AI.
Applications over infrastructure 12:00
Gill advises developing countries to focus spending on AI applications rather than costly infrastructure like data centers and large language models, since skipping that expense avoids debt that would otherwise have to be recouped by cutting skilled jobs.
Advice for a finance minister 13:00
Asked what he would tell a mid-sized country's finance minister facing pressure to pick a technology camp, Gill says the answer is not sovereignty but interoperability, aligning with both the US and China rather than choosing one, since full independence would be too costly. On regulation, he points to the earlier World Development Report on standards, arguing that effective rules should start as voluntary industry standards, with government stepping in only to keep barriers to entry low, citing Herbert Hoover's work as Commerce Secretary as a model for how that balance was struck.
Risk of Incumbents Writing Rules 17:00
Somik Lall raises a concern about letting industries write their own voluntary AI standards, since a handful of large incumbents dominate the field and could use standard-setting to block competitors, especially in countries with weak competition regulation. Indermit Gill agrees this is a real danger but says there is no better starting point than industry-led standards. He argues you simply have to be clear-eyed that large AI firms, like finance firms before them, will act in their own shareholders' interest and may use fear of catastrophic risk to lock in rules that favor themselves, much as "too big to fail" arguments once protected big banks. His proposed fix is to let companies draft the standards they think are needed, then have independent economists assess whether those rules would make it too hard for new entrants to compete, and revise the rules accordingly.
Cheap AI Access and Old Lessons 20:31
Asked whether cheap access to frontier models from a few dominant companies is worrying, Gill recalls the 2016 World Development Report on Digital Dividends, when Facebook pushed "Facebook Zero," free internet access tied to its own platform, in India. The World Bank's balanced report was being used against the deal, prompting pressure from Facebook's Sheryl Sandberg for a retraction, which Gill resisted by having a neutral op-ed written independently rather than one supplied by Facebook. India ultimately rejected that path and built its own digital public infrastructure stack, achieving some of the world's best digital access. The lesson is that cheap access today can carry hidden long-term costs.
Smaller, Targeted AI Applications 25:30
Gill is more optimistic about smaller, application-specific AI tools for health, education, and justice than about concentration fears around huge language models. These tools need local adaptation, but that adaptation can travel regionally, so something built in parts of India could work in Bangladesh or Pakistan too. He points to deep existing failures, kids learning less than before, dismal conditions even in a top public hospital in Chandigarh, and years-long court backlogs, including a Stanford researcher's account of a judge translating testimony and a stenographer typing with two fingers, as the gap AI could fill.
Countries With Strongest AI Potential 29:30
Asked which countries show the most promise, Gill names Kenya for judicial reform, Bangladesh for medical applications, and southern Indian states more broadly. On Kenya, AI helps sort cases into arbitration versus court, easing backlogs so severe that some Indian states could not clear them in fifty years. He also pushes back on fears that AI will devastate India's call-center jobs, noting these employ only a small share of workers while 45 percent work in tiny firms and another 45 percent on farms, where AI already helps with market access and farming decisions, making the overall effect productivity-enhancing for most workers. He adds that AI-driven procurement reform is the single most promising, low-cost way to cut corruption in developing countries.
Procurement as a Priority Use 33:30
Gill points to procurement as the area where AI could deliver the biggest anti-corruption gains, and the session then opens to audience questions. One attendee asks about shared or regional infrastructure. Lall notes that no single developing economy can afford AI-specific data centers or energy systems alone, since basic development needs like schools and healthcare still take priority. Gill suggests regional hubs led by larger countries, such as Nigeria for West Africa, Kenya for East Africa, and India for South Asia, since AI trained on local data, say Lagos, should transfer reasonably well to nearby places like Kano or Accra. Lall counters that without shared regional energy systems, this approach still runs into limits.
Bottlenecks, Strategy, and Spillovers 37:31
Asked how to help policymakers who don't understand AI's benefits, Gill argues the real fix is strong competition, where firms that fail to adapt get competed out, paired with simple guidance: pursue adaptation aggressively, then predictive uses. Asked what the World Bank should prioritize in lending, advisory work, and convening, he says it should execute its own AI implementation plan and frame AI as a tool for efficiency rather than promising job numbers directly. On whether benefits reach beyond entrepreneurs, Gill says the bigger gains for the poor come from better government services, health, education, and justice, which could reduce inequality of opportunity even as outcomes diverge. Colleagues add that AI also widens small firms' capabilities and that entrepreneurs generate spillovers, lower costs, and better services for everyone around them.
AI as an Equal Opportunity Tool 48:01
The discussion closes on the idea that knowledge spreads best when it is not blocked, and AI should be used to let young people express their creativity and reach their full potential. If AI is treated as a broad enabler rather than a tool for a few entrepreneurs, the benefits spill over to the whole society instead of staying concentrated.
Firms Under Pressure to Adapt 49:02
Indermit Gill predicts that in ten years, countries left behind by AI will get blamed for lacking education or electricity, but he suspects the bigger reason will be that firms simply were not pushed hard enough to adopt productivity tools like AI. He gives his wife's example of a 25-person firm using AI to cut tax-filing time drastically, needing only one skilled person instead of two.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
