World Bank Chief Makes the Case for "Small AI" as a Tool to Narrow the Global Development Gap

As governments and international bodies wrestle with how to regulate artificial intelligence, a parallel debate is gaining momentum: whether the same technology that many fear will deepen global inequality could instead be used to narrow it. The question sits at the centre of a widening policy discussion, with cautionary voices warning that AI could entrench the advantages of wealthy nations while leaving poorer ones further behind. Yet the World Bank is pushing a more optimistic argument, suggesting the technology can be steered toward the benefit of developing economies if it is designed and deployed with their specific conditions in mind.

The stakes are considerable. Developing countries often face persistent shortages of qualified teachers, doctors and engineers, and AI tools are sometimes presented as a way to extend scarce expertise to populations that currently lack access to basic services. At the same time, the infrastructure required for advanced AI — reliable electricity, high-speed internet, expensive computing hardware and large volumes of digital data — is exactly what many low-income countries lack, raising the risk that the benefits of the technology will flow mainly to those already well connected. This tension between promise and peril has made AI one of the most contentious subjects in development policy.

The World Bank's president has argued that developing economies can capture real gains from AI provided the technology is adapted to their circumstances, an approach he describes as "small AI." In his account, this means systems that are more appropriately adjusted to local conditions, better able to run offline, and capable of operating in environments where connectivity or electricity may not be at a very high level. Rather than assuming the same data-hungry, cloud-dependent models used in rich countries will simply trickle down, the concept implies building lighter, more resilient tools that work within existing constraints.

That framing matters because the standard model of AI deployment assumes constant connectivity and abundant power — conditions that are far from universal. In much of sub-Saharan Africa, South Asia and parts of Latin America, electricity supplies can be intermittent and internet coverage uneven, particularly in rural areas. A tool that only functions with a stable broadband connection is of limited use to a farmer, a clinic or a school in a region where neither is guaranteed. Small AI, in this reading, is less a technical downgrade than an exercise in matching the tool to the environment.

The World Bank chief also stressed that even as technology reshapes daily life, development agencies must keep their attention fixed on longstanding priorities: reducing inequality and meeting basic human needs such as access to clean water, healthcare and education. His argument is that the rise of AI should not distract institutions from those fundamentals, but should be integrated into efforts to address them. By ensuring that the development of artificial intelligence does not cause policymakers to lose sight of these issues, he contends, the international community can recognise the scale of the change underway while still keeping its core mission in view.

In this view, AI is neither an automatic remedy for poverty nor an inevitable driver of division. The determining factor is the policy environment surrounding it. The World Bank president makes the case that while AI undeniably brings challenges, small AI in particular — harnessed correctly and accompanied by the right policies — can play a positive role. That emphasis on governance and regulation shifts the debate away from whether the technology is inherently good or bad and toward the practical question of how it is introduced, who controls it and whose needs it is designed to serve.

The discussion comes amid mounting scrutiny of AI's concentration in a handful of companies and countries. The most powerful models are developed by a small number of firms based largely in the United States and China, and the computing power required to train them is expensive and energy-intensive. Critics argue this dynamic risks creating a new form of dependency, in which developing nations become consumers of technology designed elsewhere rather than participants in shaping it. Questions of data ownership, language representation and algorithmic bias add further layers of concern, since tools trained on limited datasets may perform poorly for populations that were never well represented.

There is also a more immediate worry about jobs. Economies that rely heavily on labour-intensive sectors, including customer service, data processing and basic administrative work, could see those roles displaced by automation before workers have had a chance to move into new industries. Wealthier countries can cushion such transitions with social safety nets and retraining programmes; poorer ones frequently cannot. That asymmetry is central to fears that AI could widen the gap between rich and poor nations rather than close it, even if the technology delivers overall productivity gains.

Against that backdrop, the "small AI" concept represents an attempt to reorient the conversation from frontier models toward practical applications. Examples might include diagnostic tools that function offline in remote clinics, educational software that adapts to local curricula and languages, or agricultural advisory systems that deliver advice by text message rather than requiring a smartphone with continuous data access. Each of these is modest compared with the large language models dominating headlines, but proponents argue their cumulative impact on daily life in developing countries could be substantial.

International institutions are not the only actors in this space. The push for AI governance has accelerated globally, with voluntary safety agreements, regulatory proposals and ethical frameworks emerging in multiple regions. Some of these efforts have been led by governments eager to attract investment while limiting harm; others have come from civil society and researchers concerned about accountability. The result is a fragmented landscape in which rules differ sharply from country to country, complicating efforts to ensure that AI development serves broadly shared goals rather than narrow commercial interests.

For developing economies, the policy choices are difficult. Attracting investment in digital infrastructure competes with spending on health, education and water — the very priorities the World Bank president says must remain central. Building local capacity in data science and engineering takes years, and retaining skilled graduates is a chronic challenge when better-paid opportunities exist abroad. Meanwhile, the rapid pace of technological change means that any regulatory framework risks being outdated before it is fully implemented. Navigating these trade-offs requires both resources and institutional strength, which are precisely what many low-income countries find in short supply.

The optimism expressed by the World Bank nonetheless rests on a testable premise: that technology adapted to local conditions produces better outcomes than technology imposed from outside. If small AI tools can demonstrably improve access to healthcare, education and financial services in places where those services are scarce, the case for targeted investment becomes stronger. If they fail to scale, or if they are deployed without the supporting policies and infrastructure needed to sustain them, the sceptics' warnings about deepening inequality will look increasingly prescient.

What emerges from the debate is a recognition that AI's effect on global inequality will not be determined by the technology alone. It will depend on decisions about design, distribution and governance — choices that are political and economic as much as technical. The World Bank's argument is essentially that developing countries need not be passive recipients of a transformation happening elsewhere, provided the tools are built for their realities and the policies around them are sound. Whether that vision can be realised at scale remains an open question, and one with consequences for billions of people.