A third-grade student in El Salvador reads aloud to a computer. An AI tool assesses his diction, accuracy, and speaking speed, using the Spanish spoken in his own country rather than applying a standard that may not reflect his linguistic context. Based on the results, it recommends reading exercises tailored to his learning needs. His teacher remains part of the process, using the technology to support learning without isolating the student from the classroom.
Thousands of miles away, a farmer in India faces a different challenge. When a crop shows signs of disease, getting the right advice at the right time can make a significant difference to the harvest. With a basic smartphone, she can photograph an affected plant and receive practical guidance on the problem and possible treatment.
These are very different applications, yet they demonstrate the same principle: AI becomes more valuable when it is designed around the people who need it and the realities they face.
This was one of the central messages of the Global AI & Digital Summit 2026, hosted by the World Bank Group in Seoul, in partnership with the Republic of Korea’s Ministry of Finance and Economy and Ministry of Science and ICT. Under the theme Small AI. Big Impact, the discussions explored how practical, locally adapted AI solutions can expand opportunity, improve essential services, and support economic growth, particularly in developing economies.
Rethinking What AI Progress Looks Like
Much of the global conversation about AI focuses on powerful models, vast computing resources, and the possibilities of generative AI. These advances are important. But they also raise a fundamental question: how much of this progress can reach people and communities that lack the infrastructure, resources, or skills needed to use it?
The scale of the challenge is significant. According to the World Bank, approximately 2.2 billion people remained offline in 2025, with around 95% of them living in low- and middle-income countries. Beyond internet access, reliable electricity, affordable devices, and digital literacy all influence whether people can benefit from digital technologies.
There is another challenge that receives less attention: the availability of usable data. In many settings, valuable information already exists, but it may be scattered across disconnected systems, stored in paper records, or recorded in formats that are difficult to analyze. Before AI can turn such data into useful insights, it needs to be digitized, organized, and made accessible.
These conditions call for a different approach to AI development. Rather than assuming that every setting can support the most resource-intensive technologies, small AI starts with a more practical question: What is the problem we need to solve, and what is the simplest, most effective way to solve it?
Small AI refers to targeted, fit-for-purpose solutions designed around specific needs. They can use smaller models or specialized AI capabilities, operate on everyday devices, and be adapted to local data, languages, and working conditions. Some can function with limited or intermittent connectivity, reducing their dependence on large computing infrastructure. The ambition is not smaller. The approach is more focused.
From Technical Capability to Everyday Impact
The examples from El Salvador and India illustrate what this shift can make possible.
In El Salvador, AI is being used to support students’ reading development in ways that reflect their local language and learning needs. By assessing how a child reads and identifying where additional practice is needed, the technology helps teachers provide more targeted support. Just as importantly, AI is being introduced within a social learning environment, where teachers continue to guide students and support their development.
In India, small AI is helping farmers access agricultural guidance through basic smartphones, including advice on crop diseases, pests, and treatments. For smallholder farmers, particularly women working in communities where specialist support may not always be readily available, timely information can help inform decisions that affect their crops, productivity, and livelihoods.
Neither application requires AI to replace the people already doing the work. Instead, the technology extends their access to information and helps them make more informed decisions.
This distinction matters. A useful AI solution does not have to perform an unlimited range of tasks or introduce an entirely new way of working. It may simply help a teacher identify a learning gap earlier, or allow a farmer to recognize a crop problem before it becomes more damaging.
Large language models may expand what AI can do. Small AI can expand who benefits from it and where those benefits can be realized.
Building the Conditions for Impact
Small AI offers an important opportunity for developing economies, but its potential should not be confused with the idea that AI can succeed without supporting infrastructure.
Even a well-designed solution needs the right conditions to move beyond an initial pilot and deliver sustained results. The Global AI and Digital Summit highlighted four interconnected priorities: expanding digital access, improving affordability, strengthening local ecosystems, and building AI readiness.
Access means ensuring that people have the connectivity, electricity, and devices needed to use digital services. Affordability means making those services economically viable for the people and organizations they are intended to serve. Ecosystem development means bringing governments, technology providers, researchers, businesses, and development partners together to build solutions that respond to real needs. AI readiness means investing in data foundations, digital and technical skills, institutional capacity, and the governance needed to deploy AI responsibly.
These priorities also highlight why local context matters. Solutions developed elsewhere can provide valuable starting points, but their successful application depends on understanding local languages, institutions, workflows, and needs. Technology must be adapted, tested, and supported by the people who will use it.
Responsible implementation is equally important. AI should strengthen human capabilities, not introduce new barriers or leave critical decisions to systems that people cannot understand or challenge. Human oversight, trust, and accountability must remain part of the design.
Without these foundations, promising applications risk remaining isolated pilots. With them, proven approaches can be adapted and scaled to reach more people and communities.
Making AI Work Where It Matters Most
The significance of small AI goes beyond making technology less expensive or easier to deploy. It offers a way to rethink how innovation is translated into development outcomes.
For a student, that may mean receiving learning support suited to their needs. For a farmer, it may mean accessing information that was previously difficult to obtain. For a public institution, it may mean delivering services more effectively. In each case, the value lies not in the sophistication of the technology alone, but in what it makes possible for the people on the receiving end.
For developing economies, this creates an opportunity to approach AI not only as a technological frontier to catch up with, but as a practical tool for improving productivity, strengthening essential services, and creating new economic opportunities.
At the ESG & Data for Good Center of Excellence, this perspective reinforces an important principle: meaningful innovation starts with understanding the challenge, designing around real needs, and ensuring that technology creates value beyond its technical capabilities.
The future of AI will undoubtedly include increasingly powerful models. But progress should also be measured by how effectively AI reaches people who have historically had limited access to its benefits.
Sometimes, the most significant impact does not come from building a system that can do more. It comes from designing one that helps someone do something they could not easily do before.