Small AI models could unlock new markets in developing countries

Small, efficient AI models capable of functioning on cell phones with poor connectivity could accelerate economic growth in emerging economies, according to a senior World Bank official. At the ongoing World Bank and IMF annual meetings in Bangkok, Paschal Donohoe, Managing Director and Chief Knowledge Officer, argued that small AI could bring development that previously took centuries to achieve within a matter of decades.

Donohoe explained that small AI, or "small language models," are lightweight artificial intelligence models that require minimal computing power, work well with weak internet connections, and consume less energy than conventional "large language models." These models, which the World Bank has been "enabling," are specifically designed to tackle local development challenges where infrastructure and skills are limited.

Evidence of small AI's potential is already evident in many developing nations, where it has helped farmers identify crop pests, enabled healthcare professionals to screen for tuberculosis without reliable broadband, and provided students with access to AI tutors. Donohoe cited an example of Indian farmers utilizing the World Bank’s agricultural scheme, AgriConnect, who needed soil analysis in their native language.

He argued that small AI, which can be accessed via mobile phones, could address the information needs of small farms in India without relying on the availability of scientists and agricultural experts. The World Bank is working towards creating such products through collaboration between governments, the World Bank, and the private sector.

However, there are significant disparities in the adoption of generative AI chatbots like OpenAI’s ChatGPT between high- and low-income countries, with only 0.7 percent of low-income country internet users having adopted the technology as of April 2025, compared to nearly a quarter of high-income country users. IMF Managing Director Kristalina Georgieva emphasized during a World Bank Group panel on small AI that achieving workable AI systems in low-income nations requires preconditions such as widespread access to electricity and digital infrastructure.