World Bank urges developing countries to build small AI models

The World Bank used this week's annual meetings in Bangkok to urge developing countries to start building "small AI" — specialised models trained on narrow datasets for a specific purpose, able to run on a mobile phone where connectivity is poor. Paschal Donohoe, the bank's chief knowledge officer and a former Irish finance minister, said countries the bank serves could see development that "could have taken centuries to happen" occur within decades.

Donohoe's example: farmers in the bank's Indian agricultural scheme AgriConnect need soil analysis in their own language, and there will not be enough scientists and agricultural experts to analyse every small farm in India. A narrow model, in the local language, on a phone, in a setting with low connectivity — that is the bank's picture of where AI pays off.

The bank's own development report, published in August, supplies the backdrop: almost a third of rural schools in sub-Saharan Africa lack electricity, and more than two-thirds lack a consistent internet connection. IMF managing director Kristalina Georgieva, speaking on a panel Monday, listed the preconditions — electricity, digital infrastructure, people geared to adoption, government protection against bad actors — and said that with the basics in place, AI adoption could be "as democratic as the cellphone." She also warned that 40 per cent of jobs worldwide are likely to be augmented or eliminated by AI over time, with the impact falling hardest on the young, who normally enter the workforce through routine jobs and climb from there; if those jobs go, "no young person would want to be a barista in a coffee shop as a professional future."

The bank's pitch cuts against the industry's central claim. The largest AI firms raise capital on the argument that capability lives in scale — larger models, larger training runs, larger power draw. The World Bank is telling poor countries that the value for them sits in narrow, cheap models running on hardware they already own. If small AI covers the clinic, the classroom and the farm, the investment case for the next round of frontier compute rests on a shrinking set of tasks.

The bank also has its own position to advance. Under president Ajay Banga it has made attracting private capital into emerging economies the focus of its work, and Donohoe announced that $120 billion of private capital was mobilised in the past year through guarantees and other financial instruments. Small AI is a product that capital can be pointed at.

Georgieva's precondition list begins with electricity. Almost a third of rural schools in sub-Saharan Africa do not have it.