The Gates Foundation warned that billions of people are at risk of being left out of the benefits of artificial intelligence because the technology does not understand their language or the conditions in which they live.
The foundation's annual report on global progress showed that the rapid increase in the use of AI could widen the gap between rich and poor if the technology is not deliberately built to serve the people of the world's poorest countries.
More than 90 percent of the data used to train early large language models came from English-language sources. At the same time, leading AI-based speech recognition systems err less than 6 percent of the time in English.
In Yoruba, which is widely spoken in Nigeria and other parts of West Africa, the error rate is 10 times higher.
The foundation found that AI tools may also fail when moved from wealthy countries to poorer ones and deployed there without local data. For example, a system designed for a commercial farmer in Iowa may not understand the crops, pests, markets, or constraints faced by a smallholder farmer in Ethiopia.
The Gates Foundation's 2026 "Goalkeepers" report, published on Tuesday, also warned that if everything is left to the market, the most powerful AI tools will likely be built first for the people and institutions that can afford them, not for those who could benefit the most.
However, the report also shows that AI can help address severe shortages of doctors, teachers, and agricultural advisers in poorer countries by putting expert knowledge into the hands of those who directly serve people.
The foundation has pledged to dedicate at least US$1 billion over the next two years to efforts to make AI more accessible. The funding will go to health, education, and agriculture sectors as well as digital infrastructure, such as local-language datasets.
Bill Gates said in the report: "AI can either become a powerful tool for reducing inequality or widen the gap. This is not a long-term forecast. This is a choice that is in front of us right now."
He added: "The decisions made in the next 12 to 18 months about how AI is built, funded, and deployed will determine whether this technology primarily benefits those who already have the most resources or reaches those who have the least."
Early projects include an AI system that has improved diagnostic accuracy in clinics in Kenya, educational tools that have shown marked learning improvements in Sierra Leone and the United States, and an agricultural service system now used by more than 740,000 farmers in India.