World Bank: AI speaks English fluently — but African languages poorly

The World Bank has issued a stark warning to the whole of Africa: artificial intelligence speaks English fluently, but speaks African languages poorly. And that is leaving millions of people behind.

In its new report, Africa Economic Update: Building AI Readiness, published in October 2026, the bank says global datasets used to train AI systems substantially underrepresent African countries, their languages, their institutions, and their economic activities.

Why does this matter? AI learns from the data it is fed. Africa has roughly 2,000 languages — but around 56% of open-source training data is in English only, according to the World Bank. When system developers lack high-quality material in a given language, AI performs worse in it: translations go wrong, answers come out incomplete, and local expressions are misunderstood.

The consequences can touch everyday life. A farmer may need agricultural advice in terms they understand, a student may need learning materials in their preferred language, and a health worker may need clear communication tools for patients.

But there is good news too. The report says the deepest returns for Sub-Saharan African countries lie in adapting AI to local data, languages, and problems. One example is InkubaLM, a small language model built by African AI company Lelapa AI, trained on five African languages — isiZulu, Yoruba, Hausa, Swahili, and isiXhosa — which the report says outperforms much larger models in those languages. It is proof that models focused on African languages can fill a gap the big global systems cannot.

Still, the World Bank raises an important question: whose languages are these? Building language datasets requires written materials, speech recordings, translations, and contributions from local researchers and communities. Clear rules are needed: permission to use the data, privacy protection, and how contributors are recognized or compensated — because commercial companies can profit economically from the language resources of communities that may not share in the benefits.

Another challenge is infrastructure: skills, access to computing equipment, and affordable internet. Even if a tool supports an African language, its value depends on whether people can access it affordably and use it with the devices they have.

What does this mean for you?

For farmers: AI advice on crop diseases or weather will only be useful if given in the Kiswahili you actually understand — not in cobbled-together translation.

For teachers and students: Learning tools that speak Kiswahili well could open up knowledge for millions — but teachers should verify AI answers, because it can make mistakes in our languages.

For journalists: This is a story to follow: who owns the Kiswahili data used to train AI? Questions of fairness and compensation will decide whether Kiswahili speakers benefit — or are merely used.

For small businesses: Many customers will ask AI in Kiswahili before buying. A business that can communicate in the customer's language will have an edge.