Researchers are trying to bridge the access gap for non-English speaking Africans using AI by releasing the largest known dataset of African languages.
The African Next Voices project collected 9,000 hours worth of conversations across Kenya, Nigeria, and South Africa, in 18 different languages, which is now AI-ready and open-access for developers to use.
Research has suggested that AI could unlock up to $100 billion in annual economic value in Africa, but access is a barrier: For instance, ChatGPT recognizes only 10-20% of requests written in Hausa, a language spoken by more than 90 million people across West Africa, Nature reported.
“We think in our own languages, dream in them, and interpret the world through them,” the University of Pretoria’s Vukosi Marivate, who worked on the project, told the BBC. “If technology doesn’t reflect that, a whole group risks being left behind.”