AI Algorithms and the Underrepresentation of Africa

The article by Akintunde Babatunde addresses the critical issue of who writes the algorithms that power AI systems, particularly in the context of Africa. It highlights that while Africa has over 2,000 languages, major languages like Hausa and Yoruba constitute less than 0.004% of the datasets used to train AI models like ChatGPT.

This underrepresentation reflects a broader issue of historical and geographical power structures that influence knowledge production. The article emphasizes the need for African knowledge to be digitized and made accessible, as much of it exists in physical archives and oral traditions that AI systems cannot currently utilize.

Babatunde argues that the lack of African knowledge in AI training datasets is not merely a technical gap but a structural disadvantage that requires urgent attention and repair. The piece calls for a reevaluation of how knowledge is valued and represented in the digital age, stressing the importance of inclusivity in AI development.