Back to Insights
4 min read
29 July 2026
The saying “When an old person dies in Africa, a library burns to the ground”has never been more pertinent. Amongst the more than 2,000 African languages, almost a third are currently endangered and 10% predicted to disappear within the next century. Not only do African languages lack meaningful digital presence, but pre-colonial history, which is typically transmitted orally across generations, is also being lost as elders pass away and communities urbanise. This gap serves to empower and perpetuate colonial narratives at the expense of indigenous knowledge and history.
The reality that AI is grounded in Western Academia and lacks a deep understanding of African history further exacerbates this chasm. When I asked premium Claude AI to recount the history of my tribe, the Shilluk – one of the largest in South Sudan – it was only able to recount a few short paragraphs of a simplified history. Claude conceded that “Much of the early history rests on oral tradition and on colonial-era anthropology, both of which carry their own biases” , thus specific dates and details are not factual but rather reconstructions.
Upon being asked to teach some Shilluk language, the AI could only provide a couple verified words, and a Shilluk missionary latin-based alphabet. The limitations extend to Juba Arabic, which is the most widely spoken language in South Sudan. In 2023 there were over 2.3 million South Sudanese refugees, with Uganda hosting almost 1 million; and yet, the AI was unable to translate a basic Uganda tenancy agreement from English to Juba Arabic.
The technocolonialism problem
Such shortcomings are particularly concerning given AI’s reliance on critical minerals from Africa, as well as cheap labour for data labeling and content moderation. Ironically, African contributors are being left out as the technology evolves, with websites, applications and digital/technical tools on African cultures, history and languages being largely excluded or non-existent. The Western-dominated AI industry has been uninterested in developing these aspects – yet the market for language learning and cultural preservation is strong: Africans represent 19% of the world’s population. In 2024, the first generation, continental and intra-continental African diaspora exceeded 45 million individuals, with strong remittance flows.
A new developing and diverse AI market
African governments, led by Ghana and Kenya, have taken a keen interest in AI, data sovereignty and linguistic justice, building on the momentum generated by startups. In February 2026, the African Development Bank and the UNDP launched the AI 10 billion initiative which plans to mobilise USD 10 billion by 2035 towards productive AI continental development.
The Masakhane African Languages Hub is a community-led research and open datasets hub for the continent, focusing on the creation of meaningful African data through a range of projects. In Uganda, SunbirdAI has created a Large Language Model (LLM) AI beta module, in collaboration with the Ugandan government, called Sunflower. It works in English and some of the most widely spoken Ugandan languages. Despite the LLM still being a work in progress, with a majority of the country’s 30+ languages not being represented when tested, it is a meaningful step towards bridging language and understanding gaps across countries.
There is often a lingual hierarchy, in which non-dominant local languages and communities are invisibilised, seeing less development and education. LLMs like Sunflower could open the door to more meaningful conversations on strengthening digital infrastructures, access and representation. AI is also used in digital technologies such as AI-assisted archiving and metadata, AI-powered data visualisations, 3D reconstructions and audio archives. Organisations such as <a href="https://africandigitalheritage.org/” rel=”nofollow noopener” target=”_blank”>African Digital Heritage and Open Restitution Africa have been actively doing so in ways that empower local communities and prioritise African history and culture.
The efficiency and equity problem
While these are just a few examples of undeniable progress and potential, key structural gaps remain. With developments mostly targeting high-rederrepresented from modern digital infrastructure. While Google offers translation in over 60 African languages, many lack dictation despite being oral dependent languages
LLMs use tokenisers to fracture text into machine readable numerical vectors. Because they are trained on English and Latin scripts, their tokenisers are inefficient for African orthographies, with an economic analysis of the Ge’ez script(the bedrock for Amharic and Tigrinya) revealing that a single word in English requiring one or two tokens, can require up to ten tokens when written in Amharic. With commercial AI Application Programming Interfaces (APIs) charging per token, and LLMs often constrained by working memory token limits, an Amharic customer service chatbot becomes overtly expensive and less reliable.
A call for preservation
There is a real need to develop the multidimensional AI market across the continent to prioritise language preservation and cultural representation. Combining the current cultural and lingual approaches with small footprint models and learning platforms tailored to the African context, such as Edge AI, would revolutionise preservation. This could help bilingual teaching in Africa, strengthening continental education.
Digital AI tools could even be used by NGOs for refugee populations to facilitate immersion and conflict resolution, by facilitating teaching and understanding amongst individuals from different ethnic groups and backgrounds. For diasporas outside the continental, individuals would have more digital res, leading to a greater sense of belonging and wellbeing
The data we have today excludes African knowledge and feeds a distorted version of it into the systems that will shape how the continent understands itself for decades. This is why it is crucial for the continent and its diasporas to prioritise language, culture and history preservation. This does not only enable better cultural ties and community, but it also allows for better African contextualised decisions. It deserves backing.
About the Author
Djourbaine Bilnyang is an intern at Africa Practice and an International Relations BSc student at LSE. Of South Sudanese heritage and passionate about African affairs and development, she became interested in AI during her role as a policy editor at the LSE Africa Summit 2026. Djourbaine can be contacted at [email protected]
