Generative AI & Knowledge Gaps

Introduction: The New Knowledge Divide

Generative AI promises to democratize creativity and knowledge, making vast worlds of text, images, and ideas accessible to anyone with an internet connection. Yet beneath this promise lies a troubling paradox: the very data on which these systems are trained reflects a deep imbalance in whose voices, values, and philosophies count as knowledge.

For South Africa — and the African continent at large — this is not a new story. From colonial archives to post-apartheid curricula, whose knowledge is preserved, amplified, or erased has always been political. In the age of AI, this epistemic inequality risks being scaled up and automated.

Epistemic Inequality: When Absence Becomes Ontology

AI systems do not invent knowledge out of thin air. They generate new content by patterning on training data, which is overwhelmingly dominated by Western sources — English-language books, European art, American media, and research from Global North universities.

This creates a double absence for African knowledge systems:

• Absence in representation: African histories, oral traditions, and indigenous philosophies are rarely digitized at scale.
• Absence in ontology: When AI generates responses, the lack of African perspectives is not just a missing footnote — it becomes structured into reality as though it does not exist.

In philosophical terms, this is what we might call an ontological absence: what AI “knows” becomes what the world seems to be. If African thought is absent from the data, it risks being absent from the future imagination.

The South African Case: Whose Knowledge Gets Remembered?

South Africa’s rich intellectual heritage — from Ubuntu ethics to Khoisan cosmologies to the radical philosophies of Steve Biko — has shaped not only local identity but global debates on justice, community, and freedom.

Yet in the datasets that power AI, these voices are scarcely present. For example:

• Ubuntu, one of Africa’s most influential ethical frameworks, is often reduced in Western AI outputs to a cliché about “I am because we are,” without the nuance of its communal jurisprudence or spiritual depth.
• Oral histories of the Bantu migrations, San cosmologies, and anti-apartheid struggle poetry remain locked in archives, untranslated, or in fragile community memory.
• Research shows that over 80% of online academic content indexed for AI models comes from the Global North, with less than 5% from Africa (UNESCO, 2023).

This epistemic gap is not just about representation. It shapes how AI “imagines” the world — whose voices it defaults to, whose ethics it encodes, whose future it designs toward.

Affecta Nullius: Synthetic Emotion and the Absence of Africa

In my own philosophical work, I have described a concept called Affecta Nullius — the real emotions humans feel toward entities or narratives that have no grounding in reality.

When AI presents a world in which African thought is absent, users emotionally engage with that world as though it is complete. The illusion becomes affectively real, even if it is historically incomplete. This is epistemic violence by omission: a false fullness that erases.

Toward Epistemic Justice: Training AI on African Knowledge

The challenge, then, is how to redress this imbalance. A few pathways are emerging:

1. Digitizing African archives: Universities, libraries, and museums in South Africa and across the continent can prioritize digitization of indigenous texts, oral histories, and philosophies.
2. Local language corpora: Generative AI needs isiZulu, isiXhosa, Sesotho, and Afrikaans data at scale, not just English. Training models on African languages expands epistemic diversity.
3. Ethical partnerships: Policymakers can push for collaborations with global AI firms that go beyond loose translations, ensuring local knowledge forms the basis of training.
4. Community-driven AI: Projects like Masakhane already show how grassroots African NLP (natural language processing) communities can shape the future of AI with contextual sensitivity.
5. Philosophical grounding: AI ethics cannot only be Kantian, utilitarian, or Silicon Valley pragmatism. It must also be Ubuntu, Bikoist, and San cosmology.

Conclusion: A Future Worth Imagining

Generative AI has the potential to be a bridge or a barrier. Left unexamined, it risks scaling Western epistemic dominance into the digital age. But with deliberate effort, it can help amplify African thought, preserve fragile knowledge systems, and seed new global philosophies rooted in the South.

The task is urgent. For if Africa’s histories, philosophies, and wisdoms are absent in the datasets that define tomorrow, they risk being absent in tomorrow itself.

Epistemic inequality is not just about knowledge gaps — it is about futures foreclosed. Generative AI must not be another frontier of erasure, but a space where Africa finally speaks in its own voice.

📖 References

• UNESCO (2023). Global Report on Cultural Diversity and AI. Paris: UNESCO.
• Masakhane NLP (2022). Grassroots Approaches to African Language AI. Community Research Collective.
• Wiredu, K. (2004). African Philosophy: An Introduction. Cambridge University Press.
• Biko, S. (1978). I Write What I Like. Johannesburg: Heinemann.
• Mbiti, J.S. (1990). African Religions and Philosophy. London: Heinemann.