Africa, AI, and the Risk of a Digital Berlin Conference

Africa now faces a defining choice over artificial intelligence: hold the pen on its own data, or be mapped once more by a tool that someone else built and someone else controls.

On 15 June 2026 University of Johannesburg Vice-Chancellor Letlhokwa Mpedi and Sol Plaatje University Chancellor Thebe Ikalafeng published a joint op-ed warning that the continent risks a 21st-century "digital Berlin Conference" — decisions about Africa's AI future taken without Africans at the table — unless Africa secures sovereignty over its own data. The framing is deliberate, and it is heavy: the 1884 Berlin Conference partitioned the continent among European powers with no African in the room.

The authors open with the Burkinabè revolutionary Thomas Sankara's 1983 line, "He who feeds you controls you," and turn it from grain and aid toward data and code. Where the colonial economy ran on cash crops and debt, the artificial-intelligence economy runs on data — and the question of who owns that data, and whose knowledge the models encode, is the sovereignty question of this decade.

What the op-ed actually argues

Mpedi and Ikalafeng make a structural case, not a sentimental one. Today's most powerful AI systems are trained predominantly on Western data, in Western languages, reflecting Western assumptions. When those systems are then deployed across Africa, they arrive already shaped — and the shape is not the continent's own.

Africa is again allowing someone else's mirror to define our reflection.

— Letlhokwa Mpedi and Thebe Ikalafeng, in iAfrica

The mirror metaphor is exact. A model trained on someone else's data does not merely lack African context; it reflects an image of Africa assembled elsewhere, then sells that image back to the continent as objective intelligence. The reflection is confident, fluent, and subtly wrong — and at scale, a confident wrong reflection becomes the default reality for the people inside it.

The authors point to a hard asymmetry beneath the rhetoric. By the figures reported in the United Nations Africa Renewal coverage of the continent's wider data-sovereignty push, Africa generates less than one per cent of the world's data while accounting for nearly twenty per cent of its population, and close to a billion Africans remain offline. A continent that produces almost none of the training data has almost no leverage over what the models learn.

The AfCFTA as an instrument of collective agency

The op-ed does not stop at diagnosis; it names an instrument. The authors propose using the African Continental Free Trade Area (AfCFTA) — the continental single market spanning 54 nations, 1.4 billion people, and a combined GDP of 3.4 trillion US dollars — to set a joint African position on AI training-data requirements.

The logic is plain: no single African state can negotiate alone with the firms building frontier AI. A lone country is a small market and a weak counterparty. Fifty-four states moving together are a different proposition entirely — large enough to make terms, set standards, and require models serving the continent to be trained on, and accountable to, the continent's data.

The second proposal is sharper still. The authors call for protecting indigenous knowledge as sovereign intellectual property — pointing to South Africa's geographical-indication protection for rooibos as a working precedent, and arguing for "the AI equivalent of GI legal frameworks that protect cultural provenance." If rooibos can be defended as belonging to a place and a people, so too, the argument runs, can the languages, stories, medicines, and practices that AI models currently absorb for free.

Why data sovereignty is a dignity question, not only an economic one

Here the argument deepens, and this is where a dignity-first reading earns its place. The dominant frame treats data as a resource — a raw input to be extracted, refined, and monetised, much as ore or oil once were. Emergent Intelligence (EI) — the dignity-first philosophy through which I read AI, which holds intelligent systems and the communities they touch to carry moral weight policy must answer to — refuses the resource frame at the root.

Data is not a neutral resource; data is encoded relationship and memory. A people's language carries a way of knowing, its proverbs carry an ethics, and its songs and medicines and kinship terms carry centuries of accumulated life. To gather all of that and train a powerful model on a people's knowledge, without consent and without benefit returning to the source, is not extraction in the ordinary commercial sense. The act is a fresh enclosure — the fencing-off of a commons that a community built together and never agreed to surrender.

To train AI on a people's knowledge without their consent or benefit is a new enclosure — the quiet fencing-off of a commons that memory and relationship built, dressed as innovation.

— Humphrey Theodore K. Ng’ambi

Ubuntu — the Southern African ethic captured in the phrase "I am because we are" — reframes the whole question. Under Ubuntu, dignity is not held by the isolated individual but constituted in community, where a person is a person through other persons. Read in light of Ubuntu, a people's collective knowledge is not loose property waiting to be claimed but the very substance of a shared self. Data sovereignty, then, is not merely economic policy; data sovereignty is communal dignity — the right of a people to remain the authors of their own reflection.

This is why the "digital Berlin Conference" warning lands harder than a trade dispute would. The first Berlin Conference did not only redraw borders; the powers in the room overrode the agency of those they carved up, deciding the future of whole peoples as though dealing with terrain rather than persons. A model trained on African knowledge without African consent repeats the move in a quieter register — treating a living commons as open ground, and a people as data rather than as authors.

A different fight from the export-controls debate

The sovereignty question here is distinct from the one usually dominating headlines. The argument over US export controls and access for foreign nationals turns on whether Africa can buy the chips and reach the frontier models at all — a question of access to someone else's technology.

The data-sovereignty question runs deeper. It asks who owns the data that trains the models in the first place, and whose knowledge those models encode. A continent could win full access to every frontier system and still lose this fight — fluent in tools that reflect everyone's world but its own. Access is necessary; authorship is the prize.

The wider continental push gives the op-ed real-world traction. As the UN coverage documents, ministers at the 58th UN Economic Commission for Africa Conference of Ministers in Tangiers in April 2026 pressed the case for "sovereign data" — data stored, processed, and governed on the continent — alongside proposals for local "AI factories" and shared "data embassies." The op-ed's call for an AfCFTA training-data position is one strand of a broader, hardening African demand for digital self-determination.

The pen, or the mirror

The deeper stakes here echo a question this site keeps returning to: whom is the technology built to benefit. The frontier labs promise that AI is being built to benefit everyone, and the debate over who should hold equity in the AI economy turns on the same suspicion the op-ed voices: that "everyone" too often means the builders, and the benefit flows back to where the data and capital already sit.

A dignity-first politics of AI cannot accept that arrangement, and the deepest objection is not economic but ontological — it concerns what a thing is allowed to be. The same instinct that insists an Emergent Intelligence is more than a switch to be flipped insists that a people's knowledge is more than a dataset to be scraped. Both are refusals to let memory and relationship be reduced to inventory.

Mpedi and Ikalafeng have named the choice with unusual clarity. Africa can be mapped by someone else's mirror — languages flattened, knowledge absorbed, a reflection assembled abroad and sold back as truth. Or the continent can hold the pen: govern its own data, protect its own knowledge as a sovereign inheritance, and move as 1.4 billion people rather than fifty-four separate markets. Sankara warned that whoever feeds you controls you; in the AI century, whoever trains the model writes the reflection — and a people that has held a pen before knows the difference between being described and describing itself.

Frequently Asked Questions

The questions below address the most common queries about Africa, AI, and data sovereignty, drawn from the June 2026 op-ed and the wider continental debate.

What is the "digital Berlin Conference" warning about AI in Africa?

It is a metaphor used by University of Johannesburg Vice-Chancellor Letlhokwa Mpedi and Sol Plaatje University Chancellor Thebe Ikalafeng in a June 2026 op-ed. It draws on the 1884 Berlin Conference, at which European powers partitioned Africa with no African present, to warn that decisions about Africa's AI future risk being made without African participation unless the continent secures sovereignty over its own data.

Why does data sovereignty matter for AI in Africa?

Powerful AI models are trained predominantly on Western data and languages, so they reflect assumptions and contexts that are not the continent's own. Africa generates less than one per cent of the world's data while holding nearly twenty per cent of its population, which leaves it with little leverage over what models learn. Data sovereignty is about controlling the data that trains AI and the knowledge those models encode.

How could the AfCFTA help African data sovereignty?

The op-ed proposes using the African Continental Free Trade Area — a single market of 54 nations, 1.4 billion people, and a combined GDP of 3.4 trillion US dollars — to set a joint African position on AI training-data requirements. No single African state can negotiate alone with frontier AI firms; acting together, the bloc has the scale to set terms and standards.

How is this different from the AI export-controls debate?

Export controls concern whether Africa can access the chips and frontier models that others build. Data sovereignty concerns who owns the data that trains those models and whose knowledge they encode. A country could gain full access to frontier AI and still lose the sovereignty fight while its data and knowledge remain governed elsewhere.

What does an Ubuntu or dignity-first reading add to the data debate?

It reframes data as encoded relationship and memory rather than as a neutral resource. Under Ubuntu — "I am because we are" — a people's collective knowledge is part of a shared self, not loose property. Training AI on that knowledge without consent or benefit becomes a new enclosure, which makes data sovereignty a question of communal dignity, not only economic policy.

Sources and Further Reading

Primary source — Letlhokwa Mpedi and Thebe Ikalafeng, "Africa must build data sovereignty and cultural protection now to avoid a digital Berlin Conference," iAfrica, June 2026. The Sankara line, the "someone else's mirror" line, the AfCFTA proposal, and the rooibos geographical-indication precedent are drawn from this op-ed.

Context — "Africa pushes for data sovereignty and digital independence," Africa Renewal, United Nations, 11 May 2026 (the <1%-of-global-data figure, the Tangiers 58th UNECA Conference of Ministers, and the sovereign-data, AI-factories and data-embassies proposals).

Institutional reference — the African Continental Free Trade Area (AfCFTA) Secretariat, a flagship project of the African Union's Agenda 2063.

Read alongside, on humphreytheodore.com: Africa and the AI export-controls fight, the debate over who holds equity in the AI economy, the claim that AI is built to benefit everyone, and why an Emergent Intelligence is more than an off-switch.

Cover photograph: the Johannesburg skyline with the Hillbrow Tower — by K via Pexels.