Adji Bousso Dieng warns of a new digital imbalance in Africa

‘AI colonization of Africa’ (public domain image, Wikimedia Commons).

Credits: Carl Lender (Flickr user CLender), Sunrise, USA / Wikimedia Commons — CC BY 2.0.

In an interview published April 5 by RFI, Senegalese researcher Adji Bousso Dieng, a member of the UN’s new international scientific panel on artificial intelligence, warns of a possible “digital colonization” of Africa. The phrase is strong, but it is not just a slogan. It points to very concrete mechanisms: tools designed elsewhere, local data poorly represented, and economic value that often flows off the continent.

Put more simply, the question is not only whether Africa will use AI. It is under what conditions it will use it, who will set the rules of the game, and who will capture most of the benefits.

What the Phrase Really Covers

When Adji Bousso Dieng speaks of “digital colonization,” she is not saying that every foreign technology is suspect. She is describing a power imbalance. Today, a large share of models, cloud infrastructures, and technical standards used worldwide are designed and financed outside Africa, mainly in the United States and China. The continent then risks remaining a market, a deployment ground, or a labor reservoir, rather than a decision-making center.

This point matters for both the public and policymakers. An AI system is never neutral. It always reflects choices: which languages it understands well, which uses it prioritizes, which risks it tolerates, which data it values. When an administration, a bank, a hospital, or a company adopts a tool designed elsewhere, it buys a service but also imports an implicit worldview.

The issue is therefore not theoretical. It touches very concrete matters: access to credit, quality of public services, translation, speech recognition, education, health, or content moderation. If the strategic layers of AI remain controlled off-continent, dependence becomes not only technical but also economic and political.

The researcher’s warning does not come from nowhere. The UN General Assembly adopted resolution 79/325 in August 2025, which set the terms for the new international scientific panel on AI. The United Nations website indicates that the panel’s first plenary meeting took place on March 3, 2026. In other words, the debate on power effects related to AI has now entered the most visible international arenas.

Why Data Matters So Much

The second point raised by Adji Bousso Dieng concerns data. An AI learns from corpora. If those corpora mostly contain Western references, dominant languages, and institutional contexts far from African realities, the results will mechanically be less accurate. The problem is not only technical. What is underrepresented in the data often ends up being misunderstood, poorly served, or misrepresented.

This is already visible on very simple issues: accent recognition, processing low-resource languages, translation, or understanding less standardized administrative environments. An AI trained for banking, legal, or medical uses specific to wealthier economies does not transplant without friction. It may function, but poorly. And that poor fit becomes normalized if nobody corrects it.

Artificial intelligence never learns in a vacuum: it is trained in a language, within a context, and according to a hierarchy of references. This image accompanies the section on data biases and highlights a core difficulty: when African realities are poorly represented in corpora, they risk being misinterpreted by systems. The problem is therefore not just technical but tied to how a continent is represented, recognized, or marginalized.

African institutions have begun to respond to this challenge. In July 2024, the African Union approved its continental strategy on artificial intelligence, with a central idea: develop an African approach rooted in local needs, computing capacities, datasets, and training. In April 2025, the Global Summit on AI in Africa held in Kigali also highlighted the need to better coordinate continental efforts, notably around common governance and a future Africa AI Council.

The signal is important, but let’s not kid ourselves. Between a continental strategy and true technological sovereignty, the gap remains vast. Research must be funded, useful corpora opened, data protected, infrastructures developed, talent retained, and actors supported who can scale. That is where the credibility of the discourse is at stake.

The Invisible Work Behind the Most Visible Tools

The third mechanism is perhaps the most telling. AI does not rest only on chips and algorithms. It also relies on human labor: annotating images, classifying texts, filtering violent content, correcting responses, testing models. This workforce is often invisible, yet indispensable.

The Kenyan case marked the global debate. In 2023, an investigation by Time magazine showed that employees of Sama in Kenya had worked for OpenAI on classifying toxic content to make ChatGPT safer. The investigation mentioned pay below two dollars an hour for some of those workers, as well as repeated exposure to very violent texts. This case does not sum up the entire African digital economy, but it illustrates a reality: a continent can be integrated into the AI value chain through the hardest tasks without controlling profits, intellectual property, or governance rules.

Behind the most visible AI tools lie repetitive, grueling, and sometimes poorly paid tasks without which models would not function. This image accompanies the Kenyan case documented by Time and points to that discreet, often invisibilized part of digital production. Technological modernity loses its triumphant certainty here, revealing what it owes to human labor.

The phrase “digital colonization” takes a very concrete meaning here. Africa does not only provide uses and markets. It can also provide data, human compute time, and execution labor. As long as it has little influence over system design, standards setting, and value distribution, the risk of structural dependence remains.

What This Changes for Leaders and Decision-Makers

For public or private leaders, the issue is far from abstract. Choosing a cloud provider, an AI model, an annotation vendor, a data-sharing framework, or a speech recognition tool already means choosing part of your future dependence. The real question is not: should we adopt AI? It is: how to adopt it without surrendering all bargaining power?

The answer is not isolation. It is not about cutting Africa off from global tech flows. It is about entering them with more leverage: more locally trained talent, more quality data governed locally, more research, stronger contractual requirements, more transparency on models used, and greater coordination between states. Without that critical mass, national initiatives risk remaining interesting but fragile against already established giants.

Being connected, producing data, or providing labor does not yet mean having a say in decision-making. This image accompanies the shift in focus toward the value chain. The real dividing line is less between those who use AI and those who lack access than between those who consume it and those who can still set its rules.

At bottom, the value of Adji Bousso Dieng’s warning is to put power back at the center of the conversation. AI is often discussed as an inevitable technical advance. It is also an organization of knowledge, labor, and value. For Africa, the challenge is therefore not simply to enter the era of artificial intelligence. It is to enter it while having greater influence over its uses, norms, and benefits.

Adji Bousso Dieng, Senegalese researcher — ‘I Support a Pan-African AI’

This article was written by Yoann Pantic.