Whose Realities Count When AI Decides? How algorithms re-shape labour, agency and governance in the Global South

AI does not descend from the cloud. It is produced through mines, cables, data centers, outsourcing firms, and annotation platforms, by people in places under conditions of power.

That was the grounding argument of the panel hosted by the Global Center on AI Governance at the AI & Equality Festival of Ideas, moderated by Selamawit Engida Abdella. The shared question: whose realities are shaping AI systems, and whose are being shaped by them? 

Dr. Rose Nakasi opened with a concrete account of what it means to design for African realities in healthcare AI. Her work at the Makerere AI Health Lab produced Ocula, an AI-powered diagnostic platform for microscopically diagnosed diseases including malaria and cancer. Ocula attaches to existing microscopes in public health facilities through low-cost 3D-printable adapters and runs on standard smartphones, offline, without requiring continuous internet connectivity. The design choices reflect the infrastructure constraints of the settings Ocula is built for, and the conviction that effective AI tools for the African context need to be trained on locally representative data that captures the pathogen morphologies, population diversity, and resource realities of the continent.

Emsie Erastus brought a political economy lens to the question of who is setting the terms for AI on the continent. Drawing on her work as head of African Voices for Women in AI Ethics and her research on cosmologies of internet infrastructure, she traced the historical continuity between colonial extraction and contemporary data flows. The undersea cable routes that carry African data today follow the same paths as the telegraph cables laid by British imperial interests in the 1800s. The mining sites powering AI hardware in the DRC occupy the same territories as earlier resource extractions. The pattern, she argued is structural. On the policy side, she pointed to the growing friction between tech companies pushing AI as economic development and civil society organizations insisting that human rights and innovation are inseparable.

Joanita Najjuko, speaking from her work with the Nawi Afrofeminist Collective, named what dominant AI narratives consistently leave out: the material conditions that make these systems function. Content moderators in Nairobi watching deeply violent content for hours. Data annotators labeling faces, weapons, and bodies for poverty wages. Women market traders displaced by platform economies that can sustain losses small businesses cannot. The gendered distribution of care work that excludes women from participating in platform economies during peak earning hours. These are the conditions under which AI is built and deployed. A Panafricanist feminist lens, she argued, makes visible what the dominant framing is structured to hide: that AI sits within a long history in which Africa has been treated as a site of extraction, and that the task is to question the economic system itself, not simply to include more people within it.

Dr. Adio-Adet Dinika highlighted the Data Workers Inquiry and the Possible Futures series at the Distributed AI Research Institute. The inquiry, drawing on Walter Rodney’s workers inquiry methodology, treated data annotators, content moderators, and labelers as the experts on their own conditions, and invited them to explain those conditions in their own forms: podcasts, animations, documentaries. These workers are making cultural judgments, interpreting ambiguity, translating lived worlds into training categories, and yet the industry is structured to erase them as both workers and knowers. The Possible Futures series extends that argument into imagination itself, asking what AI governance, healthcare, education, and justice systems could look like if communities set the terms from the beginning, grounded in ancestral knowledge systems and organized around reciprocity rather than extraction.

Across four distinct bodies of work, from diagnostic AI in Ugandan health facilities to feminist political economy to data worker conditions to speculative futures, the panel insisted that answering the question “whose realities count when AI decides?”  requires naming the structures that make certain kinds of labor invisible, holding platforms and states accountable for the conditions they create, and reclaiming the imagination of possible futures from the small number of actors who have currently captured it.