Digital Capital, Artificial Intelligence, and Education in Sub-Saharan Africa

The rapid advance of artificial intelligence (AI) in education outpaces the data needed to assess its impact. Large-scale international surveys do not yet measure schools' actual exposure to AI, making direct causal estimation impossible. This article proposes an alternative strategy-prospective and distributive-to evaluate ex ante the risks of deploying AI in highly stratified school systems. Using PASEC 2019 micro-data from fourteen sub-Saharan African countries, we construct two school-level indices: a digital capital index and a socio-educational vulnerability index. A within-country quadrant typology and a triple-risk indicator reveal that disadvantaged students are overwhelmingly concentrated in schools with low digital capital and high vulnerability-environments where converting AI into learning gains is least plausible. Estimates findings show that digital capital is positively associated with mathematics achievement, vulnerability exerts a penalizing effect, and returns to digital capital vary substantially across schools and territories, with partly compensatory effects emerging in peri-urban areas. Interpreted through the lens of Rawls's difference principle, these findings indicate that educational AI functions as a conditional amplifier: without explicit targeting and complementary investments in conversion capacity, its deployment risks widening existing inequalities.