The AI Divide: How Automation Deepens Digital Inequality

Abstract

Artificial intelligence now sits inside the systems that decide who gets credit, who is given which medical diagnosis, and who reaches a public service at all. The gains from it, though, are not shared evenly, and the gap between the countries that build these systems and the ones that only use them has been growing rather than shrinking. This paper treats that gap, which it calls the AI Divide, as a structural problem rather than a passing lag, and it breaks the problem into five parts that feed one another: access to infrastructure, disruption to work that hits developing economies hardest, algorithmic bias that carries old discrimination into new tools, the concentration of computing power in a few hands, and an environmental cost that lands unevenly. The argument draws on a structured review of peer-reviewed studies from 2016 to 2026, read alongside data from international institutions. A few numbers set the scene. Adoption in richer economies now runs about three-quarters higher than in poorer ones. Roughly 92 million jobs are expected to disappear by 2030, mostly where there is least capacity to retrain people. Facial-recognition error rates climb to nearly 35 percent for darker-skinned women yet stay under 1 percent for lighter-skinned men. Three companies hold close to two-thirds of the world's cloud capacity. Because each part of the divide reinforces the others, it tends to widen on its own unless policy intervenes. The paper ends with five governance principles, universal access, data sovereignty, representative development, algorithmic accountability, and international cooperation, and works through what each would mean for Azerbaijan and the South Caucasus, where the problem is already present rather than hypothetical.