Cognitive Inequality in the Age of Generative AI: Who Benefits and Who Falls Behind

Generative AI was supposed to democratize knowledge. The early productivity studies confirmed it: ChatGPT compresses the performance gap between high-skill and low-skill workers (Noy Brynjolfsson et al., QJE 2025). But a parallel body of evidence tells a darker story. An MIT Media Lab study found that regular ChatGPT users showed up to 55% reduced brain connectivity compared to non-users, with 83% unable to recall passages from essays they had just written (Kosmyna et al., 2025). A survey of 666 participants found significant negative correlations between AI usage frequency and critical thinking scores (Gerlich, Societies 2025). And the people who would benefit most from AI — low-income, less-educated, Global South populations — are precisely the ones who adopt it least: 24.7% of the Global North's working-age population uses generative AI, versus 14.1% in the Global South (Microsoft AI Economy Institute, 2026). This paper introduces the concept of "cognitive inequality" — a new dimension of social stratification where AI simultaneously compresses productivity gaps and widens cognitive depth gaps, with access patterns ensuring the benefits flow disproportionately to those who need them least. We synthesize evidence across six domains: cognitive effects, digital access, education, labor markets, cognitive offloading psychology, and counter-arguments. We identify a central paradox that current policy frameworks have not addressed.