The Productivity Paradox Revisited: Artificial Intelligence, Labour Market Restructuring, and the Inequality Trap

The most widely cited figure in the AI-and-jobs debate is a net employment gain. This paper argues that numbers are real and, taken alone, misleading. Using a cross-sectoral synthesis of data from the IMF, McKinsey Global Institute, World Economic Forum, OECD, and peer-reviewed academic sources, three structural problems are identified that the aggregate figure cannot show: a credential barrier that locks displaced workers out of the new roles AI creates; a geographic concentration of AI investment that widens the gap between advanced and developing economies regardless of domestic policy; and a capital-labour distribution asymmetry that routes productivity gains toward asset owners rather than workers. Investment banking is examined as an illustrative sectoral example showing how automation compresses an industry’s entry-level pipeline without eliminating it. The analysis concludes that the central policy failure is not a shortage of reskilling programmes but a mismatch between the pace of AI adoption and the capacity of educational institutions to respond. The analysis bears particular weight for India, where 90% of the workforce is informally employed and where the distance between AI’s aggregate promise and its distributional reality is greatest. Key limitations, including reliance on projections and cross-source comparability constraints, are discussed explicitly.