“Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics,” circulated as NBER Working Paper 24001 in November 2017, is by Erik Brynjolfsson, Daniel Rock, and Chad Syverson. It confronts a genuine puzzle: AI systems were demonstrating ever more impressive capabilities, yet measured productivity growth had slowed and incomes for most Americans had stagnated.
The authors lay out four candidate explanations and weigh each. The first is false hopes, that the technology is simply overhyped. The second is mismeasurement, that real gains are happening but national statistics miss them. The third is redistribution, that some actors capture private value without adding to aggregate output. The fourth, which they argue is the leading explanation, is implementation lags. Like earlier general-purpose technologies such as electricity and the computer, AI delivers its full payoff only after waves of complementary innovation, organizational redesign, and intangible-capital investment, all of which take years and are themselves badly measured.
This paper is the conceptual seed of the “productivity J-curve” the same authors formalized later. For executives it carries a practical warning and a reassurance at once: do not expect a new technology to show up in the productivity numbers immediately, and do not conclude from flat statistics that the technology is failing. The hard, slow work of reorganizing how firms operate is usually what stands between a capable technology and measurable growth.