Noah Smith
· Noahpinion
· August 19, 2026 at 23:27
· ⏱ 22 min read
| Read on Substack ↗
Summary
Noah Smith argues that a new Acemoglu et al. paper claiming low birth rates boost GDP per worker through labor-saving automation is not convincing: the country-level result weakens with controls, the US wage result likely reflects talent sorting rather than birth rates, and the thesis contradicts Acemoglu's own prior work on automation and wages. For markets, this is a macro/economic-research debate rather than an actionable trade, though it underscores uncertainty about how AI and demographic decline interact.
•Noah Smith says The Economist's article attacking Daron Acemoglu is not newsworthy, but he welcomes criticism of top economists because economics is too hierarchical.
•Acemoglu et al. (2026) claim lower birth rates are associated with higher GDP growth per working-age adult across countries and higher wage growth across US commuting zones, attributing this to labor-saving automation.
•Smith shows the country-level result loses statistical significance after controlling for education, urbanization, and geographic region; the effect weakens as controls are added.
•The paper never includes institutions as a control, which Smith calls a red flag given Acemoglu's own career emphasis on institutions as determinants of development.
•The US commuting-zone result is more significant, but Smith argues it may be explained by talented people and knowledge industries sorting into coastal cities like Boston, NYC, and SF after 1970, not by 1940 birth rates.
•Smith highlights an internal contradiction: Acemoglu's earlier work (Acemoglu & Restrepo 2017) found robots depress wages, and his 2021 'Harms of AI' paper said AI could push down wages and fail to raise productivity, while the new paper relies on automation raising wages.
•Contradicting evidence cited includes Kremer (1993), Romer (1990), Maestas et al. (2023), and Aksoy et al. (2019), which generally associate faster population growth or less aging with faster growth.
•Fernández-Villaverde warns that today's ultra-low fertility is historically unprecedented; a TFR of 1.56 vs 0.75 are qualitatively different regimes, so regression results from the past cannot be extrapolated linearly.