AI Innovation and firm productivity
Alderucci et al. present a comprehensive study on quantifying the impact of Artificial Intelligence (AI) on productivity and labor demand using U.S. Census microdata. They authors developed an ensemble of machine learning algorithms to parse the full text of U.S. patent grants and assess their AI-relatedness. The ensemble model was trained on a dataset of 2,000 manually labeled patents (1,000 AI and 1,000 non-AI). Finally, Alderucci et al. conducted an event study using the matched data to gauge the impact of AI innovations on various firm-level outcomes.
Key findings:
- AI Prevalence:
– The study found that AI-related invention is more pervasive than many previous analyses have suggested. - Employment and Productivity:
– AI-related inventions are positively associated with growth in employment.
– There is a positive relationship between AI patents and increases in output per worker (labor productivity growth). - Income Inequality:
– The study found less evidence that AI invention is expanding income inequality within firms. - Broad Impact:
– The productivity growth effects documented in this study are potentially broader in scope than those found in previous experimental literature.
– This suggests the possibility that AI could eventually lead to a significant and persistent acceleration in productivity growth across a broad range of industries.
Quantifying the Impact of AI on Productivity and Labor Demand: Evidence from U.S. Census Microdata
Authors: Dean Alderucci, Sagar Baviskar, Lee Branstetter Nathan Goldschlag, Eduard Hovy, Andrew Runge, Prasana Tambe, Nikolas Zolas
From: Carnegie Mellon University, U.S. Census Bureau, University of Melbourne, Duolingo, University of Pennsylvania, U.S. Department of State
Automation, rents, and labor market inequality
This paper by Daron Acemoglu and Pascual Restrepo examines the effects of automation in economies with labor market distortions that generate worker rents (wages above opportunity cost) in some jobs. The authors developed a task-based model where firms can allocate tasks to workers of different skills or automate them, in addition they incorporate labor market distortions in the form of task-specific wage wedges that create rents for workers. For the empirical analysis they use US data from 1980-2016 to test predictions.
Key theoretical results:
- Automation targets and displaces workers from high-rent tasks, dissipating rents
- This rent dissipation amplifies wage losses for exposed worker groups
- Automation reduces within-group wage dispersion for exposed groups
- Rent dissipation is inefficient and can offset or negate productivity gains from automation
Key empirical findings:
- Automation accounts for 52% of the increase in between-group inequality since 1980
- Inefficient rent dissipation offset 60-90% of potential productivity gains from automation
- Automation increased aggregate TFP by only 0.3-1.3% between 1980-2016 (much less than potential gains)
- Aggregate consumption increased by just 0.45-1.95% due to automation over this period
- Within exposed groups, automation caused more pronounced wage declines between the 70th and 95th percentiles of the wage distribution
The paper provides a novel framework for understanding how labor market imperfections interact with automation, showing that rent dissipation can significantly amplify negative wage effects and reduce potential productivity gains from new technologies.
Automation and Rent Dissipation: Implications for Wages, Inequality, and Productivity
Authors: Daron Acemoglu, Pascual Restrepo
From: MIT, Boston University
