aceaggregate total factor productivity gains from AI task automation over the next 10 years
Source article: The simple macroeconomics of AI
SUMMARY This paper evaluates claims about the large macroeconomic implications of new advances in Artificial intelligence (AI). It starts from a task-based model of AI’s effects, working through automation and task complementarities. So long as AI’s microeconomic effects are driven by cost savings/productivity improvements at the task level, its macroeconomic consequences will be given by a version of Hulten’s theorem: Gross Domestic Product (GDP) and aggregate productivity gains can be estimated by what fractio…
Contested: both sides are scored from claims and sources, not community votes.
"Time series of the growth factors (salinity, nutrients, illumination, temperature) for the three Gracilaria tikvahiae growth regime clusters during a global nuclear winter and overall total growth factor (product of the 4 factors)" by Authors of the study: Florian Ulrich Jehn, Farrah Jasmine Dingal, Aron Mill, Cheryl Harrison, Ekaterina Ilin, Michael Y. Roleda, Scott C. James, David Denkenberger is licensed under CC BY 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/.
This peer-reviewed paper models AI's macroeconomic impact as task-level automation and complementarity, using Hulten's theorem to translate the fraction of tasks impacted and average cost savings into GDP and TFP effects. Using existing exposure estimates, it calculates no more than a 0.66% TFP increase over 10 years, then revises down to less than 0.53% after accounting for the shift from easy-to-learn to hard-to-learn tasks.
The modest productivity finding matters because it tempers claims of large near-term AI-driven growth, while the distributional analysis matters because it suggests gains will accrue disproportionately to capital rather than reducing labour income inequality. Uncertainty remains about how to measure hard-to-learn tasks lacking objective outcomes and how to account for new AI-created tasks with negative social value like online manipulation algorithms.
- Paper uses a task-based model and Hulten's theorem to link micro task-level cost savings to macro GDP and productivity.
- Initial macro estimate is capped at 0.66% TFP increase over 10 years using existing exposure and productivity improvement estimates.
- Author argues early evidence comes from easy-to-learn tasks, while future impacts involve hard-to-learn tasks with context-dependent decision-making.
- Revised forecast lowers expected TFP gains to less than 0.53% over next 10 years.
- On distribution, AI is predicted to widen the gap between capital and labour income and shows no evidence of reducing labour income inequality.
AI automation of tasks is estimated to produce modest aggregate gains of no more than 0.66% TFP growth over 10 years based on task exposure and task-level cost savings.
Those TFP gains are likely exaggerated and even more modest, predicted to be less than 0.53% over 10 years, because future AI effects will involve hard-to-learn tasks with many context-dependent factors and no objective outcome measures.
The rundown
The paper formalizes AI effects through automation and task complementarities, applying Hulten's theorem to aggregate task-level savings into GDP and productivity estimates.
It finds AI's impact is more equally distributed across demographic groups than previous automation, so it is unlikely to increase inequality as much, but also finds no evidence it will reduce labour income inequality and predicts a widening capital-labour gap.
It flags that some new tasks created by AI may have negative social value, such as the design of algorithms for online manipulation, complicating welfare assessment of measured productivity gains.
Estimates may be overstated because they extrapolate from easy-to-learn tasks to hard-to-learn tasks where learning is limited by context dependence and lack of objective outcome measures.
Sources
- Peer-reviewedEconomic Policy2024-08-06
How should this claim be treated?
The debate