Economic Scenarios for Transformative AI
S · Shapes the field Task-based macro model by Anton Korinek and Charles I. Jones, reviewed by Acemoglu, Autor, Restrepo and others, that puts the full range of published AI growth forecasts into one comparable framework with a 10,980-person US survey.
Summary based on the full text·Original: about 75 minutes to read
In short
Anthropic Institute Working Paper 2026-02 (September 2026) by Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter and Peter McCrory. It builds one task-based model of the US economy that converts a few assumptions about AI (share of tasks affected, adoption, gain per task, automation versus augmentation) into paths for GDP, wages, the labor share and unemployment to 2030. Three scenarios cover the published range: modest (+1.6% GDP), substantial (+8.3%) and extreme (+32.4%). In every scenario output rises, cognitive employment shrinks and wages outside knowledge work rise; the scenarios differ in size and speed.
Key numbers
- 60% to 45%labor share of income by 2030 in the extreme scenario
- 17.9%unemployment rate of workers who began in cognitive occupations, extreme scenario, 2030
- +8.3%GDP above the no-AI path in 2030, substantial scenario; the median US adult's answers imply +8.6%
- ~9% of GDPtransfer that would hold cognitive workers' income at its no-AI level in the extreme scenario
As reported by the source.
English·PDF·opens www-cdn.anthropic.com
What it reports
- Modest: AI touches 4% of task instances by 2030; GDP +1.6%, growth 2.4% a year instead of 2%, labor share 60.0% to 59.4%, unemployment 3.9% against a normal 3.8%.
- Substantial: 12% of task instances (about 20% of 2025 cognitive tasks); GDP +8.3%, growth 5.4% a year (the 1999 dot-com peak was 4.7%); average wage +2.1%, cognitive wage -0.3%, other occupations +5.9%.
- Substantial: labor share 60% to 56% in four years, close to the entire US decline over four decades after 1980; cognitive unemployment 2.9% to 4.5%.
- Extreme: GDP +32.4%, growth 15.4% a year in 2030; cognitive wage -11.5%, other wages +33.6%; cognitive employment -21.5% since mid-2026; overall unemployment 11.9%.
- In the extreme case total labor income is almost unchanged (0.45 x 1.32 = 0.60); capital income is 81% above its no-AI path.
- AI speeding up research adds well under one percent to labor productivity in all scenarios, because research stays bottlenecked by physical tasks.
- Almost all divergence between scenarios comes after 2027.
- With very inelastic capital (elasticity 1), the average wage falls 9.2% below the no-AI path in the extreme case; with fully flexible wages, cognitive wages fall 42.2% but cognitive unemployment is 2.6%.
- Survey: 10,980 US adults via Morning Consult, 11-23 August 2026; medians imply GDP +8.6% and unemployment 4.6%.
How they got it
A CES task model with two occupation groups (cognitive: management, professional, sales, office; all other), an upward-sloping capital supply, a semi-endogenous ideas channel and a search-and-matching labor market. Base year 2024, labor share 0.6, task elasticity 0.5. Survey answers from 3,259 respondents who completed all five items were run through the same model.
Limits
- Scenarios carry no probabilities; key parameters (capital supply elasticity, wage rigidity) are assumed, and results swing widely with them.
- No robotics, no demand-side feedback, no business cycles; workers differ only by occupation group, so displacement costs are understated.
- Published by an AI developer whose commercial case depends on AI being consequential; the companion explorer describes the substantial scenario more aggressively than the paper.
Why it matters for companies
The model gives managers a shared vocabulary for an otherwise unbounded debate: four dials (tasks affected, adoption, gain, automation share) that can be estimated inside one company. It predicts that the pressure lands on management, professional, sales and office roles first, with wage pressure or unemployment depending on how flexible pay is. Workforce planning to 2030 should test at least the modest and substantial paths, because the paper finds they look the same until 2027. Reskilling into non-cognitive roles is the adjustment mechanism the model assumes, and it is slow.
Summary based on the full text. Read the original for the full argument and data.Open PDF on www-cdn.anthropic.com