Mapping AI Job Risk Across 3 Countries: A Leontief Cascade Model
I started with Karpathy's open-source project and built a Leontief input-output cascade model on top of it - then extended it to the US, France, and Singapore using real government labor data.
I started with Karpathy's open-source project. Then I mapped AI job risk across 3 countries.
Andrej Karpathy published a treemap scoring 342 US occupations for AI exposure - sized by employment, colored by risk. Brilliant visualization. But it only measures direct exposure.
I wanted to answer a harder question: what happens to jobs that aren't directly automatable, but depend on ones that are?
So I built a Leontief input-output cascade model on top of it. Same framework economists have used for 80 years to trace how shocks propagate through supply chains - applied to AI displacement.
Then I extended it to three countries using real government data:
US - 342 occupations, 143M workers
- 49M jobs at direct AI risk
- 77M at cascade risk - 28M more once you model the domino effect
- 54% of the workforce in the high-exposure bucket
France - 225 DARES occupations, 28.4M workers
- 4.3M direct, 8.9M cascade
- 31% of the workforce at high risk
Singapore - 149 SSOC occupations, 2.5M workers
- 1.33M jobs at cascade risk - 54% of the workforce
- Includes Singapore's advanced industrial robotics context (ports, fabs, MRT)
The real insight isn't any single number.
AI risk doesn't stop at your job description. It flows through every sector your industry buys from and sells to.
Caveat: these are stress-test upper bounds, not forecasts. Exposure does not equal displacement. The cascade model assumes full propagation without labor market adjustment - a scenario tool rather than a crystal ball.
Built with React + Vite, real government labor data, and Leontief I-O math. Slide the adoption slider from 0% to 100% and watch the risk propagate. Find your occupation. See what's connected to it.
Live visualizations (all open-source and free):
- US: raafet57.github.io/jobs/us-ai-cascade-risk
- France: raafet57.github.io/jobs/france-ai-exposure
- Singapore: raafet57.github.io/jobs/singapore-ai-exposure
Source code:
- Original: github.com/karpathy/jobs
- My fork: github.com/Raafet57/jobs
Topics: AI, Future of Work, Data Visualization, Open Source, Labor Economics