Who gains, across the income distribution
Change in average income in each income group, as a percent of that group's income before AI. Each bar splits the change into labor income and capital income; the dot is the total.
Groups are percentiles of each economy's own income distribution. Capital income includes the profits paid to equity owners (wage-markdown profits and product-markup profits).
Where output goes
Shares of output before and with AI.
Markdown profits are the gap between labor’s resource value and wage payments; markup profits come from firms pricing above cost.
Wages by occupation
Percent change in each wage decile's wage. Deciles rank occupations by their pre-AI wage.
Under the hood inequality measures, returns, and every parameter your answers set
Inequality and returns
Technology and adoption by wage decile
From the controls to the equations
For occupation group j, the task data give automation potential dj and time-saving potential ej on the remaining tasks. Questions 1–2 change these profiles while keeping each average fixed, using occupation production weights ηj. The selected task tier determines both profiles.
αj = α0 + (1 − α0) λAdj; sj = λPej; Gj = 1 / (1 − sj); mj = cPsj.
αj is the share of tasks for which automation is feasible. Firms choose the fraction θj to automate and the fraction zj of remaining tasks to augment. Gj is the augmentation productivity multiplier; mj is the capital required per unit of augmented task output.
The wage-sharing setting ξ determines the wage markdown before the employer-power adjustments:
τjξ = 1 − (1 − τj0)Gjξ − 1; τj = min{0.95, Mjτjξ}; wj = (1 − τj)ωj.
ωj is labor’s marginal resource value. The firm’s labor payment is wj = (1 − τj)ωj. Adoption uses the value of a worker’s time:
Automation: R < ωj / ψj0; Augmentation: Rmj < [ωj / ψj0](1 − 1/Gj).
Here R is the capital rental and ψj0 is baseline labor productivity. Employment is fixed within each occupation. At fixed capital, a larger markdown shifts income from workers to owners and leaves task allocation unchanged. It can change adoption through saving, capital accumulation, and the resulting resource values. At ξ = 1, the wage-sharing rule preserves the baseline markdown. Question 4 adds multipliers Mj for deciles 1–4 and 9–10; the middle four deciles use their geometric mean. The central answers set both multipliers to one. Question 3 sets equity participation χ, with an optional exact value under Advanced. The paper presets use these same equations and calibration.
How the model works
Tasks
Workers in ten wage groups perform tasks. Automation uses capital to perform a task; augmentation combines workers with AI capital to save time. Firms compare capital costs with the value of labor time freed for other tasks. We use O*NET task data, the exposure measures of Eloundou et al., and the updated automation ratings of Hosseini and Lichtinger. Questions 1–2 change how these opportunities vary across wage groups.
Wages and profits
Workers receive a fraction of labor’s resource value. The remaining fraction becomes markdown profit. Redistribution changes saving and capital accumulation. The wage-sharing setting connects markdowns to augmentation; question 4 adjusts employer power by wage group. Firms also charge a product markup that generates profits equal to 3 percent of output.
Ownership
Capital, AI, and both kinds of profit are held as risky equity that only some households own; the rest save in safe bonds. Equity returns, borrowing, and saving determine the distribution of assets and capital income. Question 3 sets the share with access to equity.
The comparison
We build on Moll, Rachel, and Restrepo (2022), keeping their household and asset-market calibration. Each result compares the calibrated pre-AI economy with a long-run equilibrium under your assumptions. Advanced settings and the paper presets show the technology, wage-sharing, and participation inputs.