Truace
Subscribe
44/100

Emerging evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Economists' model projects AI narrowing the wage gap, but adoption costs erase most pay gains

Study title: Task-Specific Technical Change and Comparative Advantage

Evidence type
Benchmark or lab evaluationPreprint, not yet peer reviewed
Sources
1 sourcePeer-reviewed
Published
3 min read
Truvace record
TRV-2026-2481cite or check it

Study title: Task-Specific Technical Change and Comparative Advantage

Two economists, Lukas Althoff of Stanford and Hugo Reichardt of CREI, built a dynamic model of the US labor market in which AI can augment, automate or simplify tasks. In their moderate AI scenario, the ratio of 90th- to 10th-percentile wages falls by 24 percent and average wages rise by 28 percent.

Read the study abstract

Artificial intelligence (AI) reshapes workers’ comparative advantage by altering the tasks they perform and the skills those tasks require. We develop a dynamic task-based model to quantify the general-equilibrium effects of task-specific technical change. Workers have multidimensional skills, choose occupations, and accumulate skills on the job; occupations combine tasks, and productivity depends on how workers’ skills match task requirements. We develop a computationally efficient procedure to estimate the mod…

58Gain side

How the two sides scoreClose call

58Problem side

The gain

In a model-based projection for the moderate AI scenario, the 90-10 wage ratio falls 24 percent and average wages rise 28 percent, with the largest welfare gains going to the lowest-skill workers.

The problem

In the same model, skill-dependent adoption costs cut the projected average wage gain from 28 percent to 0.3 percent, and AI lowers the returns to skills, with verbal skills losing the most value.

A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.

What the scores mean

G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.

P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.

In brief

The equalizing force is simplification: when AI lowers the skills a task requires, lower-skill workers can compete for jobs that were previously out of reach. Automation and augmentation have small distributional effects but drive most of the average wage increase.

The results depend on adoption. When the model includes adoption costs, which are highest for lower-skill workers, the 90-10 ratio still falls by 15 percent but average wage gains shrink from 28 to 0.3 percent. These are model projections across slow, moderate and rapid scenarios, not measured outcomes.

Main points

  1. The model was estimated on NLSY79 panel data on wages, occupations and skills, combined with O*NET data covering 19,530 tasks linked to 974 occupations.
  2. AI's task-level capabilities were estimated by prompting several AI models (Qwen 2.5, gpt-oss, GPT-4o); in the moderate scenario, generative AI saves 25.2 percent of worker time and 22.4 percent of tasks are automatable.
  3. Welfare gains for current workers in the moderate scenario equal a rise in lifetime wages of 15 to 45 percent, largest for the lowest-skill workers and most variable among older workers.
  4. Administrative occupations such as financial clerks see a large decline while construction and manufacturing occupations become more important; occupations with the biggest employment gains often see the largest relative wage falls.
  5. Mapped to college majors, engineering, mechanics and architecture gain value while law, political science and philosophy lose value.
  6. Because simplification reduces the incentive to build skills, average skill levels fall, which dampens AI's positive growth effects.
  7. In US job postings, occupations predicted to be more simplified by AI show a decline in required expertise of 0.4 standard deviations by early 2026 after the ChatGPT release.

The full story

The authors add simplification to the standard automation and augmentation channels of task-based models. Workers have five-dimensional skills (manual, math, social, technical, verbal), choose occupations each period and learn on the job by performing tasks that exceed their current skills. They estimate the model by maximum likelihood on NLSY79 data following workers from age 25 to 2022.

To quantify AI, they used AI models to rate task-level skill requirements across 35 skills, with 683,550 queries per model. These ratings correlate 0.79 to 0.93 with O*NET occupation-level measures and 0.70 with the average of two human experts on 200 skill-task combinations for economists. Slow, moderate and rapid scenarios follow earlier forecasting work.

Beyond wages, AI changes what skills are worth. Because workers build skills by doing demanding tasks, lowering requirements slows skill accumulation. Occupations with the largest employment gains tend to be those whose relative wages fall most, since simplification enlarges the pool of qualified workers and increases competition.

Adoption has been slower among lower-skill workers, so the authors added skill-dependent adoption costs estimated from current adoption data. These costs dampen but do not eliminate the decline in inequality. The authors conclude that lowering adoption costs is a key policy lever for raising wages, especially among lower-skill workers.

What this doesn’t fix

The results come from a structural model run under slow, moderate and rapid scenarios for AI capabilities, which are described as uncertain and evolving; they are projections rather than observed wage changes, and the paper is a working paper not yet peer-reviewed.

Sources

1 on file
  1. NBER Working Paperdoi:10.3386/w35353
    Peer-reviewed2026-06-01

Talk it over

Four questions to start onThis record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

Start with a question

Economists' model projects AI narrowing the wage gap, but adoption costs erase most pay gainshttps://truvace.com/item/task-specific-technical-change-and-comparative-advantage

  1. This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

    The gain In a model-based projection for the moderate AI scenario, the 90-10 wage ratio falls 24 percent and average wages rise 28 percent, with the largest welfare gains going to the lowest-skill workers.

    The problem In the same model, skill-dependent adoption costs cut the projected average wage gain from 28 percent to 0.3 percent, and AI lowers the returns to skills, with verbal skills losing the most value.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a benchmark or lab test, from NBER Working Paper. It is a preprint, not yet peer reviewed.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The results come from a structural model run under slow, moderate and rapid scenarios for AI capabilities, which are described as uncertain and evolving; they are projections rather than observed wage changes, and the paper is a working paper not yet…

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Labor