The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2481
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T07:40:19.515853Z
status: published
lens: g_space
sector: labor
headline: Task-Specific Technical Change and Comparative Advantage
dek: 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…
gain_title: In a simulation model of the US labor market, moderate AI progress cuts the 90-10 wage ratio by 24 percent and raises average wages by 28 percent, with the largest welfare gains for the lowest-skill workers.
problem_title: (none)
trace_subject: (none)
gain_reading: In a simulation model of the US labor market, moderate AI progress cuts the 90-10 wage ratio by 24 percent and raises average wages by 28 percent, with the largest welfare gains for the lowest-skill workers.
gain_evidence: In the moderate AI scenario, the 90-10 wage ratio falls by 24 percent, and average wages rise by 28 percent. | The gains are largest for the lowest-skill workers who benefit disproportionately from simplification.
problem_reading: (none)
problem_evidence: (none)
quick_read: Economists Lukas Althoff and Hugo Reichardt built a dynamic model of the US labor market in which AI can augment, automate or simplify tasks. In their moderate-progress scenario, the 90-10 wage ratio falls by 24 percent and average wages rise by 28 percent, which they attribute mainly to simplification lowering the skill needed for tasks.

The picture changes when adoption costs are included. Because lower-skill workers have adopted AI more slowly, the model has them face higher costs. The 90-10 ratio still falls, by 15 percent, but average wage gains shrink from 28 to 0.3 percent. The authors call lowering adoption costs a key policy lever.

These are model projections across slow, moderate and rapid AI scenarios, not measured outcomes. The model also projects that AI lowers the returns to verbal skills, shifts employment away from administrative jobs toward construction and manufacturing, and slows skill accumulation, which dampens growth.
limitation: The results are model-based projections that assume all workers adopt AI in the baseline, and the model omits regional, racial and gender drivers of wage inequality, so it somewhat underestimates inequality (75th-to-25th percentile ratio 1.84 versus 2.04 in the data).
tag: Evidence-backed gain
key_points: In the moderate scenario, welfare gains for current workers are equivalent to a 15 to 45 percent rise in lifetime wages, largest for the lowest-skill workers. | Simplification, meaning AI lowering a task's skill requirements, drives the narrowing of inequality; automation and augmentation drive most of the average wage rise. | Verbal skills lose the most value while manual and math skills keep much of theirs; engineering, mechanics and architecture majors gain value while law, political science and philosophy lose it. | Administrative occupations such as financial clerks decline sharply, while construction and manufacturing occupations grow in importance. | The occupations with the largest employment gains often see the biggest relative wage falls, because simplification enlarges the pool of qualified workers. | The model was estimated on NLSY79 panel data and O*NET data covering 19,530 tasks across 974 occupations, with AI capabilities scored by several large language models. | In job postings, occupations predicted to be more simplified by AI show expertise requirements falling 0.4 standard deviations by early 2026.
rundown: The model links skill formation, occupational choice and wages. Workers have five skill types (manual, math, social, technical and verbal), choose occupations that bundle tasks, and build skills on the job. It was estimated by maximum likelihood on National Longitudinal Survey of Youth 1979 data, with a new database of task-level skill requirements for 19,530 O*NET tasks.

AI capabilities came from several language models (Qwen 2.5, gpt-oss and GPT-4o) across slow, moderate and rapid scenarios. In the moderate scenario, generative AI saves 25.2 percent of worker time on average and 22.4 percent of tasks count as automatable. The authors report that these measures align with expert ratings and experimental productivity estimates.

Skill formation matters for the results. Because workers learn by doing tasks above their current skill level, simplification slows skill accumulation and lowers average skill levels. This dampens the positive growth effect of AI.

Welfare gains vary most among older workers: some have skills and occupations that happen to match the needs of the AI era, while others do not. Adoption has been slower among lower-skill workers, which is why the authors add skill-dependent adoption costs, estimated from data on current AI adoption.
sources:
- peer_reviewed | NBER Working Paper | https://doi.org/10.3386/w35353 | 2026-06-01
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

7b279b6247693fb62c4bcf78b2021de9735596a0953673b66e15ad9f9fc142e5

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000