纽约联储-人工智能暴露工人的可再培训性如何_（英）-2025.8_38页_2mb

报告摘要

Summary of "How Retrainable Are AI-Exposed Workers?"

Authors: Benjamin Hyman, Benjamin Lahey, Karen Ni, Laura Pilossoph
Source: Federal Reserve Bank of New York Staff Report, No. 1165, August 2025

Key Findings

Workers in AI-exposed occupations show high positive earnings returns from job training, with an average return of approximately $1,470 per quarter.
AI-exposure trainees outperform low-exposure peers in earnings, but those targeting high-AI occupations face a 29% earnings penalty compared to general skill training.
Training returns are driven by tight labor markets and are amplified when firms recruit from deeper skill pools.
Between 25% and 40% of occupations are "AI-retrainable," meaning workers gain higher pay by moving to AI-intensive roles after training.

Methodology

Uses a dataset of over 1.6 million job training spells from U.S. Workforce Investment Act programs (2012–2023).
Links administrative earnings data with AI exposure measures (Brynjolfsson et al. 2018; Eloundou et al. 2024).
Employs nearest-neighbor matching to compare trainees with control groups receiving only job search assistance.
Analyzes earnings trends across AI skill space and decomposes retrainability using an "AI Retrainability Index (AIR)."

Conclusions

Job training programs effectively support AI adaptation, but workers tend to avoid deepening AI-specific skills.
Evidence suggests stronger training outcomes in favorable labor market conditions.
Findings highlight the importance of retraining in addressing AI-driven labor market changes.
Data includes 1.6 million training spells linked to earnings records and occupational AI exposure scores.
Method uses matching and decomposition to isolate AI-related skill transitions.
Policy Implications stress the role of workforce programs in aiding AI adaptation, with recommendations for program design.