> **来源：[研报客](https://pc.yanbaoke.cn)** # Summary of "Labor Market Impacts of AI: A New Measure and Early Evidence" ## Core Content This document presents a new measure of AI displacement risk, called *Observed Exposure*, which combines theoretical AI capability and real-world usage data to assess the impact of AI on labor markets. The authors analyze how AI usage is distributed across different occupations and examine its effects on employment and hiring trends. ## Key Findings - **Observed Exposure**: A new metric that evaluates the extent to which AI tasks are being automated in professional settings. It weighs automated uses more heavily than augmentative ones and considers work-related contexts. - **Theoretical vs. Actual Use**: AI has not yet reached its theoretical capability. Only a fraction of tasks that could be automated are actually being performed by AI, indicating a gap between potential and current implementation. - **Occupational Impact**: Jobs with higher observed exposure are projected to grow less over the next decade. The most exposed professions include computer programmers, customer service representatives, and financial analysts. - **Worker Characteristics**: Workers in highly exposed occupations are more likely to be older, female, more educated, and higher-paid. - **Unemployment Trends**: There is no systematic increase in unemployment for highly exposed workers since late 2022. However, there is suggestive evidence that hiring of younger workers in exposed occupations has slowed. - **Job Finding Rates**: Young workers (22–25 years old) are less likely to start new jobs in highly exposed occupations, indicating a possible early impact on hiring patterns. ## Methodology The authors combine three data sources to measure AI exposure: - **O*NET Database**: Lists tasks for over 800 occupations. - **Anthropic Economic Index**: Tracks real-world AI usage. - **Eloundou et al. (2023)**: Provides theoretical AI capability estimates, rating tasks on a scale of 0 to 1 based on feasibility and speed improvement. They calculate *Observed Exposure* by weighting tasks based on their level of automation and their relevance to professional work. The metric is then aggregated to the occupation level. ## Analysis of Exposure and Job Growth - **BLS Projections**: Jobs with higher observed exposure are projected to have weaker employment growth from 2024 to 2034. - **Regression Analysis**: For every 10 percentage point increase in AI coverage, the BLS's growth projection decreases by 0.6 percentage points. - **Task Coverage**: The red area in Figure 2 (observed exposure) is much smaller than the blue area (theoretical capability), showing that AI is not yet widely adopted across all tasks. ## Worker Characteristics and Exposure - **Demographics**: Workers in highly exposed jobs are more likely to be female, white, and Asian, and have higher education levels. - **Education Levels**: Graduate degrees are more common among highly exposed workers (17.4%) than among those with no exposure (4.5%). - **Earnings**: Highly exposed workers earn significantly more, on average, than those with no exposure. - **Labor Market Outcomes**: Higher exposure is associated with more hours per week, higher hourly wages, and lower union membership. ## Unemployment Trends - **Unemployment Rates**: No significant change in unemployment for highly exposed workers since late 2022. - **Difference-in-Differences Analysis**: The unemployment rate gap between high and low exposure groups has remained stable, suggesting no major displacement effects yet. - **Young Workers**: There is a slight decline in job finding rates for young workers (22–25) in exposed occupations, though this is not statistically significant. ## Limitations and Future Work - **Data Limitations**: Job transitions may be difficult to measure accurately in surveys. - **Future Improvements**: The authors plan to incorporate more usage data and update theoretical metrics as AI capabilities evolve. - **Focus on Young Workers**: Further research is needed to understand how recent graduates in AI-exposed fields are navigating the labor market. ## Conclusion The study introduces a novel framework for measuring AI's labor market impact, emphasizing the distinction between theoretical capability and actual usage. While no significant unemployment effects have been observed yet, there are early signs of reduced hiring for young workers in AI-exposed occupations. The authors hope that this approach will help identify economic disruptions more reliably as AI continues to diffuse. ## Key Figures - **Figure 1**: Share of Claude usage by Eloundou et al. exposure rating. - **Figure 2**: Theoretical capability and observed exposure by occupational category. - **Figure 3**: Top ten most exposed occupations. - **Figure 4**: BLS employment growth projections vs. AI exposure. - **Figure 5**: Differences between high and low exposure workers. - **Figure 6**: Unemployment trends for high vs. low exposure workers. - **Figure 7**: New job starts for young workers in high vs. low exposure occupations. ## References - Acemoglu, Daron, David Autor, Jonathon Hazell, and Pascual Restrepo (2022) - Appel, Ruth, Maxim Massenkoff, Peter McCrory, et al. (2026) - Blinder, Alan S et al. (2009) - Brynjolfsson, Erik, Bharat Chandar, and Ruyu Chen (2025) - Eloundou, Tyna, Sam Manning, Pamela Mishkin, and Daniel Rock (2023) - Gans, Joshua S. and Goldfarb, Avi (2025) - Gimbel, Martha, Molly Kinder, Joshua Kendall, and Maddie Lee (2025) - Handa, Kunal, Alex Tamkin, Miles McCain, et al. (2025) - Hui, Xiang, Oren Reshef, and Luofeng Zhou (2024) - Johnston, Andrew and Christos Makridis (2025) - Massenkoff, Maxim (2025) - Ozimek, Adam (2019) - Tamkin, Alex and Peter McCrory (2025) - Tomlinson, K., Jaffe, S., Wang, W., Counts, S., & Suri, S. (2025)