Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
AI Task Exposure and U.S. Employment Patterns Across Demographic, Educational and Occupational Groups, 2022–2025
Citation. Chintapalli, S. S. & Shivaram, M. (2026). AI Task Exposure and U.S. Employment Patterns Across Demographic, Educational and Occupational Groups, 2022–2025. American Journal of Student Research, 4(4), 994–1015. https://doi.org/10.70251/hyjr2348.449941015
Summary. This observational analysis combines 2022–2025 IPUMS Current Population Survey aggregates with BLS projections and Anthropic’s occupation-level AI Exposure Index to describe how measured task exposure and employment patterns vary by sex, race/ethnicity, age, education, and occupation. It reports a stable aggregate employment-to-population ratio alongside heterogeneous occupational and demographic patterns, but does not identify AI’s causal effect on employment. The study frames its results as early directional evidence because the observation period overlaps post-pandemic labor-market normalization and the exposure measure is a proxy derived from Claude-use patterns.
Key claims
- Women represented 70% of workers in the study’s AI-displacement exposure category, compared with 46% in the AI-complement category and 44% in the no-exposure category; the reported association between sex and exposure tier was χ²=2799.9, p<0.001.
- Weighted CPS estimates show women’s employment rose from 73.5 million in 2022 to 76.9 million in 2025 and men’s from 81.5 million to 86.6 million, while the employment-to-population gap narrowed from 10.8 to 10.0 percentage points.
- The Gradient Boosting Model had modest discrimination (AUC 0.56); occupation and education were associated with the model’s predictions, but the analysis cannot establish that AI exposure caused employment changes.
- The authors report near-zero projected growth for workers without a college degree and more than 9% projected growth for workers with advanced degrees, while cautioning that BLS projections are not AI-specific effect estimates.
Evidence & limitations
- The analysis uses population-weighted repeated cross-sectional CPS data from 2022–2025, BLS occupational projections, O*NET data, and an Anthropic occupation crosswalk; its model is associational rather than causal.
- The short window may conflate AI diffusion with post-pandemic recovery, the Claude-based exposure index is a proxy rather than direct automation measurement, and occupation-level aggregation can mask within-occupation variation.
- The row-level train/test split can place the same occupation in both sets, creating possible information leakage; the authors recommend occupation-grouped or time-based validation.