AI-exposed jobs deteriorated before ChatGPT

TL;DR

Analysis of millions of LinkedIn profiles and university syllabi finds that graduates taking more AI-exposed curricula had higher first-job pay and shorter job searches after ChatGPT, which point to forces pre-dating generative AI and to the ongoing value of LLM-relevant education.

Abstract

Public debate links worsening job prospects for AI-exposed occupations to the release of ChatGPT in late 2022. Using monthly U.S. unemployment insurance records, we measure occupation- and location-specific unemployment risk and find that risk rose in AI-exposed occupations beginning in early 2022, months before ChatGPT. Analyzing millions of LinkedIn profiles, we show that graduate cohorts from 2021 onward entered AI-exposed jobs at lower rates than earlier cohorts, with gaps opening before late 2022. Finally, from millions of university syllabi, we find that graduates taking more AI-exposed curricula had higher first-job pay and shorter job searches after ChatGPT. Together, these results point to forces pre-dating generative AI and to the ongoing value of LLM-relevant education.

Figures (14)

• Figure 1: Unemployment risk for AI (LLM) exposed workers increased starting before the launch of ChatGPT.(A) Unemployment risk for occupations with high exposure compared to occupations with low exposure (see SI Fig. S2 for seasonally-adjusted plot). (B) Unemployment risk for occupations grouped into exposure quintiles. (C) Unemployment risk relative to first quarter 2022 by major occupation. Lines are colored by major occupation's AI-exposure. We compute unemployment risk monthly across states and report quarterly series as the mean of monthly values across all states within each quarter (see SI Fig. S3 for non-normalized version).
• Figure 2: College graduates from 2022 onward spend more time securing LLM-exposed jobs, but poor labor market performance starts months before ChatGPT. We consider graduates first job within three years after completing their terminal degree. (A) Graduate's whose first job had high exposure typically spend less time job seeking compared to peers from the same graduation cohort, but this trend reverses after 2022. (B) As a comparison, graduates whose first job was high salary also experienced increasing job delays after 2022, but delays that were still comparable to those in previous years. In (A) & (B), points show regression-adjusted differences; bars are 95% confidence intervals. (C) For each graduation cohort by month from May of their graduation year, the share of LLM-exposed jobs started compared to the overall share of jobs. Positive values indicate that graduates within a cohort are securing LLM-exposed jobs at a higher rate than less exposed jobs. Vertical lines indicate 95% confidence intervals. Dashed lines indicate the launch of ChatGPT for graduation cohorts 2021 and 2022. (D) Similar to (C) examining each graduation cohort's share of low exposure jobs started each month.
• Figure S1: Number of states with non-zero unemployment claims by major occupation group and month. A state reporting zero claimants may be accurate or may reflect a failure to connect a state's internal occupation classification system with the Standard Occupation Classification (SOC) used by the U.S. Department of Labor.
• Figure S2: Same as Figure \ref{['fig:uiRisk']}A, but seasonally adjusted by quarter. However, note that long-term annual trends, such as the COVID-19 pandemic in 2020 and 2021, are not controlled for.
• Figure S3: Same as Figure \ref{['fig:uiRisk']}C, but trends by major occupation are not normalized to the first quarter of 2022.