For workers ages 22 to 25, the results were substantially different. Since 2022, employment among that group in the 40% of occupations most affected by AI declined by roughly 11%. In the remaining 60%, where AI exposure was lower, employment increased by about 10%.
The researchers based their analysis on a large sample of anonymized payroll information from ADP. They assessed occupations using an existing measure of potential AI effects on the labor market as well as Anthropic’s Economic Index, which tracks how Claude is being used across different types of work.
The data also points to hiring as the primary source of the divergence. The researchers found that lower recruitment rates, rather than an increase in firings or workers leaving their jobs, accounted for most of the employment decline among young people in AI-exposed occupations. The effects were also more apparent in the number of people employed than in their wages.
How AI is used within an occupation appears to matter. Anthropic’s Economic Index distinguishes between “automative” activity, where AI performs work previously done by a person, and “augmentative” activity, where the technology assists someone who remains responsible for the task.
Occupations including accountants, auditors, receptionists and information clerks were among those with greater exposure to automative AI use. Chief executives and registered nurses, by contrast, were among the occupations where augmentative use was more prevalent.
The employment results followed a similar division. Entry-level employment was weaker in occupations where automation-oriented uses were more common, while jobs characterized by AI augmentation produced less consistent declines. “The findings are consistent with automation-oriented uses of AI substituting for labor while complementary uses are associated with flat or rising employment,” the researchers wrote.
The study also examined whether the type of knowledge required for a job could help explain why younger workers are experiencing different outcomes. The researchers distinguished between formal knowledge that can be documented and taught through education or written procedures and knowledge developed through experience, practice and mentorship.
Using education requirements from the O*NET occupational database as a proxy, they found slower entry-level employment growth in occupations that rely more heavily on codified knowledge. Work involving more tacit knowledge showed stronger employment growth among mid-career and senior workers.
Education produced another distinction. Occupations with larger proportions of college graduates had smaller differences between jobs with high and low AI exposure. Among occupations with fewer college graduates, employment was growing in less-exposed fields while declining in those facing greater AI exposure.
The updated findings do not show the same employment decline spreading evenly throughout the workforce. Instead, they indicate that the effects identified so far are disproportionately concentrated among younger workers attempting to enter occupations where AI is more capable of automating existing tasks.
Lead researcher Erik Brynjolfsson said the persistence of that pattern has increased his concern about what it could mean for people beginning their careers. “The entry-level effects we’re measuring are real, persistent and widening,” Brynjolfsson told The Washington Post, “and I’m more worried than I was about a labor market that keeps its overall employment level while quietly closing the on-ramp for people starting their careers.”
This analysis is based on reporting from ars TECHNICA.
Image courtesy of Resume Professional Writers.
This article was generated with AI assistance and reviewed for accuracy and quality.
