Men’s employment grew more than women’s in occupations with greater exposure to artificial intelligence, with the largest gender gaps among workers without a bachelor’s degree, according to a new Pew Research Center analysis.
The report released Thursday compared employment in 2017-19 with January 2023 through June 2026. For each one-point increase in an occupation’s AI exposure score, men’s employment grew by 6.3%, while women’s grew by 0.2%.
The pattern concerns changes in the number of people employed, rather than their share of an occupation. Pew compared men’s and women’s growth within the same occupation, then examined whether the gap widened as AI exposure increased.
The findings show an association between exposure and employment changes. They do not establish that AI caused the differences: Other developments also affected employment between the two periods, Pew said.
What AI exposure measures
Researchers scored 480 occupations using information about their work tasks from the Labor Department’s Occupational Information Network, known as O*NET. The score reflects how much generative AI could perform or assist with important tasks, rather than how extensively employers actually use the technology.
Scores range from zero to 3.5. Zero identifies the least-exposed occupation in the analysis, not a job with no tasks AI could help perform.
Work involving reading, writing and information analysis tends to have higher exposure. Accountants and auditors scored 3.09, software developers 3.10 and financial managers 3.20. Construction laborers scored 0.70, carpenters 0.80 and home health aides 1.14. Medical assistants and bank tellers fell in the middle, at 1.54 and 1.64.
Pew’s model showed the employment-growth gap widening toward the top of the scale. At the least-exposed end, the pattern reversed, with women’s employment growing faster than men’s. Employment in the most-exposed occupations grew overall, rather than shrinking.
Education changes the pattern
Among workers with a high school diploma or less, each additional exposure point was associated with 8.8% employment growth for men and a 9.5% decline for women.
The gap was smaller among people with some college education, a group that includes associate degree holders and people who attended college without earning a degree. Men’s employment grew by 3.6% per exposure point, while women’s fell by 6%.
Among workers with at least a bachelor’s degree, the estimates were much closer: 0.8% for men and 0.7% for women. Neither group’s employment showed a statistically significant relationship with increasing AI exposure.
Pew said the gains for men without four-year degrees in highly exposed occupations apply to a relatively small share of those men. Many instead work in less-exposed fields such as truck driving, construction and moving freight.
Women without four-year degrees are more often employed in jobs such as secretarial work, reception and customer service. Those occupations fall closer to the middle or upper part of the exposure scale, where chatbots and automated support can play a larger role.
How researchers tested the findings
The occupational task information came from O*NET version 30.3. May 2023 federal employment estimates helped combine detailed jobs into the occupation groups used in the household survey.
The analysis used roughly 4 million records from the Census Bureau’s Current Population Survey, weighted to estimate employment. It omitted 2020-22 because the pandemic reshaped employment for reasons unrelated to AI.
Researchers built the exposure measure from 41 work activities. Sixteen activities involving information, communication and advice were classified as highly exposed, while nine involving physical tasks or in-person work were classified as having low exposure. Sixteen mixed activities were left out of that calculation. Scores combined the relative importance of high- and low-exposure activities for each occupation. Researchers subtracted the average importance of low-exposure activities from the average for highly exposed activities, then adjusted the scores so the least-exposed occupation began at zero.
The statistical models accounted for occupation and year differences, with uncertainty calculated to reflect records grouped within occupations. They used pre-2020 employment weights so later changes in job counts would not themselves change an occupation’s weight in the analysis.
Researchers checked their findings against four other AI exposure measures, including one based on how people actually use Anthropic’s Claude at work. Across all five measures, including Pew’s, four showed a gender gap of about 5% to 7% for each standard-deviation increase in exposure. An index focused on image generation did not show a statistically significant gap. Those standardized comparisons account for differences among the indexes’ scoring scales.
FingerLakes1.com is the region’s leading all-digital news publication. The company was founded in 1998 and has been keeping residents informed for more than two decades. Have a lead? Send it to news@fingerlakes1.com.
