AI job losses and women: what labor history means for the Fed

The news

A CNBC analysis published Monday argues that, if past technology shocks are a guide, the spread of artificial intelligence is likely to hit women’s employment harder than men’s. The reasoning rests on occupation: AI is strongest at routine tasks, numbers and pattern-finding, which describes much clerical and administrative work, and women hold most of those jobs.

• Ariane Hegewisch of the Institute for Women’s Policy Research is quoted saying that 90% of secretaries and clerical workers are women.
• The International Labour Organization concluded in March that nearly 1 in 10 women in high-income countries work in occupations at the highest risk from AI.
• The historical record cuts both ways: telephone operators lost jobs and careers, but later cohorts of women found comparable work elsewhere.

Our analysis

The labor market is half of the Federal Reserve’s dual mandate, and the question of who bears AI-related job losses is no longer abstract for the System. The October Beige Book, published October 14, said five Districts reported firms that had reduced staffing through attrition, delayed hiring or anticipated less job creation after boosting productivity through AI investment. Overall employment was little changed in most Districts. In the September 15-16 FOMC minutes, some participants observed that strong demand for skilled workers tied to the AI buildout was driving strong wage gains for those workers, while several participants pointed to unusually low labor market dynamism, with low hiring and layoff rates.

The CNBC piece supplies the historical analogs. The automated telephone switch is the clearest. In the 1940s nearly 95% of US telephone operators were women, and an estimated 75,000 operator jobs vanished in the decades after AT&T’s rollout. Research by economists James Feigenbaum and Daniel Gross found that many incumbent operators ended up in lower-paying jobs or left the workforce. The same research found the next generation of women fared better, finding similar-quality work in other fields.

The washing machine and other household appliances also cut time spent on unpaid housework, making it easier for more women to take paid work. The 1980 census counted about 4.7 million secretaries, stenographers and typists, more than 98% of them women. The IBM personal computer in 1981 and Microsoft Word two years later shifted much of that work to other employees, and many women adapted by taking on new, more skilled tasks. A 2022 study cited in the piece found that office software adoption lowered employment in administrative work by 1% while raising wages for college graduates by 3%.

For monetary policy watchers, the point is that task-level automation and job-level loss are different, and the Fed’s data on aggregate employment may not show the distribution. The FOMC’s September statement described job gains as keeping pace with the workforce and the unemployment rate as little changed.

Room for disagreement

The historical comparison has limits. Earlier automation waves also created new work, and much of the adjustment fell on later cohorts rather than on the workers already in the jobs. Whether AI creates comparable demand for the workers it displaces is an open question, and AI is expected to take on more skilled tasks, not only routine ones, which adds to the uncertainty.

Aggregate numbers also point away from alarm for now. The FOMC minutes describe the labor market as stable and close to maximum employment, and the Beige Book reports adequate labor supply in most Districts, so the pressure is concentrated rather than broad.

What to watch

The signals to track are the next Beige Book, the unemployment rate for women alongside the overall rate, and any FOMC language on AI and hiring. If staffing restraint spreads beyond the five Districts that reported it this month, it becomes a larger factor for the employment side of the mandate. The committee meets October 27-28, and its December meeting brings new projections for unemployment.