To start, a quick clarification. It might be the day before (a very, very strange) Thanksgiving, but this year at Broadstreet, you won’t find any seasonal content. Next year, I’m going to try and convince the fellow editors to write up “How to Argue About Politics and Economics with Your Relatives… And Win Using History” but this year, you’ll just have to pretend you’re having connection problems on Zoom while you speed-read on Wikipedia to impress your cousins with your encyclopedic knowledge of esoteric topics.
Rather than thinking about our futuristic Thanksgivings, I want to think today a bit about the future of work. Pandemic and health anxiety are clearly having their moment, but automation anxiety has not left us. As AI improves and computing gets ever cheaper, scholars and pundits (and cranks) have estimated what share of jobs could be automated or replaced by machines or software. Predictions range: the OECD calmly predicted that fewer than 10% of American jobs were at “high risk,” but PwC said 38%. McKinsey thinks half of jobs are already automatable globally.
There’s even a website oracle to ask if (and when) robots might take your job: https://willrobotstakemyjob.com/. Economists are at 43% and I should start worrying but many of my fellow Broadstreet editors will be delighted to know Political Scientists face only a 4% risk and Sociologists a 6% risk. This suggests these “models” know nothing about how labor markets or academic research fields actually work, but that’s not my main point here.
How worried should we really be? I remember Larry Katz, one of my PhD advisors, quipping that when imagining the future of work, he’d rather study history than science fiction. Me too. So what does history tell us about automation?
In a new working paper with Daniel Gross, we study the effects of automation on telephone operators. The telephone industry was huge in the early 20th century, serving 15 million telephones and connecting more than 65 million calls per day. The “market” was dominated by AT&T and its operating companies.
The setting, we think, is a nice reminder that while industrial robots in manufacturing are important, that is far from the only setting where we’ve seen automation (and could see more of it), well before AI and machine learning and smart-whatevers.
For a half-century starting around 1920, AT&T replaced operators with mechanical switching and dial service. Operators went from looking like this:
By 1940, over 60% of the AT&T network was already dial. My coauthor Daniel (and probably some of you technology nerd readers…) was excited to note just how long the diffusion process of the dial technology was and that it followed a classic S-curve pattern.
But what about the effects of automation? In the 1920s, before the shock, telephone operation was one of the most common entry-level jobs for young women (especially young, white, US-born women) and employed hundreds of thousands of people. As such, our focus in the paper is on the effects of the technology shock on the incumbent women workers and the next generation—the women who in the years and decades after the automation could have become operators if only those jobs had still existed. To do this, we draw on census data and a longitudinally-linked sample of women.
The technology had bite: after a city went dial, the number of young women in subsequent cohorts who were telephone operators immediately and permanently dropped by 50-80%. This abruptly eliminated around 2% of employment for this subset of the population.
Yet, despite the magnitude, the shock did **not** reduce future cohorts’ employment rates. Instead, we think that some other jobs grew to take its place. These alternatives included typists and secretaries, which required similar skill levels and paid similar wages. But another was lower-skill, lower-wage service jobs in restaurants or beauty parlors. Working women of the youngest ages (16 to 18) tended to be in these lower-paying jobs, while women in their 20s were in the similar-paying jobs. On the whole, both groups seemed to have fared okay, at least in terms of employment.
But this only describes how *future generations* adapted to a labor market where telephone operation was automated. What about the women who were operators just before mechanical switching was adopted? At some point, I’ll spend a post (or two or twenty) on census linking (I have many thoughts), but to make a long story much shorter, linking women census to census (and across changing marital status) is hard and we think we have a clever way to do it.
Comparing the outcomes of women who were telephone operators in cities that were vs. weren’t automated, relative to those of same-age, precisely demographically-similar working young women from the same neighborhood, we find that shocked-incumbent operators were subsequently less likely to be working. Conditional on working, the incumbents were more likely to have switched to lower-paying occupations. But the magnitudes of these effects are tempered by the fact that in the early 20th century, many women exited the workforce as they aged (and married and had kids). But telephone operator was one of the few jobs for women with the potential to be a career, and the loss could be costly for those who would have chosen to keep them.
Collectively, our results suggest local economies can adjust to automation shocks over relatively short horizons and continue to absorb the steady stream of young workers entering the labor market. The incumbent workers who get replaced may be most at risk.
If you’ve made it this far, I have hopefully convinced you that economic history can help us think about our technological future. But if you are a dedicated Broadstreet reader, you probably have a different question: what does this have to do with historical political economy? Or, rather, the history and economy are here but where is the political? This will be more speculative—it isn’t in the paper, so don’t blame my coauthor—but I think the case of telephone operation automation connects with questions today.
More or less, the soon-to-be-replaced operators themselves were left to the mercy of their benevolent employers at AT&T. Despite PR noises about retraining them, we find very few incumbents are still working in the telephone industry in non-operator jobs (and few find work as operators in other industries like department stores or office buildings or hotels). Instead, it appears the jobs were eliminated and the women left to fend for themselves. A sign of the times comes from the Indianapolis Star in 1925 which reported on the automation in an article titled “Cupid Beats Out Old Man Hunger.” The sexist punchline: don’t worry about the telephone operators, as 25 of the 250 women were soon to be married (allegedly nine on the Saturday immediately after the automation Friday night!). Empirically, we find no evidence that marriage (or child-birth) went up for displaced operators.
What about policy responses? With hundreds of thousands of operators and a slow diffusion of dial, did public policy (eventually) address the technologically-displaced workers? We have a clue about the lack of policy imagination from a Joint Committee report from 1956 on “Automation and Technological Change.” The top policy recommendation was, essentially, hope the economy is growing and dynamic enough to absorb any workers displaced by technology:
We actually find some evidence for that in the telephone operator example: cohorts of young women in labor markets hit with automation during the Great Depression fare a lot worse than those adjusting to automation before or after the downturn.
Today, are we doing better? There are a few very general policy responses; for example, however politically unrealistic it is, Andrew Yang ran an entire political campaign on the universal basic income as a shield against robots but UBI backers tend to see it as the solution to a million problems, not just automation. But in the face of a flurry of predictions about the technological replacement coming for a few workers/some workers/all workers (depending on your sources), what is the plan? I’m not sure. I suppose we could hope the economy is growing (but don’t we always do that?). What else? I think we’re better off looking to history than to science fiction but it seems a bit of an open question about what policies have been used or used successfully to respond to technology shocks in the labor market.
2 thoughts on “Adjusting to Automation”