My tradition on this blog is to take a break (mostly!) from current events in the later part of August. Instead, I pre-schedule daily posts based on things I read during the previous year about three of my preoccupations: economics, editing/writing, and academia. With the posts pre-scheduled, I can then relax more deeply when floating on my back in a Minnesota lake, staring up at the sky.
______
One of the classic examples of jobs lost to automation is the case of telephone switchboard operators. As late as 1950, there were about 350,000 women working as switchboard operators working for phone company, and maybe another million working as switchboard operators at offices , factories, hotels, and apartments. Roughly one of every 13 working women was a switchboard operator. Of course, now the number of switchboard operators is nearly zero. The example is often given to point out that in a dynamic economy, even when hundreds of thousands of jobs are “lost,” workers do manage to transition to new jobs.
But that basic story lacks detail. James Feigenbaum and Daniel P. Gross have been digging into two aspects: 1) What happened to the women who were displaced from switchboard operator jobs; and 2) for AT&T, what determined speed and timing of investing in automation to replace switchboard operators?
Feigenbaum and Gross tackle the first question in “Answering the Call of Automation: How the Labor Market Adjusted to Mechanizing Telephone Operation” (Quarterly Journal of Economics, August 2024, 139: 3, pp. 1879–1939). They focus on the period between 1920 and 1940, with data on 3,000 cities. During this period, over 300 cities switched to mechanized switchboards. You can then compare the labor market in cities that switched sooner, later, and not at all. during this period. You can then compare the labor market for women across these different situations. They find:
As a first step, we show that after a city was cut over to mechanical [switchboard] operation, the number of 16- to 25-year-old women in subsequent cohorts employed as telephone operators immediately fell by 50% to 80%. These jobs made up around 2% of employment for this group, and even more for those under age 20—and given turnover rates, this shock may have foreclosed entry-level job opportunities for as much as 10% to 15% of peak cohorts. The effect of this shock on incumbent operators was to dis- possess many of their jobs and careers: telephone operators in cities with cutovers were less likely to be in the same job the next decade we observe them, less likely to be working at all, and conditional on working were more likely to be in lower-paying occupations. In contrast, however, automation did not reduce employment rates in subsequent cohorts of young women, who found work in other sectors—including jobs with similar demographics and wages (such as typists and secretaries), and some with lower wages (such as food service workers). … Though wage data for this era are more limited, using available data we also do not find evidence that local labor markets re-equilibrated at significantly lower wages.
The stability of both employment rates and wages is consistent with demand growing for these categories of workers in other sectors of the economy—and, in turn, with the predictions of Acemoglu and Restrepo (2018) , who suggest that firms will endogenously develop new uses for labor when automation makes it abundant. Buttressing this interpretation, our evidence indicates some occupations expanded to new sectors of local economies after cutovers—that is, the emergence of new work (Autor et al. 2024 ). Taken together, these results suggest that although existing workers may be exposed to job loss, local economies can adjust to large automation shocks over medium horizons.
The overall message is a conventional one for economists. Yes, labor markets do get disrupted, sometimes severely. There is a case here during the transition for government programs to provide some combination of unemployment insurance and training for other jobs. But the ultimate answer is growth of other employment opportunities.
Feigenbaum and Gross discuss the automation of switchboard operations from the perspective of AT&T in “Organizational and Economic Obstacles to Automation: A Cautionary Tale from AT&T in the Twentieth Century” (Management Science, published online in advance of being assigned to a specific issue, February 27, 2024). They point out a puzzle of timing: mechanical call switching technology is invented in the 1880s. However, AT&T doesn’t install the first dial telephones until 30 years later at the Chesapeake & Potomac Telephone Co. in Norfolk, Virginia, in 1919, and the process of phasing out all of the switchboard operators isn’t completed until 1978. Why did it take so long?
Telephone systems were initially designed to have operators physically connecting calls—a task known as “call switching”— putting them at the center of both the telephone network and AT&T’s production system. Manual switching, in turn, shaped choices and activities across the business, including service offerings, plant and equipment, operations, prices, accounting, billing, customer relations, and more.
Although manual switching served early telephone networks well, expansion revealed its limits, as its complexity rose quickly in large markets with billions of possible connections, and switchboards became system bottlenecks. As AT&T grew, its service quality thus fell, and operator requirements exploded: by the 1920s AT&T was the largest U.S. employer, with operators over half its workforce. Company records show the limits of manual switching were known as early as the 1900s, when automatic technology was already being tested—yet it took AT&T several more decades to adopt it widely. We show in this paper that automation was hindered by interdependencies between call switching and the rest of AT&T’s business: the mechanization of call switching required complementary innovation and adaptation across the firm, which were only resolved over time.
In retrospect, one wonders if AT&T would have moved faster to mechanical switching if it wasn’t a monopoly! But there is also a more charitable lesson here. Big innovations require widespread organizational adjustment. Such adjustments often require a substantial up-front investment–some of it monetary, some of it an investment in organizational change in business practices. There are firms and government agencies out there that are still adjusting to information technology and the web, and just starting to make widespread use of tools that have been around for a decade or more. Even if the new artificial intelligence innovations turn out to be the greatest thing since sliced bread, it will take years and decades for them to filter through the economy–indeed, after discovering a new technology, one often has follow-up discoveries of additional uses for that technology.
