If you were a young (white) woman looking for work in the early 1920s, you could do worse than becoming a telephone operator.
In the early 1920s, AT&T, the telephone monopoly that grew out of Alexander Graham Bell’s Bell Telephone, was America’s largest employer, and specifically employed many women as operators, who manually connected callers by plugging wires into inputs on switchboards. In 1929, when employment peaked in the last months before the Great Depression, a government report estimated the number of operators working for AT&T at 161,669.
The company refused to hire Black operators until 1944, immigrants rarely got hired, and some exchanges barred Jewish women too. But for white, gentile, American-born women, especially young and unmarried women (as women often left the labor force after marriage, and married women faced discrimination in hiring), connecting calls at the switchboard was a common way to make a living.
“In 1920, telephone operators were roughly 2 percent of the US female workforce and 4 percent of nearly three million young, white, American-born working women,” economists James Feigenbaum and Daniel Gross observe. “As much as 15 percent of cohorts born at the turn of the century might have ever been an operator.”
Then it all went away. In the 1920s, as telephone coverage was expanding and the ranks of operators were growing, AT&T began to roll out a “mechanical switching” system in which people would manually dial other numbers from their home, using a rotary system. Human operators were no longer needed. The profession took decades to die out completely, as AT&T switched gradually, exchange by exchange. But eventually, automation killed off the telephone operator as a profession, by around 1978.
That’s what made telephone operators so interesting to Feigenbaum and Gross, two economic historians who wanted to examine a clear case where automation led to an entire job class being automated away.
For existing operators, they find that automation had real costs. Operators in a city that transitioned to mechanical switching were substantially less likely to have any job 10 years later than operators in cities that were slower to automate; those that did find work tended to find worse, lower-paying jobs.
But Feigenbaum and Gross also examine the results for young white women coming of age during automation, who just a few years earlier would’ve been ideal candidates for telephone operator jobs. Remarkably, they find little or no negative effects at all: they were just as likely to find work as they would have been before, and job openings in fields like secretarial work and restaurants increased even as telephone operation was automated away. Some of those jobs (like restaurant work) paid less, but others were competitive with telephone operation.
This is just one case, and economists have a long way to go in understanding how automation affects workers — a question that is more important than ever with the rapid progress in AI. But telephone operation appears like a mostly heartening example. Even though a job that once employed 2 percent of all working women was automated away, new workers entering the labor market were not significantly worse off.
The curious case of the completely automated job
Of course, automation leading to job losses in a particular job category or whole sector of the economy is pretty common. As websites like Expedia and Kayak and Google Flights emerged, the number of travel agents in the US fell from 100,000 in 2000 to 45,000 last year, even as the working population grew by 29 million people. From 1948 to 2019, a recent Department of Agriculture report found, the amount of labor on US farms fell by 74 percent, while the output of those farms grew by 175 percent. We grew nearly three times as much foodwith a quarter of the labor because of intensive investment in advanced combine harvesters, fertilizers, and other innovations.
But that didn’t eliminate the need for farm workers, and travel agents still exist (in fact, the Bureau of Labor Statistics expects the number of travel agents to grow rapidly in the next decade as part of the travel industry’s overall recovery from Covid). It’s pretty rare for a job to be fully automated out of existence the way telephone operators were. The economist James Bessen, for instance, has argued that since 1950, only one job (elevator operators) has ever been fully automated away. The Bureau of Labor Statistics still estimates some 4,000 people working as telephone operators, though their work is highly specialized and very different from that of early 20th-century women on switchboards who saw their jobs swept away.
“Jobs are bundles of tasks,” Gross told me. “We had a job that was defined by one task: call-switching. … Part of why there aren’t as many examples of entire categories being eviscerated is that most jobs have workers doing multiple things.”
My job as a reporter, for instance, can be divvied up into many individual tasks: scheduling calls with sources, conducting interviews, transcribing those interviews, conducting online research and reading past coverage and academic papers, collating all of the above into a final article. Even if one of those tasks is automated (as transcription largely has been in recent years), the rest remain. Most jobs, from janitorial labor to factory assembly to medicine and law, are like this: complex combinations of discrete tasks, and the job itself doesn’t vanish if one task is automated.
The technology to automate call-switching emerged in the 1890s, only 16 years after Bell’s invention of the telephone. Almon Strowger, an undertaker in Kansas City, Missouri, developed the so-called “Strowger switch,” the first electric system for connecting phone lines without a human operator.
A possibly apocryphal but extremely funny origin story alleges that Strowger was inspired to invent his switch because he thought the operator at the local telephone exchange, who was married to a rival undertaker, was conspiring to divert calls from bereaved families to her husband instead of Strowger. I haven’t been able to source this to anything other than a series of poorly footnoted books and articles but I like the anecdote too much to leave it out. It also seems to fit later anecdotes from people who knew Strowger and attested to his … difficult … temperament.
In any case, switching failed to take off in the 1890s. It didn’t offer clear cost savings over human operators, and it produced more errors. It wasn’t until 1917, Feigenbaum and Gross note in a companion paper, that “mechanical switching could match manual operation on connection times and error rates, and internal estimates suggested it may generate savings in large cities.”
A major factor was the exponentially rising complexity of telephone networks as more and more people got phone lines in their homes and workplaces. “It only takes 50,000 subscribers to have a billion possible pairwise connections,” Gross said. “Adding a 50,001st subscriber adds another 50,000 potential connections. Having the mechanisms to connect that many different people manually is incredibly costly and complicated.” While human operators had managed this complexity for a few decades, it beggared belief that they could handle a country where every home had a phone.
Automation proceeded in stages, city by city, and with important limits. Initially more complex tasks, like long-distance switching, were reserved for human operators even in cities that transitioned to mechanical switching. The Great Depression slowed investment in mechanical switching systems, as did restrictions on non-military uses of copper imposed during World War II. (Copper was the main material for phone lines). The full transition to mechanized call switching only ended in 1978, Feigenbaum and Gross observe, at which point computerized switching systems far more complex than anything Almon Strowger imagined were beginning to be implemented.
The staggered rollout is a godsend for economists: they let Feigenbaum and Gross compare employment outcomes for young white women before and after AT&T transitioned to mechanical operation in a given city, and by combining these before/after comparisons in the 261 different cities they examine through 1940, and roughly 2,500 additional cities which were not yet converted to mechanical service, they can estimate an average effect of the transition.
What automation did to existing telephone operators — and those who would’ve taken their place
Transitions to mechanical switching led, unsurprisingly, to a dramatic reduction in the share of young, white, US-born women working as operators: in cities instituting the change, the share fell by 1.7 percentage points, which is a huge change given that on average 3.9 percent of this group was working in telephone operation before automation.
Operating was a relatively high-turnover job; among operators in cities that didn’t transition to mechanical switching, only 24 percent were still operators 10 years later. But the share was even lower in cities that automated: only 16 percent stayed in the field (presumably moving to cities or exchanges that hadn’t yet been automated). A large share of operators who dropped out of the profession post-automation didn’t find other work at all. Older operators (meaning those who were over 25 when automation occurred) were 7 percentage points less likely to be working, which Feigenbaum and Gross note accounts for “more than half of the displacement of operators in this age group.” They had no future in telephones, and most of them got booted out of the labor force entirely.
Those who kept working tended to get worse jobs. About 10 percent of operators exposed to automation were in a lower-paying profession a decade later, compared to only 1 percent of operators not exposed to automation.
So that’s the bad news: getting hit head-on by a wave of automation had serious negative effects on these women. But what about women coming of age in the 1930s who might have earlier been telephone operators? Were they worse off for lacking this job opportunity? Surprisingly, Feigenbaum and Gross find the answer is no.
“We find no effects on the fraction of young women working, in school, married, or with children for any group,” they conclude. This is true even after they narrow their analysis to white, American-born women, and down to relatively narrow age bands (16 to 20, say, or 21 to 25). What appears to have happened is that other professions open to young women with just a high school diploma saw job opportunities increase as those in telephone operation were shrinking. Secretarial work, for instance, boomed, as did restaurant work.
“This is the era of the drugstore lunch counter, the soda fountain,” Gross says. “There’s growing demand in this broad line of work in new places.”
The idea that demand for young women in the workforce surged in these industries exactly enough to offset the jobs lost to automation in telephony seemed almost magical to me. It’s such a neat story, and a such hopeful one for automation generally.
One possible story is that the spread of the telephone, enabled by automated switching, led to increased productivity elsewhere in the economy which enabled more hiring in positions like secretarial labor. Secretaries spend a lot of time on the phone, after all. That’s not what seems to have happened, though. “We don’t really think there are any kind of direct productivity impacts of the technology outside of AT&T itself,” Gross says. “If there are, they’re minuscule, too small to explain these effects.”
So what did happen? The closest thing to an answer we have is that the overall economy adapted. Moving to mechanical switches didn’t reduce the total amount of spending in the economy. The money that used to pay operators’ salaries, the money AT&T made from telephone bills and then spent on wages, was still there, and it went to something. Moreover, the presence of a sudden glut of young women available to work gave businesses a reason to try out what Feigenbaum and Gross call “organizational innovations”: new ways to structure their firm to make use of these female workers. Around this time, doctors and hospitals had begun hiring “medical stenographers” to take down symptoms and other information from patients, in person or via phone. None of the tech behind that job was new, but the availability of young women to do it was new.
“There’s a time dimension that’s really important,” Feigenbaum says. “If you’re an incumbent worker, the technology shock is bad for you. If you’re a future worker, you have time to adjust.”
Feigenbaum and Gross are hesitant to draw overly broad conclusions from this work for the whole economy. “We’d need to study 10 more, 100 more automation events to really know how, this phenomenon operates,” Feigenbaum says. “Are there some cases where the other jobs are not growing at the same time?” It’s possible. We just don’t know.
But the ability of the next generation of female workers to adapt to the telephone automation shock gives me some hope as we face a new wave of automation led by AI. Of course, sufficiently general AI threatens to automate vast swathes of tasks at once, quite quickly, without giving us much time to transition. If that happens, rapid job loss seems inevitable. But it hasn’t happened so far, and smaller shocks like mechanical telephone operation seem more common.
The telephone operators’ example gives me some reason to think the next generation of would-be truck drivers, or radiologists, will be able to sort into new work. And maybe, if we’re lucky, we can avoid existing drivers getting hurt the way existing telephone operators were.