The short version
Some piece of work can now be done without a person. Everyone jumps straight to whether the people currently doing it lose their livelihoods, and the honest answer is that the task itself tells you remarkably little. Automation acts on tasks. Livelihoods belong to people, and the two come apart in both directions.
Elevator operators were eliminated, near enough completely. Bank tellers grew in number right through the arrival of the machine built to replace them. Telephone operators went from one of the most common jobs in America to a rounding error. Same century, same kind of technology, three entirely different endings.
The elevator problem is settling the question by reaching for whichever of those three agrees with you.
Where the name comes from
Not from a textbook, and not, strictly, a term of art. The underlying error already has a proper name: the lump of labour fallacy, which the British economist D. F. Schloss identified in 1891 in a paper called “Why Working-Men Dislike Piece-Work.” The fallacy is treating the total amount of work in an economy as a fixed quantity — a lump — so that any task handed to a machine is a job permanently subtracted from the human supply.
The elevator earns the label anyway, because the story holds both halves of the trap and because its dates are inconvenient for everybody.
Automatic elevators worked from around 1900. Otis announced Autotronic, the first fully automatic high-speed system, in 1948, and completed the first installation in Dallas in 1950. But capability was never the bottleneck. The bottleneck was that passengers would not step into a box with nobody in it. Otis shipped a recorded voice, Elevoice, in 1948 — it welcomed you aboard and scolded you for holding the doors, “Please let the doors close. You are delaying service.” — and deliberately installed it in elevators that still had attendants, two years ahead of the attendant-free system, to get people used to being spoken to by the building.
What finally broke the stalemate wasn’t a better elevator. In September 1945, some 15,000 New York elevator operators, doormen and porters walked out. Around 1.5 million people couldn’t get to work and the city reportedly lost more than $100 million. Building owners suddenly had a reason to remove a role the technology had been capable of removing for four decades. The operators’ union had been seventeen thousand strong in 1920. A second strike in 1950 changed nothing at all — buildings simply went automatic, and the trade ceased to exist.
What it looks like
- Automatic elevators will end the operator's trade
- ATMs will end the bank teller
- Direct dialling will end the switchboard operator
- Correct. Eliminated inside about fifteen years
- Wrong for thirty years — then correct, for a different reason
- Correct, and faster than nearly anyone expected
The teller is the case everybody quotes, so it’s worth having straight. ATMs rolled out across America from the 1970s; by 2010 roughly 400,000 were installed. Over the same stretch, teller employment rose, from around 500,000 to nearly 600,000. The economist James Bessen worked out why, and the mechanism is specific: ATMs cut the tellers needed per branch from about 21 to 13, which made a branch cheap enough that banks opened far more of them — urban branches up 43%. Fewer people per branch, many more branches, and the total went up. The work changed too. Less counting cash, more sitting with customers and their problems.
That is the most-cited reassurance in the whole automation debate. It is also almost always told as though the story stops in 2010.
It doesn’t. Teller employment peaked around 2007, shed more than 107,000 positions by 2017, stood near 442,000 in 2019, and was down to roughly 248,000 by 2024. The Bureau of Labor Statistics projects a further 13% decline through 2034. The prediction everyone mocks for being wrong turned out to be right, thirty years late, about a technology nobody in 1975 was watching — not the cash machine in the wall, but the phone in your pocket closing the branch entirely.
And the switchboard shows how bad it can get. In 1950 there were 342,000 operators at Bell and perhaps a million more on private boards in offices, hotels and factories — about one in every thirteen working women in the country. Direct dialling spread from the 1930s; Bell was under 200,000 by 1940, down to 40,000 across telecoms by 1984. The BLS now counts fewer than 2,000 telephone operators.
Why it doesn’t work
The lump of labour fallacy is genuinely a fallacy, and the standard rebuttal to it is sound. The quantity of work in an economy is not fixed, cheaper output can mean more demand rather than less, and new trades appear that nobody could have named in advance. All true. All of it, note, a claim about aggregates.
Which is where the comfort quietly stops applying. ”The economy will be fine” and “you will be fine” are different sentences, and the history only supports one of them. The elevator operators did not go on to have a fine time in aggregate; they lost a trade. Someone else, elsewhere, later, got work that hadn’t existed before. The total was fine. The operators were not the total, and telling them about the total would have been a strange thing to do.
Run the trap in the other direction and it fails just as badly. “This task can be automated, so the job goes” is exactly what people said about tellers, and it ignores what actually happened: automating the expensive part of running a branch made branches cheap, and cheap branches wanted staff. Cutting the cost of one task can raise demand for the person who does the rest of them. You cannot get from a task to a headcount without knowing what the change does to cost and what cost does to demand — and the person making the confident prediction usually knows neither.
Then there is timing, which is what the elevator makes unforgettable. The capability arrived around 1900 and the trade ended in the 1950s. Nothing about the machinery changed in between that explains the gap. What changed is that people got used to riding alone, and then somebody acquired a reason to force the issue. Capability decides what is possible. Trust and incentive decide when. A confident schedule for an occupation’s death is very often a claim about institutions wearing a claim about technology.
This is not the elevator problem
- Expecting a particular job to change. Tellers still exist and do noticeably different work than in 1970 — less cash handling, more advising. “This occupation is about to be transformed” is a far more defensible claim than either “doomed” or “safe,” and it’s usually the true one.
- Pointing out that a trade really was destroyed. Elevator and switchboard operators aren’t myths, and saying so isn’t fallacious. The trap is the leap from one case to every case.
- Worrying about your own position. The aggregate argument was never intended as personal consolation and doesn’t work as any. Assessing your own exposure is reasonable.
- Arguing about the rate. Whether this wave moves faster or reaches further than previous ones is a real empirical question. It simply isn’t settled by anecdote in either direction.
The question worth asking instead of reaching for an example: what does this do to the cost of the work, what does that do to demand for it, and who has a reason to push adoption through the resistance?
How to head it off
- Ask for the mechanism, not the precedent. Anyone can produce a supporting anecdote; there are at least three and they disagree. Tellers grew for a statable reason — cheaper branches, therefore more branches. If nobody can name the equivalent mechanism in the case at hand, they are pattern-matching and calling it analysis.
- Keep the occupation and the worker separate. Both get called “jobs” and they behave differently. An occupation’s headcount can hold steady while nearly everyone in it is replaced by different people with different skills, which is excellent news for the statistic and no news at all for the incumbents.
- Ask who has to be convinced. Fifty years of working technology sat largely unused because passengers wouldn’t ride alone, and a strike, not an invention, ended that. Look for who bears the adoption cost, who would have to be reassured, and who profits from reassuring them.
- Check where the story you’re being handed stops. The ATM reassurance is true through about 2007 and false afterwards. A historical case quoted only up to its high-water mark isn’t evidence. It’s editing.
Two of the three occupations here no longer meaningfully exist. The third grew for thirty years and has since halved. The lesson is not “it’ll be fine” and it is not “you’re finished” — it’s that the task won’t tell you, the timing won’t cooperate, and whichever anecdote just got handed to you, somebody chose it.