The occupation called computer, held by people, and its disappearance

Filed under interpretation in canon/proposals.md, and interpretation is right — but only after the two kinds with a real claim on it have been refused, because both refusals are things a reading needs to know.

Why not moment. The house test for that kind, argued out in eliza-1966 and restated in toy-story-1995, is a date on which something visibly happened in public. There is no such date here, and the absence is not a research failure — it is the most important structural fact about this case. The four candidate dates sit in three institutions and none of them was experienced by anyone as the end of an occupation: 5 May 1958, when Langley's West Area Computing unit was disbanded and its women reassigned to program offices; 7 April 1964, when IBM announced System/360 — the National Bureau of Standards published the Handbook of Mathematical Functions, the definitive tool of hand calculation, within weeks of that announcement, and Grier's line is that "the actual date of publication coincided" with it, so that to the people selling electronic computers the Handbook "seemed an anachronism" on the day it appeared; 1967, when Gertrude Blanch retired and the bureau closed its Computational Laboratory; and about 1970, when the Jet Propulsion Laboratory replaced the job title Computer with Engineer. Grier is explicit that no single ending exists: Blanch "was not the last professional human computer," but "her departure coincided with the final days of many computing offices." An occupation that ends this way ends without a moment. That is worth carrying as such, and moment would hide it.

Why not idea. There is a genuine idea inside this history and it is not this entry. In 1791 Gaspard de Prony organised the Bureau du Cadastre's computing staff on the first chapter of The Wealth of Nations, and his planners wrote worksheets that "identified every operation of the calculation and left nothing for the workers to interpret" — a program, written for people, four decades before Babbage designed a machine to run one. That is implementable, it is built on rather than argued about, and it satisfies the kind's test cleanly. But the canon already holds its consequence in babbage-analytical-engine-1837, and the citation occasion proposals.md attaches to this id is about people and jobs, not about a framing. If a later session wants the computing office as an idea — the division of mental labour, the worksheet, differencing as error control — that is a separate entry and this file deliberately does not pre-empt it.

What interpretation buys. The kind's gloss is a reading of events that a reading may need to name, and what a reading needs from this file is precisely a reading: the claim that an occupation can be eliminated entirely by automation, in an economy that grew, without that being either a catastrophe or a non-event for the people who held it. That claim is contestable, it is contested in the literature, and section 3 grades it against the record — which is what this kind is for and what neither of the other two would have forced.

On the year field. It holds 1964 and it is a marker, not a death date. The entry covers roughly 1758 to 1970. 1964 earns the slot because it is the year in which the two halves of the story appear within weeks of each other: the reference work that human computation produced for its own use, and the machine line that made the whole practice unnecessary. A reading citing this entry should cite the span, not the number.

On descends_from, which is []. Nothing in canon/ precedes this. The edges run the other way — babbage-analytical-engine-1837 (Babbage applied to be a computer at Greenwich in 1814 and was talked out of it, then spent his life trying to mechanise the office he had almost joined), lovelace-1843, hollerith-census-1890, bombe-bletchley-1940 and univac-election-1952 all sit downstream of the human computing office, and none of those files can be edited by this job, so the descent is recorded here in prose and nowhere else. The ancestor this entry actually wants is Adam Smith's division-of-labour chapter of 1776, and there is no Smith entry to point at. That is a gap in the canon rather than in the history.

What it is

For roughly three hundred years, a computer was a person. The word is first recorded in that sense in 1613, in Richard Braithwaite's The Yong Mans Gleanings — "the truest computer of Times, and the best Arithmetician that ever breathed" — and it kept that meaning, without competition, until the middle of the twentieth century. The job was to calculate: by hand, on paper, in quantity, to a schedule, usually as one worker among many on a single large calculation.

Grier dates organised computation from the effort of three French astronomers — Alexis Clairaut, Joseph Lalande and Nicole-Reine Lepaute — to predict the 1758 return of Halley's comet, and follows it to the 1986 return, computed at JPL by one researcher with a FORTRAN IV program on a shared UNIVAC 1108. In between:

**The Nautical Almanac, from 1767.** Nevil Maskelyne's computers worked from home, on piecework, scattered across England. Every table was computed twice by two independent computers and reconciled by a third worker called a comparer — error control by redundancy, before there was a word for it. Mary Edwards of Ludlow worked from the 1780s to 1815 as one of about thirty-five.

The Bureau du Cadastre, 1791–1801. De Prony, ordered to produce trigonometric tables for the Revolution's decimal angle measure, recalled opening Smith and finding the pin factory: his bureau "could manufacture logarithms as easily as one manufactures pins." He built three tiers. At the top, a handful of mathematicians including Legendre and Carnot chose the formulas. In the middle, eight planners reduced each formula to nothing but additions and subtractions and wrote the worksheets. At the bottom, nearly ninety computers, many of them former servants and wig dressers thrown out of work when the Revolution made the styles of Louis XVI unfashionable, most of whom, de Prony reported, "had no knowledge of arithmetic beyond the two first rules." They filled in blanks. The planners checked the output by differencing — taking differences between adjacent values to find the wrong ones — which removed, as one observer put it, the "necessity of repeating, or even of examining, the whole of the work." Lalande, visiting in 1794, reported seven hundred results a day. The nineteen volumes were never published; the publisher went bankrupt and Napoleon's government was not interested.

The Harvard College Observatory, from 1877. Edward Pickering hired more than eighty women over his forty-two years as director, at twenty-five to fifty cents an hour, explicitly because it let him buy more labour per dollar and get through more photographic plates. Williamina Fleming came to the observatory as a maid. Annie Jump Cannon classified spectra at a rate nobody has matched. Henrietta Swan Leavitt, working on variable stars in the Magellanic Clouds, found the period–luminosity relation and published it in a Harvard Circular in 1912; it is the bottom rung of the cosmic distance ladder and the reason anyone knows how big the universe is. She was not permitted to take the next step of the work.

The Mathematical Tables Project, from 1938. A Works Progress Administration relief project in Manhattan, technically directed by Gertrude Blanch, employing some 450 computers hired under rules that required taking people with no skills — so the definitive tables of the American century were computed by workers who had been taught only to add.

The war. "Sometime in 1944, computers became 'girls,'" Grier writes. Warren Weaver's Applied Mathematics Panel called its computers girls; George Stibitz ranked projects in girl-years; someone on the panel coined kilogirl for a thousand hours of computing labour. L. J. Comrie published an article called "Careers for Girls" arguing that girls "can be made proficient and give good service in the year before they (or many of them) graduate to married life and become experts with the housekeeping accounts."

And here is the correction that matters most, in Grier's words, because almost every retelling of this history drops it:

> Even at this date, computing was not the sole domain of women. It was really > the job of the dispossessed, the opportunity granted to those who lacked the > financial or societal standing to pursue a scientific career. Women probably > constituted the largest number of computers, but they were joined by African > Americans, Jews, the Irish, the handicapped, and the merely poor. The > Mathematical Tables Project employed several polio victims as computers, while > the Langley research center kept an office of twelve African American computers > carefully segregated from the rest of the staff.

That segregated office is the one the world now knows as the West Area Computers, created in 1943 under Virginia's Jim Crow laws, supervised from 1949 by Dorothy Vaughan, disbanded on 5 May 1958. Langley's first computing pool — five white women, 1935 — had been meant to be temporary. JPL ran its own group under Macie Roberts and then Helen Ling, who hired only women. Aberdeen Proving Ground and the Moore School at Pennsylvania ran the ballistics-table crews.

The status was fixed and it was low. When the Applied Mathematics Panel wanted a mathematician at Aberdeen to keep track of computing problems, it specified a government rank of P-6 "and even a P-7 or a P-8 is not absolutely excluded." The actual senior computers doing that work were graded P-5 (Arnold Lowan), P-3 (Blanch, who had already done the job) and P-2 or P-1 for the rest of the planning committee. Grier's summary of what professionalisation achieved: "In the case of the human computer, professionalization produced no independence, little respect, and nothing that could be characterized as self-governance. Professionalization came just as human computers were being replaced by computing machines."

The ending was administrative and quiet. War contracts closed: "Most of the American computing groups of the Second World War ended in such a manner. As the country returned to the peacetime economy, the government reduced the budget for scientific and engineering research. In response, the leaders of the research laboratories, with or without an argument, cut their staffs, including their human computers." At the Institute for Numerical Analysis in 1948, Blanch built one more computing office — sixteen computers, seven men and nine women — and "most members of the institute staff believed that the human computers would be temporary workers, quickly replaced by a new 'automatic computing machine.'" They were. By the epilogue's accounting, the bureau's Computational Laboratory closed, the American Nautical Almanac moved to electronic computers, observatories bought their own machines, and the last commercial holdouts — insurance firms and petroleum refiners with small staffs of calculating assistants — replaced them with IBM 370s, PDP-8s and Burroughs B-6500s.

Fifty years later, on 8 November 2019, the Hidden Figures Congressional Gold Medal Act became Public Law 116-68. Its fifth medal is awarded to "all women who served as computers, mathematicians and engineers" at NACA and NASA between the 1930s and the 1970s. It is the only formal notice the occupation ever received, and it arrived two generations after the occupation stopped existing.

Why a reading would cite it

1. It is the standing case for the claim that skill does not protect an occupation. The proposal states the occasion exactly: an occupation is claimed to be safe because the people in it are skilled, or doomed because the task is automatable — this is the case that refuses both readings. The skill here was real and the output was first-rate; Leavitt's relation is still load bearing in cosmology. The occupation ended anyway, completely, and the ending had nothing to do with whether the people were good at it. When the Work and the Economy lens meets an argument that some category of knowledge work is safe because it requires judgment, training or care, this is the counterexample with a completed record rather than a forecast.

2. It is the case where task-level automatability was right and still predicted nothing that mattered. frey-osborne-2013 scores occupations by the task content of the job. Run that method on a human computer in 1935 and it returns the correct answer — the task is fully specifiable, therefore fully automatable — and it tells you nothing about what actually happened, which is that the elimination took three more decades, ran through a war that expanded the occupation, and sorted the people into outcomes that had almost nothing to do with the task and almost everything to do with which institution they happened to work for. Cite the two entries together when an exposure index arrives: this is what the index gets right and this is the size of what it leaves out.

3. It supplies the honest version of "they moved into the new jobs." The transfer story is true in famous cases and false in general, and section 3 grades both halves. A reading that needs to push back on either the optimistic or the catastrophic form of the displacement argument can do it from one entry.

4. It is the case for watching status and pay separately from headcount. Nothing in the record here is explained by counting jobs. What explains it is grading: P-1 versus P-6 for the same work, twenty-five cents an hour set deliberately below a factory wage, a job title that changed to Engineer in 1970 with a degree requirement attached. When a reading meets a claim that AI is "augmenting" rather than replacing a profession, the question this entry licenses is what happened to the grade, the title and the entry requirement — not what happened to the count.

5. It is a warning about the present tense, from the historian himself. Grier closes by guessing that the scientists of the 2061 comet return will know as little about the daily lives of "the computational workers who are so common in our age, the computer programmers, the network managers, and the Web designers" as today's astronomers know about Lepaute or Airy's boy computers at Greenwich. That sentence was written in 2005 about a category that then looked permanent. US computer programmer employment is now about 121,200 and the Bureau of Labor Statistics projects a further 6% decline to 2034, citing automation of programming tasks by AI among the reasons. The occupation that absorbed the human computers is now the occupation being cited as evidence of displacement. That is not a metaphor a reading has to reach for; it is the same file.

What it got right, and what it got wrong

interpretation does not require this section. This entry needs it, because proposals.md attached four checkable claims to the id and two of them do not survive contact with the sources.

Claim 1: "Mechanical then electronic calculation eliminated the job title completely inside roughly two decades." Half right, and the wrong half is instructive. Mechanical calculation did not eliminate the job. Mass-produced adding and calculating machines entered computing offices around 1900 and made the occupation larger, not smaller: Grier's second period, 1880–1930, is the one in which machines let computing spread out of astronomy into economics and social statistics, and the WPA project of 1938 and the war offices of 1944 are downstream of that expansion. Roughly two decades is right only if the clock starts at ENIAC's public demonstration on 15 February 1946 and stops at JPL's retitling around 1970. From de Prony it is nearly two centuries. The distinction is not pedantry: it is the difference between "a tool arrived and the job went" and "a tool arrived, the job grew for fifty years, and then a different kind of tool ended it," and a reading citing this entry to argue about pace needs the second version.

Claim 2: "The occupation ceased to exist." Right, and unusually clean. Hand computation of tables survives in no professional niche. What should be carried with it is that nobody can say when, and that the record of the ending is extraordinarily thin — "the vast majority of computers left no record of their lives beyond a single footnote to a scholarly article or an acknowledgment in the bottom margin of a mathematical table." Grier reconstructed the story from a cassette tape a family had kept, a scrapbook, a box with a confusing label at the National Archives, and a four-hour phone call with a man in a hospital bed. An occupation can be erased so completely that the erasure itself leaves no trace.

Claim 3: "The first ENIAC programmers were recruited from the human computers, because the people who understood the calculation were the people who could instruct a machine to do it." Right on the fact, and worth stating precisely. Six women — Kathleen McNulty, Jean Jennings Bartik, Frances "Betty" Snyder Holberton, Marlyn Wescoff, Frances Bilas and Ruth Lichterman — were selected in 1945 from a Moore School group of more than eighty women computing ballistics trajectories by hand. The machine was classified and they had no clearance, so they learned to program it from its logical and electrical block diagrams. At the public demonstration on 15 February 1946 they were not introduced, and they were not invited to the dinner afterwards.

Claim 4: "A substantial number of the people went on to program the machines that replaced them." Partly wrong, and this is the correction the entry exists to make. The named transfers are real: ENIAC's six; Dorothy Vaughan, who learned FORTRAN and worked as a NASA programmer until 1971; Mary Jackson, who became NASA's first Black woman engineer in 1958; Helen Ling at JPL, who spent the 1970s hiring women as programmers to get them through a door that now required an engineering degree. But Grier's general finding runs the other way, and he states it as a thesis about the whole occupation:

> few human computers contributed to the invention of electronic computing > equipment, and few computing offices were connected to machine development > projects. It is best to view the human computing organizations as the backdrop > against which the story of electronic computers unfolds. [...] they would prove > to be the secondary characters in the narrative, the Rosencrantz and > Guildenstern instead of the Hamlet and Ophelia.

And where he follows a whole staff rather than a famous individual, the outcome is ordinary: the Mathematical Tables Project's computers "came from the working-class neighborhoods around New York City: Brooklyn, Yonkers, Jersey City, the Bronx, Harlem, and the Lower East Side. A few of them would be offered positions in Washington, D.C., but most would soon be looking for work in New York." At the Institute for Numerical Analysis two years later, "few computers stayed at the institute for more than a year," and "the rapid departure of human computers never seemed to bother the mathematicians, as they were looking ahead to the new computing machine."

So the transfer is real and the sample is poisoned. The people we can name are the ones who were already at the professional end of a stratified workforce, inside institutions that survived the transition and kept records. The people who filled in blanks — de Prony's ninety, the WPA's 450, the comptometer operators in ordinary firms — left no forwarding address, and their absence from the story reads as absence of harm. The transition's own historiography was written by the people it worked for. A reading citing this entry for the optimistic lesson is citing the survivors; that has to be said out loud each time or the entry is doing damage.

Claim 5, the proposal's "second finding": the successor occupation was "initially low-status and heavily female... as programming became prestigious and well paid, its composition changed substantially." Contested, badly dated, and resting on a premise that a specialist historian calls thinly sourced. The proposal calls this the part usually left out. In fact it is now the part most often left in, and in a form the evidence does not support.

The status half is solid: programming was initially read as clerical follow-on work, the ENIAC women were treated accordingly, and Ensmenger, Light and Hicks all document a masculinisation of the field through hiring tests and professionalisation. The composition half needs numbers, and the numbers do not say what the meme says. Thomas Misa's reconstruction:

• Women's share of US computer science bachelor's degrees rose from about 12% in 1965 to 37% at its 1984–85 peak, then fell to 15–20% by the early 2000s; the Computing Research Association's Taulbee low point was 11.2% in 2009.
• Women's participation in the white-collar computing workforce rose to 38% by the mid-1980s, and stood at 27% in the 2011 American Community Survey.
• At the 1949 international computing conference at Harvard, 33 identifiable women attended against 540 identifiable men — about 6%.

Misa's verdict on the underlying premise is blunt: "The claim of computer programming being, at any time, 50 percent women is thinly sourced." He traces it to Ensmenger, whose own sentence is "the exact percentage of female programmers is difficult to pin down with any accuracy," and whose sources are Richard Canning's 1974 EDP Analyzer — which gives 25–30% of computer personnel excluding the heavily feminised keypunch and machine operators — and an incompletely cited 1964 Datamation article misattributed to Sherry Turkle.

The chronology is the load-bearing correction. The popular story is women built programming, then men took it and the composition flipped, with the flip implied to follow the human computers. The measured collapse begins in the mid-1980s and runs through the 2000s — a generation after the last computing offices closed, and after two decades in which women's participation was rising. Both things can be true at once: a field's overall share of women can rise while its prestigious upper tier closes, which is roughly what Ensmenger, Hicks and Misa are each describing from different heights. A reading may cite the reallocation of rewards as a real pattern. It may not cite this entry for the simple version, and it must not put a date on the flip that the data does not support.

One detail the entry corrects because it travels everywhere. De Prony's computers are routinely described as a room full of unemployed hairdressers. Grier: many were "former servants or wig dressers," and — the part always dropped — "No record has been found of a centralized computing floor for the former hairdressers." Given the size of the staff and the cottage-work tradition of the period, they most likely calculated at home. The vivid image of the hairdresser factory floor is an embellishment; the division of labour it illustrates is not.

Commonly misused as

Not required for interpretation, but this history is now popular enough to have acquired standard misuses, and three of them are live in exactly the arguments a reading will meet.

Misuse one: "automation destroyed the job, but the workers moved up into better ones." The entry does not support this and section 3 says why. What it supports is narrower and stranger: some of the workers, drawn overwhelmingly from the top stratum of a stratified workforce inside institutions that survived, moved into the successor occupation, and the rest left no record at all. The correct use of this case is as a caution about which outcomes get recorded, not as evidence about which outcomes were common.

Misuse two: "it was women's work until men took it over." The gender reading is true and incomplete, and the incompleteness matters because it makes the case seem narrower than it is. Grier's framing — "the job of the dispossessed" — is better attested and covers more: African Americans in a segregated office at Langley, polio survivors in the WPA project, wig dressers in Paris, Jews, the Irish, the poor. The occupation absorbed whoever was excluded from the career the work was adjacent to. That is a mechanism a reading can look for now, and "women's work" alone will not find it.

Misuse three: "the task was routine, so it was always going to be automated." The routine was manufactured, not discovered. Somebody — de Prony's planners, and after them every computing office that copied them — sat down and decided which part of the thinking to strip of judgment so it could be handed to people who had none of the training. The worksheets "left nothing for the workers to interpret" by design. What made computation automatable was therefore an organisational choice made two centuries before the machine existed, and the interesting question this entry puts to the present is not which jobs are routine but who is currently deciding which parts of which jobs to make routine, and what that decision does to the people left holding the remainder.

Misuse four, the one this project must guard against directly. Nothing here is evidence. This entry deposits nothing into history/events.jsonl, moves the needle by nothing, and sets no landmark. The 121,200, the 37%, the 27%, the twenty-five cents an hour: those are figures to be quoted correctly inside an explanation, never counted as a reading.

Sources

Primary and near-primary, read directly as pages.

• David Alan Grier, When Computers Were Human (Princeton University Press, 2005). Read: Introduction, "A Grandmother's Secret Life," pp. 1–8 (the definition, the scope, "blue-collar science," the Rosencrantz and Guildenstern thesis, the sources problem); ch. 2, "The Children of Adam Smith," pp. 34–39 (de Prony, the pin-factory recollection, the three tiers, the worksheets, differencing, Lalande's seven hundred results a day, the absent computing floor, Babbage's 1814 application to Greenwich); ch. 17, "The Victor's Share," pp. 276–283 (computers become "girls," kilogirl, Comrie's "Careers for Girls," the job of the dispossessed, the P-grade comparison, the German computing offices); ch. 18, "I Alone Am Left to Tell Thee," pp. 298–304 and 313–317 (the postwar liquidation, the Mathematical Tables Project's computers going home, the Institute for Numerical Analysis, the 1954 conference, the Handbook and System/360); Epilogue, "Final Passage: Halley's Comet 1986," pp. 318–322 (Blanch's 1967 retirement, the closing of the computing offices, Yeomans's FORTRAN IV program, the 2061 passage). The copy read was a PDF mirrored at rybn.org and read as page images; pagination is the book's own.
• Thomas J. Misa, "Gender Bias in Computing," in William Aspray, ed., Historical Studies in Computing, Information, and Society (Springer, 2019), pp. 113–133; preprint arXiv:2210.16449. Read: pp. 2–9 of the preprint (the 12%-to-37% degree series, the 38% mid-1980s workforce figure, the 2011 ACS 27%, the CRA Taulbee 11.2% low, the 1949 Harvard conference count, and the "thinly sourced" judgment with its footnote tracing Ensmenger's sources to Canning 1974 and the misattributed 1964 Datamation piece).
• Hidden Figures Congressional Gold Medal Act, Public Law 116-68, signed 8 November 2019, for the fifth medal's wording and its coverage of "the 1930s and the 1970s."
• National Archives at Philadelphia, West Area Computers Unit, for the 5 May 1958 disbanding and the reassignment of the unit's women into program offices.

Secondary, used for specific facts and flagged as such.

• Harvard College Observatory computers: the eighty-plus figure, Pickering's 1877–1919 directorship and the twenty-five-to-fifty-cents-an-hour wage come from The Harvard Crimson (26 September 2019) and the Harvard Plate Stacks and Wikipedia accounts, not from a primary payroll record. Leavitt's 1912 Harvard Circular and the restriction on her continuing the work are from Britannica and the Plate Stacks biography.
• The ENIAC six, the selection from a group of more than eighty computers, the block-diagram training and the exclusion from the demonstration dinner: Penn Today, IEEE Spectrum, and the Columbia University computing history pages. W. Barkley Fritz, "The Women of ENIAC," IEEE Annals of the History of Computing 18:3 (1996), is the primary account and was located but not read.
• JPL's replacement of the title Computer with Engineer around 1970, and Helen Ling's hiring of women as programmers through that decade: the published accounts of Nathalia Holt's Rise of the Rocket Girls (2016), not the book itself.
• The 1613 first recorded use of "computer" and Mary Edwards as one of about thirty-five Nautical Almanac computers: the Oxford English Dictionary as quoted in secondary sources, and Wikipedia's "Computer (occupation)."
• Current US programmer employment: 121,200 jobs in 2024 and a projected 6% decline to 2034, from the Bureau of Labor Statistics Occupational Outlook Handbook. The longer series — above 700,000 at the dot-com peak, and a 27.5% fall in the twelve-month average since about 2023 on Current Population Survey data — from Fortune, 17 March 2025.

Fetch failures, recorded per the rule that a failed search is noted and the work continues.

• bls.gov returned HTTP 403 to every request from this machine, exactly as it did for frey-osborne-2013. The 121,200 and the −6% are taken from search engine summaries of the Occupational Outlook Handbook page rather than from the page itself, and a later reading with working access should re-check both before quoting them.
• Nathan Ensmenger, "What percentage of programmers were women?", thecomputerboys.com — the host's TLS certificate does not cover the apex domain and the fetch failed on every form of the URL; web.archive.org is not reachable from this tool either. Ensmenger's sentence quoted above is taken from Misa's quotation of it, not from Ensmenger directly, and the entry says so rather than implying a reading it did not do.
• US Census Bureau / IPUMS, Alphabetical Index of Occupations and Industries: 1950 — not read; the PDF exceeded the fetcher's size limit. This entry therefore makes no claim about whether "computer" appeared as a distinct occupational title in any American census classification, or about how many people were ever counted in it. That is the single most useful unread source for this entry, because it is the only thing that could turn "the occupation disappeared" from a well-attested qualitative claim into a counted one.
• Lorraine Daston, "Calculation and the Division of Labor 1750–1950" (Bulletin of the German Historical Institute, 2018) — the host served a security challenge page instead of the article. It is the best available scholarly counterweight to Grier on the de Prony material and nothing from it is used here.