When a computer was a person doing arithmetic

There is a settled way of talking about human computers. It is worth asking how much of it survives contact with the detail.

The job came first and the machine took the name

For a long period, a computer was a person, usually employed in astronomy, navigation, ballistics, surveying or actuarial work. The transfer of the word to machinery happened gradually, with qualifiers like electronic computer used to distinguish them during the changeover.

That the word survived and the workers did not is a fairly exact summary of what happened. A job title becoming the name of the thing that replaced it is rare and worth noticing.

Big calculations were industrialised

Large computations were broken into simple repeated operations distributed across many workers, with results checked by duplication or by difference methods. This is division of labour applied to arithmetic, and it was organised on explicitly industrial principles. Error checking was structural rather than individual: the process was designed to catch mistakes rather than relying on nobody making them.

That design philosophy carried directly into machine computing and remains central to it.

The workforce was substantially female in many settings

Computing groups in observatories, laboratories and wartime projects employed large numbers of women, often at lower grades and lower pay than men doing adjacent work. Several were mathematicians doing substantive analytical work rather than mechanical calculation, and the boundary between the two was frequently blurred by job classification. Recovery of these histories has been substantial in recent decades and remains uneven across countries and institutions.

In hindsight, naming specific individuals requires care, since popular accounts have sometimes compressed or misattributed roles.

The transition ran through the operators

When electronic machines arrived, the people who understood how the calculation was organised were frequently the ones who set them up and ran them. Early programming was therefore continuous with the previous work rather than a break from it, which is not how the story is usually told. The status of the work changed later, as programming became professionalised and better paid.

Task continuity and status discontinuity are different things and are easily confused.

Published tables were the product

The output was often printed tables — logarithms, tides, ephemerides, ballistics — which were then used by everyone else. Errors in such tables propagated widely and were taken seriously enough to drive both verification methods and, historically, the case for mechanising calculation.

What is easy to forget is that the desire to eliminate table error is one of the recurring motivations in the early history of computing machinery. The machine was built partly to solve a quality control problem in a labour process.

How widespread this was is genuinely contested.

Some of the work never went away

Verification, sanity checking and knowing whether an answer is plausible remain human tasks and have become more important as computation has become cheaper. A wrong answer produced instantly and confidently is a more dangerous artefact than a slow one produced by hand.

Look closer and the skill that survived is judgement about results rather than production of them. That is the durable half of nearly every job that automation has taken.

The takeaway

The word did not change meaning. It just stopped referring to anybody.

The loss is small and cumulative, which is why it goes unremarked.