■ CRITICAL RISK ■ Science & Research
This occupation — applying standardized formulas so engineers and scientists don't have to — was rendered redundant by cheap computation long before AI, and AI removes the last excuse for its existence. The math survives; the technician performing it does not.
“A calculator has been doing your job since 1972.”
Our AI replacement risk score — how we score jobs
The mathematical technician was the human middle layer between a scientist's problem and its numerical answer: applying standardized formulas, reducing raw experimental data, computing trajectories and stress tolerances, checking tables, and grinding through calculations too tedious for the credentialed staff upstairs. The role descends directly from the human 'computers' of aerospace fame — rooms of people, disproportionately women, doing arithmetic in shifts. It was skilled, meticulous work performed with slide rules, mechanical calculators, and eventually early software.
Electronic computation dissolved the job's reason to exist in stages. Scientific calculators killed the routine formula work; spreadsheets and numerical packages like MATLAB let engineers run their own computations without an intermediary; and statistical software swallowed data reduction whole. Government occupational statistics have long tracked this as one of the smallest and fastest-declining categories they measure — the label persists mostly as a classification artifact. Now AI adds the final layer: language models can set up the calculation, not just execute it, translating a messy word problem into equations and code, which was the one cognitive step the technician still owned.
What genuinely resists automation lives above this job title, not within it: framing which problem to solve, judging whether a model's assumptions fit reality, and catching the answer that's precisely computed but physically absurd. Those belong to mathematicians, statisticians, data scientists, and engineers — careers that absorbed the interesting parts of this role and left the shell. A 98 here isn't a prediction of decline; it's a description of a category that computation already hollowed out, now getting its interior swept.
Automatability: our editorial assessment of current and near-term AI capability
Mostly historical. The occupation began shrinking when electronic calculators arrived and never stopped; today it's among the smallest tracked job categories, with remaining positions concentrated in a few defense and research installations running legacy workflows. AI closes the final gap this decade by automating problem setup, not just execution. Anyone holding this title now is effectively in a role the org chart hasn't gotten around to renaming.
Barely. It survives as one of the smallest occupational categories government statistics track, mostly in legacy defense and research settings. The work it once covered — routine calculation, data reduction, table checking — migrated into software decades ago, and the interesting remnants were absorbed by analyst and engineering roles. New hiring under this exact title is close to nonexistent.
Hardware first, then software, then job-title inflation. Scientific calculators and mainframes ate the arithmetic; MATLAB, spreadsheets, and statistical packages let scientists compute for themselves; and the judgment-bearing parts of the work were folded into better-paid titles like data analyst and statistician. AI now automates even setting up the calculation, which was the last distinctly human step.
The math is worth it; the technician framing isn't. Quantitative skill remains one of the best career assets available — but packaged as data science, statistics, actuarial work, or engineering, where you choose and interpret the computation rather than execute it. Pure execution of someone else's formulas has been a machine's job since the pocket calculator.
It executes known methods well and increasingly sets up problems from plain-language descriptions, which is genuinely new. It still produces confident nonsense at the edges — mis-stated assumptions, unit errors, plausible-looking garbage — so high-stakes computation needs human review. But that reviewer is an engineer or statistician, not a technician performing the arithmetic by hand.