Zum Inhalt springen

Automation and the Future of Work

Zusammenfassung

Will the machines take our jobs? It is one of the oldest anxieties in industrial history, and computing has made it perennial. From the Luddites smashing looms in 1811 to economists debating whether large language models will hollow out white-collar work, every wave of automation has provoked the same fear — and, until now, repeatedly defied it, destroying particular jobs while creating others and, on aggregate, raising living standards. This article traces the long argument over technological unemployment: the historical pattern of automation creating more work than it destroyed, the mid-20th-century computing alarms, the “polarization” of the labor market by software, the universal-basic-income debate, and the open question of whether AI represents merely the next turn of the wheel or a genuine break with the past.

The Oldest Fear

Anxiety about machines displacing workers is at least two centuries old. The Luddites — early-19th-century English textile workers — were not mindless technophobes; they were skilled artisans destroying the specific power looms and knitting frames that threatened their livelihoods and bargaining power. John Maynard Keynes coined the term “technological unemployment” in 1930, warning of joblessness “due to our discovery of means of economising the use of labour outrunning the pace at which we can find new uses for labour” — even as he predicted a 15-hour work week by 2030 as productivity soared.

The recurring historical pattern, through the first two industrial revolutions, was the lump-of-labour fallacy in reverse: automation destroyed jobs in some sectors but raised productivity and incomes, which created demand and entirely new categories of work that no one had foreseen. The classic illustration is agriculture: mechanization cut farm employment from a majority of the workforce to a few percent in industrialized nations, yet unemployment did not rise to match — those workers (or their children) moved to factories, then offices, then services that didn’t previously exist.

Computers Enter the Argument

Computing reignited the fear in the mid-20th century. Norbert Wiener, founder of cybernetics, warned in the late 1940s and 1950s that automatic machines could produce “an unemployment situation in comparison with which… the depression of the thirties will seem a pleasant joke.” In 1964 a group of intellectuals sent President Johnson the “Triple Revolution” memorandum, arguing that cybernation would sever the link between jobs and income and demanding a guaranteed income.

These alarms proved premature — the postwar decades saw computing accompany, not prevent, near-full employment and rising wages. But the nature of the impact became clearer: computers were spectacularly good at routine, codifiable tasks (calculation, record-keeping, repetitive assembly) and poor at non-routine work requiring judgment, dexterity, or interpersonal skill.

Polarization, Not Mass Unemployment

The dominant late-20th-century finding, associated with economists David Autor, Frank Levy, and Richard Murnane, was the routine-biased technological change hypothesis: software automated the middle of the skill distribution — clerks, bookkeepers, routine factory work — while complementing high-skill cognitive work (which it made more productive) and leaving low-skill manual service work (cleaning, caregiving) hard to automate.

The result was job polarization: a “barbell” labor market growing at the high-wage and low-wage ends while the middle hollowed out, contributing to wage stagnation and inequality. This reframed the debate. The problem was not a quantity of jobs vanishing (employment kept growing) but a distribution problem — automation reshaping who won and who lost, demanding higher skills and concentrating gains among capital owners and the highly educated. Moravec’s paradox captured the surprising contours: tasks humans find hard (chess, calculus) are easy for machines, while tasks humans find effortless (walking, recognizing a face, folding laundry) are extraordinarily hard — which is why robots replaced accountants’ arithmetic long before they could replace a janitor.

The AI Inflection and the UBI Debate

A 2013 Oxford study by Frey and Osborne estimated that 47% of US jobs were at “high risk” of computerization within two decades — a headline that, whatever its methodological contestation, crystallized a new wave of anxiety. The rise of generative AI from 2022 sharpened it in an unprecedented direction: unlike earlier automation, LLMs target non-routine cognitive and creative work — writing, coding, analysis, illustration, customer service — the very white-collar tasks long assumed to be automation-proof. Early studies suggested AI compresses skill gaps (helping novices more than experts) and could affect a large share of tasks in knowledge jobs, even if not entire occupations.

This revived the Universal Basic Income (UBI) debate: an unconditional cash payment to all, proposed as a buffer against automation-driven displacement and championed by figures across the political spectrum, from Silicon Valley (Sam Altman’s funded UBI experiments) to politicians like Andrew Yang. Real-world pilots (Finland’s 2017–2018 trial, GiveDirectly’s long-run Kenya study, various city programs) produced encouraging effects on well-being but no consensus on affordability or labor-supply effects at national scale.

Dead End: The Confident Job-Loss Forecast

The most reliable failure in this entire field is the precise prediction of mass technological unemployment — and, equally, the precise prediction of which jobs will vanish.

Every confident forecast has been humbled. Wiener’s and the Triple Revolution’s mid-century mass-unemployment predictions were wrong for half a century. The Frey-Osborne “47%” became a media staple but conflated tasks with jobs — most occupations are bundles of tasks, some automatable and some not, so automation usually reshapes a job rather than eliminating it (the classic example: ATMs did not eliminate bank tellers; the number of tellers grew for decades as cheaper branches proliferated and tellers shifted to relationship work, before eventually declining). Predictions of imminent fully self-driving trucks displacing millions of drivers repeatedly slipped past their deadlines as the last few percent of edge cases proved brutally hard.

The honest pattern is that economists can confidently say automation will change work and redistribute its rewards, but they have a near-perfect record of being wrong about the timing, the magnitude, and the specific occupations. The genuinely open question — never resolvable by extrapolation — is whether AI is the same wheel turning again (disruption, then new jobs and higher productivity) or a qualitative break in which machines finally substitute for human cognition broadly enough to break the historical pattern. Anyone claiming certainty on that question is repeating a two-century-old mistake.

📚 Sources