Automation
Automation is the use of machines, software or control systems to perform tasks that people previously did, with less human effort or intervention. It spans mechanical devices such as looms, electromechanical controllers on factory floors, software scripts that move data between applications, and, more recently, machine-learning systems that handle perception and language. The word describes a process, not an endpoint: a task can be automated partly, fully or not at all, and the interesting questions are usually about which tasks, who gains and what new work emerges.
The historical pattern begins with machines that stored instructions. The Jacquard loom, demonstrated in 1801 and patented in 1804, used punched cards to control which warp threads were lifted, so a loom could weave complex patterns without a skilled operator selecting each thread; the cards later inspired data storage for early computing. Resistance was real: in 1811–1816 the Luddites, English textile workers, smashed machines in protest at changes in their trade, and the term has since become shorthand, often unfairly, for opposition to technology. Henry Ford's moving assembly line, introduced for the Model T at Highland Park in 1913, broke manufacturing into short repeated steps and cut assembly time drastically. Industrial robots followed: the Unimate, installed at a General Motors plant in 1961, is generally cited as the first.
Software brought automation into offices through robotic process automation (RPA). RPA tools such as those from UiPath, Automation Anywhere and Blue Prism record or script the clicks and keystrokes a clerk performs — copying an invoice number from an email into an accounting system, say — and replay them without human involvement. RPA is rule-based and brittle: it works on structured, stable processes and breaks when a screen layout or a document format changes. Machine learning widens the reach by reading unstructured documents and handling variation, which is why vendors speak of intelligent automation, but the underlying principle remains a defined workflow with exceptions routed to people.
Economists analyze automation at the level of tasks rather than whole occupations. Autor, Levy and Murnane's 2003 paper "The Skill Content of Recent Technological Change" argued that computers substitute for routine tasks, those that follow explicit rules, and complement non-routine analytical and interpersonal tasks, and David Autor's 2015 article "Why Are There Still So Many Jobs?" explained why employment kept growing despite automation: tasks are bundled, human adaptability fills gaps, and demand responds to cheaper output. Michael Polanyi's observation that "we can know more than we can tell" explains why tacit tasks resisted automation for decades. Acemoglu and Restrepo's task-based model, developed in papers from 2018 onward, adds a key idea: automation displaces workers from tasks, but the creation of new tasks reinstates labor, and the net outcome depends on the balance between the two.
Automation and augmentation are choices as well as outcomes. The same technology can be designed to replace a worker or to extend one — a diagnostic model that makes the decision alone versus one that flags findings for a radiologist. Erik Brynjolfsson called the emphasis on human-like replacement "the Turing Trap" in a 2022 essay, arguing that augmentation tends to create more shared value. Historical examples complicate any simple story: the spread of automated teller machines coincided for decades with continued growth in bank teller employment, as analyses by James Bessen and others describe, because cheaper branches multiplied and the teller's role shifted toward relationship tasks, though that pattern has not held in every occupation or era.
The standard pitfall is to equate automatable with automated, or automated with jobs eliminated. Technical feasibility ignores cost, regulation, reliability, liability and worker and customer acceptance, so adoption is slower and more uneven than capability suggests. Conversely, a job that survives may still change greatly, with lower wages or deskilling for some people, even as aggregate employment holds up. Past outcomes also do not guarantee future ones; Keynes warned of "technological unemployment" in 1930 but also expected it to be temporary, and the honest summary today is that long-run adjustment has been real while the transition costs have been unevenly borne.
Read automation as a task-by-task reallocation between people and machines, where the outcome for work depends less on what the technology can do than on which tasks it takes, which it creates, and who shares in the resulting gains.