Will AI Really Swipe Our Jobs?

‍       I’ll admit it: I was a champion of the theory that AI will eventually make off with most of our jobs.

      Early 20th-century industrialization had already automated assembly, and early 21st-century information technologies had automated communication. But now, a quarter of the way through the 21st century—the argument goes—artificial intelligence “models” are beginning to master additional large areas of 21st-century work: specific tasks like driving and writing, larger business skills like organizing, planning and—most recently, thanks initially to Anthropic’s AI model, Claude—computer coding itself.

      Top of Form

Bottom of Form

Hence, there is a reasonable chance that humans will become, to use the British term, redundant.

      And—the argument goes—your job, if you are still working, and the jobs of most of your fellow humans, will go poof!

      Indeed, I had recently played the role of Chicken Little (“Employment is falling!”) with this quick video.

      For you don’t have to be a technophobe nowadays to predict an AI-generated job apocalypse. About a year ago Axios published this rather frightening warning: “AI could wipe out half of all entry-level white-collar jobs—and spike unemployment to 10-20 percent in the next one to five years.” That warning came from Dario Amodei, the CEO of Anthropic, himself.

      However, I did have one additional, significantly more controversial view on AI and jobs: I also believed that humans—with the help of an okay, if not generous, “universal basic income”—could create an exciting post-work world, with machines handling the labor and humans having the adventures and the fun. So, I also was Dr. Pangloss.

      But what if I was wrong?

      It certainly would be easy to be wrong about that additional, optimistic prediction. For that “exciting post-work world” would probably require some Robin Hooding of the rich to fund a UBI. And our billionaires have not been demonstrating much of an inclination to surrender their third homes or private planes to fund an ex-worker benefit fund—particularly given the unabashed greed and corruption of the Trump era.

      However, what I really want to rethink in this essay is that first prediction: What if AI does not, despite what Dario Amodei and I and many others have predicted, in fact lead to a large human job deficit? What if we are wrong in thinking that a whole lot of positions, even professions, are going to be disappearing, leaving us and our dependents unemployed—on a permanent vacation.

      Let’s start here: We are hardly the first generation to fear being replaced by machines. And machines certainly can steal jobs. Look what happened to horses: Karl Benz’ invention in 1886 of a “motorwagon,” powered by an internal-combustion engine—what we would call today an “automobile”—would put a lot of horses out of work over the next century. There were more than 26 million farm horses and mules in the United States in 1918, by one estimate, fewer than 3 million by 1960.

      However, except during the Great Recession, which was hard to pin on automation, and a few less “great” recessions, what’s called the “headline” unemployment rate in the United States has stayed at relatively low levels.

      Historical estimates, based partly on census data, suggest that about 5 percent of U.S. workers were unemployed in 1900.

      The 20th century saw the mechanization of both agriculture and manufacturing. It saw increased industrialization, then electrification and eventually computerization, all of which made it possible to get work done with fewer human workers: fewer factory workers, typesetters, secretaries, switchboard operators, travel agents, etc.

      And more people were chasing those jobs: the 20th century also saw the size of the working-age population in the United States almost triple and the number of women in the (paid) labor force more than tripled.

      Yet the unemployment rate at the end of the 20th century in the United States was only 4.2 percent. (Only? Yes, we wish it were lower still, but this does include people who are between jobs.)

      And the unemployment rate in the United States as I write in June of 2026 is nearly the same, 4.3 percent, although, thanks to the internet, cell phones and further computerization, there are many, many fewer people currently employed as bank tellers, retail clerks, journalists, etc.

      In other words, the notion that technological advances must lead to a decrease in employment opportunities has proven, over and over again in modern times, to be a fallacy. Indeed, that fallacy has a name—a cool name often traced to an economist named David Frederick Schloss in 1891, near the start of this period of rapid technological change: the “Lump of Labor Fallacy.”

      The mistake—back at the end of the 19th century and, still, a quarter of a century into the 21st—is to assume that there is a certain amount of work to be done—a “lump”—and, therefore, if machines are doing more work there will be less work for people. It is a fallacy because human beings, in modern times, keep coming up with useful new kinds of work to do: building automobiles, for example, or finding telephone numbers or designing websites or building data centers.

      The number of things horses can do to earn their keep is quite limited. That has not proven true of humans.

      So, yeah, the electric light eventually cost lamplighters their jobs. Sucked for them. But most of them or their children would have found new positions in a lightbulb factory or as electricians or in another field entirely.

      The author of this article worked for many, many decades as a journalism professor. It is possible to imagine that that job may no longer be available as AI takes over much teaching, as AI takes over much journalism, even much writing. That might hurt people I care about, former colleagues. But it is possible, even likely that new jobs will be created for humans as—and we are necessarily improvising here— “AI ethicists,” perhaps, or “AI orchestrators.” (Not to mention Substack writers.)

      Or maybe, if AI produces goods and services ridiculously inexpensively, maybe this ancient notion that one must have a job will fade, and instead of jobs we will have hobbies, interests and a Universal Basic Income.

      This essay first appeared in Mitch Stephens’ Substack, Ideas on Ideas.