Elon Musk, Sam Altman of OpenAI, and some other major executives of artificial intelligence (AI) firms are sure that AI will destroy millions of American jobs and that many of those who lose work will not find gainful employment. Musk, Altman, and Marc Benioff, CEO of Salesforce, advocate a universal basic income (UBI) for those who they think will never find work. I’ve written elsewhere about the fact that a UBI, even one that replaced means-tested welfare programs, would enormously increase both government spending and our federal budget deficit.
There’s another problem, and it’s the one I focus on here. A large UBI would assure that millions of people will never work. As a result, we would miss the products and services that those people would have produced. One way to see that is to imagine that at various times in our history the federal government had implemented a UBI.
From farm to factory
Imagine that we’re back in 1900. Approximately 40 percent of American workers work on farms. A visionary back then foresees the replacement of horses with tractors and the use of fertilizers and irrigation. Together those improvements in technology will multiply farm output and shrink farm employment. He foresees correctly that by 1980, only 3.4 percent of American workers will work on farms. That visionary also advocates a federal government program to give everyone enough money to live modestly without working. Such a proposal is now known as a universal basic income (UBI.) People don’t take him seriously. Very few people believe that the federal government should have any role in subsidizing people so that they don’t have to work.
But what if key politicians back then had taken him seriously and, in the next few years, had implemented a UBI?
Then many of the millions of people who lost jobs on the farm would not have become factory workers. Manufacturing employment replaced many jobs on the farm. In 1900, 11.1 million people were employed in US agriculture.; the total labor force then was 27.6 million. By 1980, that was down to 3.4 million, even though the labor force in 1980 was 99.3 million, which was 3.6 times as large as the labor force in 1900. With a UBI, there would still be millions of people leaving farm jobs and working in manufacturing. But there would also be millions who would not work. Think of the cars, trucks, televisions, automatic washers and dryers, and other valuable products we would not have. Fortunately, we dodged a bullet. We didn’t get a universal basic income.
Bastiat’s insight
It’s important to remember the insight that French economist Frédéric Bastiat had about issues like this in the mid-nineteenth century. In a famous essay, “What Is Seen and What Is Not Seen,” Bastiat wrote:
In the sphere of economics an action, a habit, an institution, or a law engenders not just one effect but a series of effects. Of these effects only the first is immediate; it is revealed simultaneously with its cause, it is seen. The others merely occur successively, they are not seen; we are lucky if we foresee them.
What Bastiat could have foreseen is that if a UBI had been implemented early last century, many of those valuable items—cars, trucks, automatic washers, etc.—would not have been produced.
Bastiat went on to point out the difference between a bad economist and a good economist: “A bad one relies on the visible effect while the good one takes account both of the effect one can see and of those one must foresee.” He could have generalized beyond economists. One key difference between a bad thinker and a good thinker, economist or not, is whether he takes account of both things that are seen and things that must be—and can be—foreseen.
From manufacturing to services
Now fast-forward to 1979, when US manufacturing jobs peaked at 19.4 million. Imagine that a visionary in 1979 had foreseen the huge increase in manufacturing productivity that, over the next forty years, would raise manufacturing output by 98 percent while reducing the number of manufacturing jobs to 12.8 million, a drop of 34 percent. What if that person had persuaded the government to provide a universal basic income? Think, a la Bastiat, of the unseen. With a few million people displaced from manufacturing jobs but not working because they didn’t have to, we wouldn’t have had the explosion in service jobs that we actually had. We would have had less medical care, less gardening, fewer restaurant meals, fewer airplane rides, fewer HVAC repairs, and fewer of several other services.
There are two lessons here, both about the unseen.
First, if at any time in our history, you had predicted that new jobs wouldn’t replace old jobs, you would have always been wrong. If you predict that AI will be the exception, you will probably be wrong. Second, implementing a UBI early last century to replace farm jobs or late last century to replace manufacturing jobs would have caused many valuable goods and services not to exist. We, therefore, would have been poorer for two reasons: higher taxes to pay for a UBI and fewer goods and services for us to enjoy.
The evidence so far
One piece of evidence that supports my optimism about future jobs is the unemployment rate. The latest report of the Bureau of Labor Statistics states that the unemployment rate in April was still a low 4.3 percent. Only a few decades ago, economists thought that the unemployment rate consistent with full employment was between 4 and 5 percent.
Moreover, we can already see what has been happening in labor markets that the skeptics feared would be most hurt by AI. We have often heard that AI will destroy jobs for software engineers. On its face, that claim has a certain plausibility. But it’s also wrong. Here’s what Alistair Barr wrote in an April 3, 2026, news story in Business Insider: “Data from TrueUp, a tech hiring analytics firm, shows more than 67,000 software engineering job openings, the highest level in over three years. Listings have roughly doubled since a trough in mid-2023.”
Why did that happen? A reasonable explanation is that because AI brought down the cost of producing software, more software was demanded. It’s true that fewer workers per quality-adjusted unit of software were demanded. But the greater amount of software seems to have increased the demand for software engineers to monitor and add value to the final product.
We’ve seen this movie before. In my 2004 review of Henry Hazlitt’s 1946 classic book, Economics in One Lesson, I retold his story about the effects of the cotton-spinning machinery invented by Richard Arkwright in 1760. At that time, England had 5,200 spinners using spinning wheels and 2,700 weavers, for a total of 7,900 people producing cotton textiles. But twenty-seven years later, the number of people producing cotton textiles had mushroomed to 320,000. Why? The machines brought down the cost of cotton textiles so much that sales of cotton textiles exploded.
Or consider coffee making. University of Chicago economics professor Alex Imas, in an insightful Substack post titled “What will be scarce?” writes:
If the entire economy is soon to be automated, with labor being replaced with increasingly more sophisticated capital, Starbucks should be a canary in the coal mine—the technology for removing labor from its stores and replacing it with automated capital has been around for years.
But, notes Imas, Starbucks is pulling back from automation because, according to CEO Brian Niccol, customers were missing the personal touch.
Conclusion
Can we know what kinds of jobs will replace the jobs that higher productivity destroys? No. But we couldn’t know in 1900 that manufacturing jobs would replace most agricultural jobs by 1980. And we couldn’t know in 1979 that service jobs would replace one-third of manufacturing jobs by 2019.
Of course, I could be wrong. AI could replace workers so effectively that the number of jobs shrinks. If you want to see extensive reasoning on this, read Brian Albrecht’s excellent May 7 Substack post titled “You are not a horse.” He doesn’t tell you whether AI will replace many jobs, but he gives strong reasons for thinking it won’t.
What if I’m wrong and AI does destroy, on net, a large percentage of jobs? What then? First, if that happened, productivity and, with it, real gross domestic product, would explode. That would then substantially reduce the federal budget deficit. Second, with the federal budget deficit falling to zero or even turning into a budget surplus, a UBI would be easier to finance, and we could do it then. But let’s not short-circuit the process and reduce economic growth by paying people who aren’t working. I’m humble about what we can know about the future of employment. Experts in AI, who know more than I do about AI—but less than I do about economics—should be similarly humble.
