Ricardo's Ghost: Two Centuries of Being Wrong About Machines

From Loom to LLM, the Jobs Apocalypse Never Arrives

In 1817, the most rigorous economist of his age changed his mind in public, and in doing so he founded a tradition of error that we have not escaped to this day. David Ricardo had long taught that labor-saving machinery was a general blessing, good for the landlord, the manufacturer, and in the end the workman himself. Then, in the third edition of his “Principles of Political Economy and Taxation,” he inserted a new chapter titled “On Machinery” and confessed that his opinion had “undergone a considerable change.” Machinery, he now believed, could be “very injurious to the interests of the class of labourers.” The workers who feared the new looms and engines were not, he wrote, victims of “prejudice and error.”

This was a remarkable thing for Ricardo to say. Here was the man who systematized classical economics, who gave us comparative advantage and the theory of rent, granting intellectual respectability to the very fear that had sent the Luddites into the mills of Nottingham with hammers six years earlier. The first major Luddite riot had broken out in Arnold in March of 1811, and within months hundreds of machines lay broken across the English Midlands. The rioters already had a folk hero in the mythical General Ludd of Sherwood Forest, but after 1817 they had something better. They had Ricardo, and the hammer now had a theory.

I want to examine that theory carefully, because everything that follows from it, straight through to the present panic over artificial intelligence, is a variation on a single mistake Ricardo made in that chapter. The mistake is not moral and it is not stupid. It is mechanical, and it is seductive precisely because it wears the clothing of rigor.

Ricardo built a small numerical model. He imagined a capitalist with £20,000 who employs that sum as circulating capital, the fund that pays wages and buys the raw materials labor works upon. Now suppose this capitalist converts part of that fund into fixed capital, a machine. He still produces goods, but he now pays out less in wages, because the machine has taken over work that hands used to do. Ricardo showed, correctly within his own arithmetic, that the “gross produce” of the economy could fall even as the “net produce,” the profit and rent left over after wages, could rise. And since it is the gross produce, the wage fund, that supports the working population, that population could become, in his word, “redundant.”

Notice what makes this argument feel airtight. It is internally consistent, the numbers add up, and the reader arrives, almost against his will, at the conclusion that the machine has thrown men permanently out of work. This is the static fallacy in its purest and most respectable form, and it is the direct ancestor of every modern forecast that counts the tasks a computer can perform, subtracts them from a fixed pile of human work, and reports the remainder as unemployment. The Oxford economists Carl Frey and Michael Osborne did exactly this in 2013 when they concluded that 47% of American jobs were at risk of automation within two decades. They assigned probabilities to 702 occupations and summed the result. The arithmetic was impeccable. The premise was Ricardo’s.

So where is the error? It lies in what the model holds still. Ricardo treated the wage fund and the level of consumer demand as roughly fixed quantities, as a pool of given size from which a machine simply removes a bucket. But an economy is not a pool. It is a circulating system, and the act of installing the machine sets in motion three forces that Ricardo himself glimpsed, half acknowledged in the very same chapter, and then drastically underweighted.

Consider the first force. The capitalist who saves on wages does not bury the money in his garden. His higher profit becomes new savings, the new savings become new capital, and the new capital employs labor somewhere else, often in a trade that did not exist the year before. Ricardo conceded this point explicitly. He admitted that the funds saved would be reconstituted as capital and would set additional labor in motion. He simply assumed the process would be slow, partial, and painful, and he let that assumption swallow his conclusion.

Consider the second force, which is the deepest of the three. When a machine makes goods cheaper, the people who buy those goods are left with money to spare, and that surplus is real income they did not have before. A family that spends less on cloth can afford shoes, and then a hundred things the cloth-maker never imagined. Ricardo knew this, for he had read Adam Smith on the boundlessness of human wants, the observation that the desire for food is limited by the narrow capacity of the stomach but the desire for conveniences and ornaments seems to have no limit at all. Rising real income does not sit idle. It chases new satisfactions, and the chase creates the industries that absorb displaced labor.

Consider the third force, the one no model can capture because it concerns the goods and trades that do not yet exist. When the cost of doing something collapses, we do not simply do the old amount more cheaply and pocket the difference. We do vastly more of it, and we do new things adjacent to it. William Stanley Jevons noticed this with coal, that more efficient engines burned more coal rather than less, because efficiency made coal worth using everywhere. Lower the cost of an activity and you do not shrink it, you multiply it, and the multiplication is where the jobs are.

Here is the decisive point. History did not leave Ricardo’s model as a matter of speculation. It ran the experiment, at the largest scale imaginable, for two centuries, and the three compensation forces won every time. Take agriculture, the greatest labor displacement in the history of the US. In 1900 roughly 41% of the American workforce lived by farming. By 2000 the figure was 1.9%. The tractor, the combine, and the chemical revolution did to the farmhand precisely what Ricardo feared the loom would do to the weaver, on a scale a hundredfold larger. Had the static fallacy held, the result would have been tens of millions of Americans permanently idle, a redundant population on a continental scale. Instead the displaced moved into factories and then into offices, the nation that once needed 40% of its hands to feed itself learned to feed the world with under 2%, and the labor set free built the richest economy the species has ever known.

Take a cleaner case still, because skeptics will say agriculture took a century. When the automated teller machine arrived in the 1970s, every sensible person knew it would exterminate the bank teller, for the machine did the teller’s central task, the handling of cash, and it did it tirelessly. The economist James Bessen documented what actually happened. Banks installed hundreds of thousands of ATMs, and the number of tellers did not fall. The machine made each branch cheaper to run, so banks opened far more branches, and urban branches rose 43%. Fewer tellers per branch, many more branches, and so more tellers in total. In 1985 the US had 60,000 ATMs and 485,000 tellers; by 2002 it had 352,000 ATMs and 527,000 tellers. The machine built to kill the job grew the job, and shifted it upward toward sales and service, exactly what the compensation forces predict and exactly what the static model cannot see.

And the panic of our own decade is falsifying itself in real time. In 2016 a Nobel laureate and pioneer of machine learning told the world to stop training radiologists, declaring it completely obvious that the software would surpass them within five years. A decade later the Mayo Clinic employs more than 400 radiologists, a 55% increase, even as a national shortage pushes salaries toward $571,000. The man was brilliant, and he was wrong in the identical direction Ricardo was wrong, for the identical reason.

The broader data tell the same story without drama. The Budget Lab at Yale studied the first 33 months after the launch of ChatGPT and found stability, not disruption, across the whole economy, with no clear link between AI exposure and unemployment. As of May 2026 the US unemployment rate stands at 4.3%, holding within a narrow band for nearly a year. This is the number against which 200 years of prophecy must be measured, and it is a refutation.

Most telling of all, the prophets are recanting. The chief executive of OpenAI now says he was “pretty wrong” about the destruction of entry-level jobs and is “delighted to be wrong.” The author of last year’s prediction that AI would erase half of entry-level white-collar work now reaches instead for Jevons, observing that if you automate 90% of a job, everyone simply does the remaining 10%, which then expands. When the men selling the apocalypse abandon it on the eve of their own offerings, the careful observer should notice.

None of this means the transition is painless. The weaver in 1817 suffered, the farmhand in 1930 suffered, and some clerk in 2026 will suffer too, and a decent society owes them seriousness rather than slogans. But suffering in transition is not the same thing as permanent mass unemployment, and the policy that confuses the two is the most dangerous policy of all. To freeze technology in place, whether with the Luddite’s hammer or the regulator’s pen, is to protect a handful of existing jobs by impoverishing everyone else, forfeiting the new industries before they can be born, and handing the future to rival nations who decline to be so sentimental. Ricardo himself saw even this. He warned that a country which refused the machine would not save its jobs but merely export its capital, and its jobs with it.

The vocabulary keeps improving. We have traded looms for large language models and redundant populations for displacement curves. The fallacy underneath has not changed since 1817. It is the belief that the quantity of human work is fixed, that a machine which performs a task therefore subtracts a job, and that the subtraction is permanent. Two centuries of relentless mechanization, ending in near-record employment and unimaginable wealth, have answered that belief as thoroughly as history ever answers anything. Ricardo’s ghost is still in the room, and he is still, with the greatest possible rigor, wrong.

If you enjoy my work, please subscribe https://𝕏.com/amuse/creator-subscriptions/subscribe

---

Alexander Muse is a Fellow at the John Milton Freedom Foundation and publishes daily political analysis at amuseonx.com. Primary sources cited in this piece are linked inline; campaign finance figures are drawn from FEC filings, polling data from publicly released crosstabs, and legal claims from filed pleadings. Corrections are posted to the original URL with a dated changelog. Readers who identify errors are invited to contact the author directly. Data in sponsored partnership with Polymarket.