ROGER PARTRIDGE: The Market for Doom: Why We Keep Predicting the End of Work

Last month’s Schumpeter Comes to Wellington argued that the outcry over public service cuts in New Zealand sits in a long tradition of failed catastrophism. This month’s Long Read widens the lens and asks why generation after generation makes the same mistake – and why warnings about AI destroying work are no more convincing than the warnings that preceded them.

Thirty years ago, Jeremy Rifkin predicted the end of work. Within a generation, he wrote in 1995, automation would leave most of the world’s workforce without work. The book was a bestseller, was translated into multiple languages and made its author a celebrity. The generation is now up. Global unemployment stands at about five per cent, employment-to-population ratios are at or near record highs across the OECD, and Rifkin has moved on to his twenty-third book.

Rifkin is the most recent figure in a long line. Paul Ehrlich, who died in March 2026 aged ninety-three, opened The Population Bomb in 1968 with the prediction that hundreds of millions would starve in the 1970s. India could not feed an additional 200 million; England would not exist by the year 2000. India became a net food exporter, and England is still there.

The Triple Revolution memorandum sent to President Johnson in 1964 by thirty-five intellectuals, including Linus Pauling and Gunnar Myrdal, predicted that cybernation would soon make work obsolete and that the link between income and employment would have to be broken. US unemployment hit 3.5 per cent five years later. The Club of Rome reported in 1972 that copper, lead, zinc and tin would all be exhausted before the millennium. Each prediction was made with confidence, marketed with skill and falsified within a generation. Each was succeeded by the next.

The latest is artificial intelligence and the end of work. We are told that this time the machines really will displace us, that the lawyer, the analyst, the accountant and the radiologist are about to follow the harness-maker and the telegraph operator into history.

The chorus is louder than the data. Tech-sector layoffs in the first quarter of 2026 ran 40 per cent ahead of the same period in 2025, and the firms making the cuts attributed the acceleration to AI. Yet, aggregate employment has not noticed. The unemployment rate has not moved. The Washington Post reported in March that there was no measurable evidence so far that AI was putting Americans out of work.

Daron Acemoglu, the 2024 Nobel laureate and one of the leading academic voices on AI’s labour-market risks, told an interviewer in mid-May that the macroeconomic data still shows no material impact on employment from AI. He was reaffirming an observation he had first made two years before.

What we are actually doomed to is a torrent of AI doomsayers.

The static mind

Our receptiveness to these predictions is so consistent that it appears cognitive before anything else.

Each prediction takes the world as it is, identifies a class of human activity about to be displaced, and projects forward an economy with that activity removed and nothing put in its place. Ehrlich saw a 1968 world with a given food supply and a given population growth rate; he projected forward to a world with insufficient food. The Club of Rome saw 1972 reserves and 1972 consumption rates; it projected forward to a world without copper. Rifkin saw 1995 productivity gains in manufacturing and services; he projected forward to a labour market without enough work to go round. Each prediction was a snapshot extended into a film.

But what no snapshot can show is the labour, capital and ingenuity that will be drawn into the gap the displaced activity leaves behind. New crops, new mines, new sectors, new occupations: these do not appear on the chart of what is, because they do not yet exist. The doom prediction extrapolates the chart forward without them. The actual economy adds the labour and capital that the displacement releases, and then whatever the released resources end up doing. The doom merchant works in two dimensions, the economy in three.

In the early 1980s, the introduction of the automated teller machine prompted confident predictions that the bank teller’s job would be eliminated within a decade. Tellers performed cash dispensing and balance enquiry; the ATM performed both faster, cheaper and around the clock. The prediction was correct about the displacement and wrong about everything that followed. With ATMs handling the routine work, branches became cheap enough to open in places that had not previously supported one. Banks opened more branches, each needing people to handle the work the ATM could not, including advice, loans, complex transactions and the cross-selling of products. By 2010, the United States employed more bank tellers than it had in 1980, in more branches, doing more interesting work. What the static prediction could not see was the new branch that would not have existed without the ATM, and the new work the teller in it would do.

The static mind is not a personal failing. It is a feature of how we reason about complex systems we cannot see as a whole. Karl Popper argued in The Poverty of Historicism that the future cannot be predicted in detail because the knowledge that will shape it has not yet been generated. The harness-maker is easy to picture. The mechanic who does not yet exist is not.

As I argued last month, this is French economist Frédéric Bastiat’s distinction between what is seen and what is not seen: the visible side of any technological transition is the work it displaces, the invisible side the work it makes possible. The first appears in the news; the second appears in the data a decade later, when the children of the displaced are doing it.

The species of writer who attends only to the visible side and builds a career on the warnings he draws from it has a recognisable shape. He is usually credentialled and often distinguished in an adjacent field, precise about the displacement and vague about the absorption. His books sell because the warnings find an audience attending to the same side – what Daniel Kahneman called the availability heuristic, the tendency to judge probability by what comes to mind first. He is the doom merchant. Ehrlich was the most famous of his kind, and Rifkin remains the most relevant to this essay. The chorus now telling us that AI will end the labour market is the present generation of them.

The economy in three dimensions

The first law of economics is the one the doom merchant forgets. Human wants exceed available resources, always, and at every level of material progress. The peasant of 1600 wanted shoes; the worker of 1900 wanted shoes, lighting, transport and a Sunday paper; the citizen of 2026 wants all of those plus broadband, dental care, foreign holidays and a yoga class. There is no level of supply at which the wants stop. When a technology releases labour and capital from one set of uses, the released resources do not pile up at the door. They are drawn into the satisfaction of the next set of wants, most of which the previous generation had not yet thought to articulate.

No central authority directs the labour and capital released by a technological transition. As Friedrich Hayek argued, the information is dispersed across millions of producers, employers and investors, and the price system aggregates it. Joseph Schumpeter – the focus of last month’s essay – called the process creative destruction. The destruction is the visible loss, the creation the dispersed gain.

The agricultural revolution between 1700 and 1850 displaced workers on a scale that dwarfs any subsequent change. The English agricultural workforce fell from a clear majority of employment to about a fifth, as enclosure, the seed drill and the threshing machine made the smallholder, the casual labourer and the threshing gang redundant in succession. The Swing Riots of 1830 burned the threshing machines across southern England. The displacement was real and the distress was real. But the country that came after had Manchester, Birmingham, Leeds and the Lancashire mill towns, real wages that roughly doubled between 1820 and 1900, and new occupations – the clerk, the engineer, the telegraph operator, the railway worker – that had barely existed at the start of the century.

The manufacturing-to-services transition between 1950 and 2000 ran the same pattern at a faster speed. The US manufacturing workforce fell from about a third of total employment to under fifteen per cent. Ewan Clague, an American labour economist, had predicted in 1935 that displacement would outrun reabsorption and that the surplus would be eliminated by age and death. In the early 1980s the Nobel laureate Wassily Leontief warned that workers would go the way of horses. So far, both have been proved wrong. The medical technician, the financial analyst, the management consultant, the marketing manager and the software engineer became mass categories of employment that had barely existed in 1950.

The information transition, beginning in 1995, was forecast in The End of Work as the moment the labour market broke. It coincided with twenty-five years of almost continuous US employment growth, broken only by the recessions of 2001 and 2008. Total US employment rose from 124 million in 1995 to 158 million by early 2020. New occupations no statistical office had categorised in 1995, including software developer, web designer, mobile applications developer, social media manager and data scientist, became standard listings. Rifkin’s prediction was falsified at its own stated horizon.

One case in the record runs the other way. Between roughly 1890 and 1930, the working horse population of Britain peaked at over a million animals and then collapsed to near zero within a generation. The displacement was complete in a way no human transition has been.

If he had asked his customers what they wanted, Henry Ford is supposed to have said, they would have said faster horses. The line is probably apocryphal. It captures what no doom merchant ever sees. The human workers in the horse economy moved into work no customer of 1890 would have specified, because no customer of 1890 could have. They became mechanics, fitters, traffic engineers, road builders, oil refiners, drivers, suburban developers and the early salesmen of an automotive industry that did not yet exist. The horses, who could not retrain, are the only species in the record where the doom prediction came true.

The future the doom merchant cannot frame

Acemoglu’s recent books have argued that the gains from new technologies are shared only when institutions and policies steer them that way. The implication, on this view, is that without active redirection AI may displace workers faster than the economy can absorb them.

A year after Acemoglu received the Nobel prize in economics for his work on the institutional foundations of prosperity, the 2025 prize went to three economists working on innovation-driven growth. Joel Mokyr received half of it for the historical work showing that the long-run effect of innovation has been to raise living standards on a scale the worriers of any given generation could not have imagined. Philippe Aghion and Peter Howitt received the other half for the formal modelling behind the empirical record: creative destruction does not merely subtract jobs from old sectors but reallocates resources to new ones, and the new ones, on average and over time, produce more than the old. Two prizes, in two successive years, frame the AI question as economics now understands it. Acemoglu warns of what could go wrong. Mokyr, Aghion and Howitt explain what has, transition after transition, gone right.

The empirical record is on the second side of the ledger. Acemoglu has been saying so for two years. His warning is conditional, prospective and policy-shaped: this could go wrong if we let it. The Mokyr-Aghion-Howitt account is empirical, historical and mechanism-based: this is how the process has worked across every transition in the record, and there is no reason embedded in the economics of AI to suppose it will work differently this time.

What the second account makes possible is a future the first cannot frame. If the labour released from analytical, drafting and processing tasks does not pile up at the door, where does it go? The static mind cannot answer the question, because the wants it serves do not yet exist. The dynamic answer is that it goes wherever the next generation of human wants turns out to be.

Some of those wants are easy to anticipate, because we already pay for them and would pay for more. We want better care for the elderly, more attentive teaching for our children, more skilled trades for our houses, more reliable mechanics for our cars, more interesting food, more personal training, more therapy, more hospitality, more craft. A generation ago, the nail bar barely existed in Britain. It now employs tens of thousands of people in work no central planner specified and no industrial policy promoted. The same was true once of the software engineer, who was not a mass occupation before the 1980s, and of the professional gamer, whose career did not exist before 2000. Neither would have appeared in any 1970 forecast of the labour market. The new occupations of 2050 are invisible to us for the same reason.

The relative price of this work has been rising for thirty years, and AI will accelerate the rise. As the cognitive tasks of the lawyer, the analyst and the accountant become cheaper to perform, the work that cannot be done by a machine becomes relatively more valuable. The masseuse, the carpenter, the carer, the chef and the live musician do work that requires a body, a presence, a judgement or a hand that the large language model cannot supply.

In the labour market the doom merchant cannot see, the good barista may earn more than the junior lawyer, and there is nothing in the economic logic that should make us regret it. The displaced cognitive worker who retrains as a physiotherapist or a guide or a craftsman is not stepping down. She is stepping into work the next generation will pay more for.

This is not utopia, and it is not painless. The fifty-year-old compliance officer at a London bank whose work is absorbed by AI in 2028 has a real problem in 2028, and so does her mortgage and her children’s school fees. The aggregate case for absorption does not erase her case.

Nor is Acemoglu wrong that institutions matter. The adjustment mechanism is not automatic and never has been. It runs through secure property rights, open markets, the rule of law, the freedom to start a firm and the freedom to leave one, prices allowed to signal where labour and capital are wanted, and education that produces workers capable of moving when the signals change.

Where those preconditions hold, the mechanism works, transition after transition, even against the resistance of existing institutions. Where they are deliberately suspended – Venezuela in the past quarter-century, Cuba across six decades, the planned economies of the twentieth century – the mechanism stops. Growth stalls, and the displaced do not get reabsorbed; they emigrate, queue or stand idle. The doom merchant’s predictions, made about capitalism, come true in the economies that abolish it.

This is the worry the doom merchant is too distracted to notice: not that AI will be different, but that we will.

The real dismal science

Economics is sometimes called the dismal science. The label has stuck for nearly two centuries, fixed by the assumption that there is something joyless about a discipline that counts and measures, that talks of supply and demand and marginal product when others reach for catastrophe. Economists are the killjoys who say that resources are scarce and that costs are borne even when they are not seen.

But the ledger runs the other way. The economists who took scarcity, comparative advantage and dispersed knowledge seriously have been the optimists of the past two centuries. The dismal predictions came from elsewhere, each confidently made and widely sold, each falsified within a generation, each answered eventually by economists pointing to wants, prices, knowledge and adaptation.

The label fell on the wrong science. The economics that explains how we muddle through is the antidote. Schumpeter, Bastiat, Mokyr, Aghion and Howitt are on one side of the ledger. The doom merchants of two centuries are on the other. Where the institutions held, economics has been right. Where they were suspended, growth stalled.

Jeremy Rifkin will publish another book. The next generation of doom merchants is already at work writing on AI. They will sell, and the economy they describe will continue to do what it has done wherever it has been allowed to: release labour and capital from old uses, draw them into new wants, and produce the country no static prediction could have foreseen.

Economics is not the dismal science. Humanity is the dismal species. The economists, when they are doing their proper work, are the ones reminding us why the dismal is almost always wrong – and what we must defend to make it right.

This essay is part of an ongoing series on liberalism, democracy, and the international order. Related writing in Persuasion, Quadrant, Quillette and on Plain Thinking is collected here.