The Machine

When a machine can do what you are paid to do, it does not merely compete with you, as another worker would, for a share of the wage; it threatens to abolish the wage altogether, for a task done by a machine is a task for which no one need be paid. This is the oldest fear in the history of industrial work, and it arrives, like a tide, in waves. In 1811 the skilled framework-knitters and weavers of the English Midlands and North, watching the new powered frames and looms turn out cheaper goods and put them out of their trades, did the direct thing: they broke the machines.¹1The Luddites (1811–13) — skilled textile workers — smashed the powered frames and looms that were destroying their trades. The breaking of machines was made a capital offence, and fourteen were hanged. David Ricardo, in the 1821 third edition of his Principles (ch. 31, ‘On Machinery’), reversed his earlier optimism and conceded that machinery could be gravely injurious to the labouring class. They were not fools raging at progress; they were craftsmen who understood, precisely, that their hard-won skills were about to be made worthless — and they were right, for their wages did vanish. The state hanged fourteen of them. A decade later the fear reached the calmest mind in the science: David Ricardo, who had taught in his first edition that machinery must benefit every class, added to his third, in 1821, a chapter recanting. He had become convinced, he wrote, that the substitution of machinery for human labour was ‘often very injurious to the interests of the class of labourers.’ And a century after that, Keynes gave the fear its lasting name, warning in 1930 of a ‘new disease’ — technological unemployment, the unemployment that comes when our skill at economising labour outruns our skill at finding new uses for it. The names change; the dread is one. And in our own day it has taken its most unsettling form, for the newest machines threaten not the hand but the mind.

And yet the dread has been, in its central prophecy, wrong — wrong for two hundred years, wrong with a consistency that ought to humble it. For the machine that can do your work has a second face. Where it does not replace the worker it complements him: the weaver at the power-loom made more cloth in a day than the handloom weaver could in a week, and the productivity that poured from the machine, spread through the economy, lifted wages it never touched. And by making goods cheap, the machine freed the spending of everyone to flow toward wants that had gone unmet — wants that became new trades, new industries, new kinds of work the displaced, or their children, could turn to. The proof is written in the largest single fact of the modern economy. In 1870 almost half of all Americans worked the land; in our own century, scarcely one in a hundred does²2In the United States, agriculture employed about 46% of the workforce in 1870 and roughly 1% today — a displacement of labour more total than any machine has yet wrought elsewhere, yet it ended not in mass idleness but in reabsorption into new work, as real wages rose many-fold across the two centuries of mechanisation. — a displacement of the workforce more total than any robot has yet wrought — and it produced not mass idleness but the richest society in history, the displaced absorbed, over generations, into work their great-grandparents had no names for. The lamplighter and the elevator operator vanished; the programmer and the radiographer appeared. Employment did not collapse. Across the whole long sweep, through every panic, real wages rose.

The machine is the rival and the benefactor both.

The error in the recurring fear has a name, and naming it is the beginning of the reassurance. It is the ‘lump of labour’ fallacy — sometimes, in honour of the machine-breakers, the Luddite fallacy — the belief that there is a fixed quantity of work to be done in the world, a lump, so that every task handed to a machine is a task taken from a human being for good. But the lump does not exist. The quantity of work to be done is not fixed; it grows, because cheaper production and higher incomes call forth new demands without end, and meeting them takes new labour. The clearest way to see this is to stop thinking of whole jobs and think instead of tasks, as the economists who study automation now do.³3In the task framework of Acemoglu and Restrepo (‘Automation and New Tasks,’ 2019), technology has two opposed effects: a displacement effect, as machines take over tasks once done by people, and a reinstatement effect, as new tasks are created for which human labour is needed. Whether employment falls or grows turns on whether reinstatement keeps pace with displacement. A machine does not usually swallow an occupation whole; it takes over particular tasks within it and leaves others — while the wider change calls into being entirely new tasks, and new occupations, that did not exist before. Automation thus works at two frontiers at once: it displaces labour from the tasks it absorbs, and it reinstates labour at the new tasks it creates. Whether work, on the whole, shrinks or grows is a race between those two frontiers — and for two centuries, the second has kept ahead of the first.

But the reassurance, however well-earned across centuries, conceals two hard things, and honesty requires holding them up to the light. The first is the matter of when. ‘In the long run,’ the optimist says, the machine has always been a benefactor — and it is true. But, as Keynes also wrote, in the long run we are all dead;⁴4‘In the long run we are all dead’ — Keynes, A Tract on Monetary Reform (1923), rebuking economists who console the present with the equilibrium of a distant future. His 1930 guess that the rich countries would be four to eight times better off within a century has largely come true in the aggregate — but the gains pooled unevenly, much of the rise accruing above the median rather than to the worker in the middle. and the short run, the only run a person actually inhabits, can be merciless. The handloom weaver displaced at forty did not live to see his grandchildren prosper in the mills; he lived to watch his wage fall to nothing and his skill turn to dust. The twenty thousand Nottingham knitters thrown out of work in a single year had no retraining scheme and no safety net; they were, in the plain phrase of the day, cast onto the scrap heap. The rising line the centuries draw is, seen close, a sequence of cliffs — and it is never the same people who fall and who climb. The gains are general and the losses particular, and a statistic that nets the one against the other describes no one who ever lived. Even the great enrichment Keynes foresaw arrived crookedly, much of it gathering at the top, so that the average climbed far faster than the wage of the man in the middle.

The second hard thing is whether this time is like the others at all. Every machine until now has had a ceiling to what it could take. The early engines took muscle, and labour withdrew into the work of the hand and the eye and the mind. The computer took routine — the predictable, rule-bound tasks, clerical and manual — and the labour market split, or ‘polarised’: the middle, where the routine work clustered, was hollowed out, while the top (the cognitive, the creative, the work of judgment) and the bottom (the non-routine manual work of care and service) grew, the displaced of the middle pressed downward more often than up.⁵5Job polarisation (Autor and Dorn, 2013; Goos and Manning, 2007): computerisation hollowed routine middle-skill work while high- and low-skill work grew, often pushing the displaced into lower-paid jobs. Artificial intelligence may break the pattern: studies of exposure (e.g. Frey and Osborne, 2017, and later work on large language models) find its reach greatest among higher-skill cognitive tasks — and expert forecasts diverge more widely than on almost any question in economics. But always there was higher ground — a domain of human judgment, imagination, and understanding into which the wage could retreat, secure in the faith that no machine could follow it there. The artificial intelligence of the present age is the first machine that climbs toward that ground. It writes, it draws, it diagnoses, it argues, it codes; the very faculties that were labour’s last redoubt — analysis, language, the weighing of judgments — are the ones it now contests. No one knows what this means. The forecasts diverge more wildly than on almost any question in the science: some see the same false alarm that has sounded for two centuries, the lump-of-labour fallacy in newer dress; others see the first machine for which the old comfort fails, because the high ground, this time, has no higher ground above it. It may be that new work blooms as it always has. It may be that this is the time the tide does not turn. The honest answer is that we are inside the experiment, and cannot yet read its result.

Always, before, there was higher ground.

There is one thing more the machine does to the wage, and it does it even to those it never replaces. It sets a ceiling. No worker is paid, for long, much more than it would cost to do without him — and the cost of the machine that could do his work is precisely that limit, the silent ceiling above his wage.⁶6A worker is not, for long, employed at a wage above the cost of the machine that could replace the job. The machine need not be installed to bind the wage; its availability, and its ever-falling price, cap what the worker can demand. As the price of machines falls — its one reliable tendency — the ceiling descends. He need not be replaced for the threat to bind: the mere existence of the machine, and its falling price, caps what he may ask, for he cannot demand more than his replacement would cost. And the price of machines does, over time, only one thing — it falls. So the ceiling descends, year upon year, pressing down toward the wage from above, whether or not the machine is ever installed. To earn a wage at all is, in this sense, to be paid in the shadow of one’s own replacement: the worker keeps his place only so long, and only so far up, as the machine that could take it remains dearer than he is.

This, then, is the last of the forces the folio has traced through this single price. We have watched the wage measure the worth of things and sell the hours of a life; stand as a pledge and fall under compulsion; settle by power and bound the world of work; refuse from dignity to drop; serve as the measure of all value; buy the obedience that makes a firm; and shelter the worker from the firm’s own risk. Here, at the last, we find the rival that shadows all the rest — the machine that can do the work, and so threatens, in the end, to make the wage itself unneeded. The wage has always lived in that shadow, and lives there still. What is new is only how high up the light the shadow has begun to climb.