Will AI take my job? Start with what already happened to pay

Most of the worry about artificial intelligence is about the future. Will the machine take my job, and when.

It is the wrong place to start. Something already happened to the relationship between work and pay, decades before any of this technology existed. Understand that, and the AI question gets much easier to think about.

The thing that already happened

Picture two lines on a chart, running from 1948 to today. One is how much an hour of work produces. The other is what an ordinary worker takes home for that hour, after inflation.

Until 1973 they move together so closely they look like one line. Production per hour rises 96.7 per cent over that quarter century. Pay rises 91.3 per cent. Pay was bolted to production, and everyone assumed that was simply how an economy worked.

In 1973 the bolt shears. Production keeps climbing. Pay flattens. Through everything that followed, three recoveries, two booms and years of very low unemployment, the lines never came back together.

People have been producing more and not being paid for it for fifty years. That is not a forecast. It happened to your parents.

How big, honestly

Economists have argued about the size for forty years, and the argument gets used to wave the whole thing away, so it is worth knowing.

The largest estimate says that between 1973 and 2014 production per hour rose 72.2 per cent while typical pay rose 9.2 per cent. Critics say the two lines are adjusted for inflation using different baskets of goods: one follows the prices of what workers buy, the other the prices of what firms produce. It is like judging whether a salary kept up by comparing it to the price of televisions rather than the price of rent. Televisions collapsed. Rent did not.

They have a point, and on its own it is worth about a third of the gap. So use the number nobody disputes: after every correction the critics have proposed, this one and all the others together, twenty to thirty percentage points still remain. That floor is conceded by the sceptics, and it is still one of the largest transfers of income in modern history. It is not only American either. Across fifty countries, the share of national income going to people who work for a living has fallen since the early 1980s.

What it looked like from the factory floor

Outside Marseille there was a tea-packing plant, bought by Unilever in 1972. Over two decades the workers rebuilt the production line themselves, redesigning machinery and fixing the faults that caused the worst stoppages. Without consultants or new investment, they made the line more than 50 per cent faster. The plant became considerably more profitable.

Asked later what the company did with the money, one worker said it did not reach them, and did not go into hiring or machines either. It went into managers sent from headquarters, looking for something to justify the post. Eventually one of them proposed the obvious efficiency: close the plant, dismiss the workers, move production to Poland.

The plant closed in 2010. The workforce occupied it for 1,336 days and in 2014 won the assets outright, turning it into a cooperative that still trades.

Hold on to that story. The workers were not replaced by a machine. They made themselves more productive, and the gain went elsewhere.

Where the money went

Some went to the top. In 1965 the chief executives of America's largest firms earned about 21 times what an ordinary worker earned. By 2024 it was 281 times, and most of it arrives as stock rather than salary. Two people work at the same bakery: one paid for her shifts, the other paid in slices of the bakery. Only one gets richer when the bakery does.

Some went to whoever owned the company.

Most went somewhere less visible. Think of a high street where every shop keeps paying its staff the same share of takings, year after year, and no shopkeeper ever decides to pay less. Over thirty years the shops with twenty employees lose their customers to a new kind of shop serving the whole town with three. The street's wage bill collapses anyway.

That is roughly what the American census data show. When a team of economists led by David Autor went inside the records firm by firm, they found the average firm had barely reduced the share going to its workers. Around 96 per cent of the decline came from activity moving to firms without large payrolls. Those firms often pay their few staff very well. There are just not many of them.

The part that matters for you

The comfortable explanation is that workers stopped keeping up: skills fell behind, so retrain.

The evidence does not support it. Over the same fifty years the workforce got older, more experienced and considerably better educated. By the standard measure, labour quality rose slightly faster after 1973 than before. People did not become less productive.

They became less able to claim a share. Picture a firm's profit as a pot in the middle of a table, with workers, managers, owners and the taxman around it. What each takes depends less on what they contributed than on the strength of their claim. What was dismantled between 1980 and 2010 was not anyone's contribution. It was the seats.

That decides everything about what to do next. If the problem is skills, the answer is training. If the problem is the claim, training is a distraction and the answer has to be ownership.

So what about AI

Now the fear can be stated precisely.

For fifty years firms took a growing share of what workers produced while the worker was still in the building. That took effort: Marseille had to be threatened with Poland, the union had to be broken, the seat had to be removed. There was always someone on the other side of the table who could refuse.

What is different now is that the effort is no longer needed. In June 2026 one technology company raised roughly 80 billion dollars in days, against a capital-spending programme of 185 billion. Not a cent of it carried a wage claim, because what it funds does not need a workforce to bargain with.

So the honest answer to the fear is this. The machine is not doing something new to you. It is finishing something that started in 1973, and finishing it faster. Which also tells you where the answer is not: another training course. It is in who owns the machine.

---

Where the evidence lives. The fifty-year evidence, the critics' corrections and the Marseille story fill chapter 4 of The Horse Is Here to Stay. The running record of new events as they confirm or complicate the argument is kept in the Stable.

One of a series of free guides to the chapters of The Horse Is Here to Stay. The book carries the evidence, the sources and the counter-arguments in full. These guides are published under a Creative Commons Attribution-ShareAlike licence: translate them, teach from them, quote them, build on them.