Fighting for Growth: Labor Scarcity and Technological Progress During the British Industrial Revolution

Why did the Industrial Revolution begin in Britain? The question feels all the more relevant now that we’re staring at another technological revolution, artificial intelligence, that might dwarf it.

I was reading this really clever paper by Bruno Caprettini, Alex Trew and Hans-Joachim Voth. They start by laying out the two competing explanations. The economic historian Robert Allen argued that on the eve of the Industrial Revolution, Britain had high wages and cheap energy in the form of abundant coal, which made it uniquely profitable to adopt labor-saving technology. The Nobel laureate Joel Mokyr argued that it wasn’t just wages: Britain had an unusual concentration of skilled mechanics, craftsmen and tinkerers who could design and build practical machines.

So who’s right, Allen or Mokyr? According to the authors, both. And the way they test it is really clever. To see whether labor shortages drove the adoption of machines, they exploit a natural experiment. Between 1793 and 1815, Britain was at war with France. At the peak of mobilization, around 1813, the army and navy had about 350,000 men under arms, roughly 10 to 14% of the adult male labor force. Recruitment was uneven across the country, and in places that saw heavy enlistment, the population skewed noticeably towards women.

People noticed at the time. In 1812, an official survey of agriculture in Dorset observed:

A considerable number of thrashing [sic] machines have been erected in this county … the principal inducement for using them is a scarcity of labourers, which, in a state of warfare, may be expected to be felt most in maritime districts.

To make sure they were capturing genuine labor shortages, not something else about those places, the authors looked at where the navy did its recruiting. Much of it was done by press gangs sent ashore from warships, so it mattered how close the big ships could get. Coastal places near deep anchorages, where large warships could anchor close to shore, saw heavy recruitment. Places where the waters were shallower saw much less.

Why did anchorages matter? Because much of the navy’s recruitment was involuntary. Press gangs, often sent ashore by individual ships’ captains, forced men into service in the Royal Navy, frequently against their will. Only about 14% of the sailors in the authors’ sample were recorded as volunteers.

The authors use this labor shock to see what happened next, and they do a lot of clever legwork. To measure recruitment, they hand-collected Royal Navy muster rolls, the personnel records of individual ships, and digitized where each recruit came from. To measure mechanization, they combed through 60 regional newspapers for ads selling or leasing farms, which listed the machines on them, nearly 5,000 machines in all. Then, among coastal places, they compared those close to deep-water anchorages with those farther away.

The difference was marked. Places that saw heavy recruitment adopted labor-saving machines much faster, above all threshing machines. Threshing was brutally labor-intensive, and one 1832 account describes a machine doing in two months what used to take men ten. Machines that didn’t save labor, like turnip cutters, showed no such link with recruitment.

The other interesting finding: in places that also had lots of skilled craftsmen, like blacksmiths, millwrights and watchmakers, machines not only spread faster but improved faster too. And when the men came home from war, the machines stayed.

And here’s the funny, delicious irony: this is a paper about labor-saving technology, and the authors used a labor-saving technology to write it. They had OpenAI’s models read through thousands of old newspaper ads for farm sales, which is how they tracked the spread of machines, labor-saving and otherwise.

I think we’re about to see many more clever papers like this. LLMs have gotten very good at extracting information from messy, unstructured sources, and at analyzing data. All sorts of weird datasets that would have been impossible to build before LLMs are now within reach. Hopefully that gives us a much better understanding of the past.

The paper also has relevance for the AI era, where I think we have the opposite problem. We’re mostly at peace, and there are few labor shortages. The exceptions are places like Russia and Ukraine, where casualties run into the hundreds of thousands. That’s part of why both sides are leaning so heavily on drones, which have become cheap to produce in Ukraine, a country with far fewer men to throw into the war than Russia. But for most of the world, labor is abundant, and if AI makes it more abundant still by putting people out of work, I’m not sure what happens. B.