Haruki Murakami wrote a memoir about running called What I Talk About When I Talk About Running. Until I googled it, I thought the title was What I Think About When I Think About Running. I had been recommending it under the wrong name for years, which tells you both about how carefully I read book covers and what I wanted the book to be about.

I run short distances: usually ten kilometres, in Jonkershoek, in the early morning as the sun rises over the peaks. It’s spectacularly beautiful. These are also the best times to think. I think about a lot of things while running. My research. My family. Blog post topics. More research ideas, ideas that I’m confident will be publishable in a top journal. An idea to start a new company. Wait, that’s actually a brilliant idea. Why has no one founded a company like that? A bug walking across the road. I’m getting tired; my leg hurts. Oof, I’m not sure clients will pay that much for my great idea. Another bug. (There must be some poop nearby.) I need to remember to renew the family Wildcard, otherwise I’ll have to pay to get into Jonkershoek. Oh, and the car licence while I’m at it. The research idea won’t work because of endogeneity issues. You get the point.

Which is why I find it fascinating when people say that ‘writing is thinking’. No, writing is sitting down and forcing yourself to think about something. Which is why you think. It is not the writing that does the thinking. It is the thinking that does the thinking. You don’t need the writing to do the thinking, but the discipline of the writing helps the thinking, yes.

So I wrote a paper about it…

Read more

Over the last two months, I have accepted almost every invitation to review academic papers. On almost every one, I recommended rejection. Nearly all had been written with a large language model, which is fine by me, since I, too, use Claude Code or Codex for every paper I write. The trouble was that they were terrible in the same way, with an abstract opening on a question, an identical section order, irrelevant and old literature, and so much unnecessary mathematics (even in econ history).

In fact, I was not capable of following the math in most of the papers, so my own model would check the algebra. In almost all cases the result reported in the text was the opposite of the result the model had derived. Few of the authors had taken the trouble to actually check – themselves or with a different model – whether the math actually works out. The worst was a submission to a well-known journal with a section headed ‘placeholder results’, explaining that the next step was to download the data and run the analysis. The three figures were blank. The authors had not even taken the trouble to check their own paper for completeness. (The editor should have caught it, sure, but editors are drowning too, so I have sympathy for that oversight.)

Which is why the news of 6 August was good news. Refine – an AI reviewing tool built by the economists Ben Golub and Yann Calvó López, and one I use far too heavily – announced partnerships with the American Economic Association and the Econometric Society. Both now run its technical verification before publication, while mistakes are still cheap to correct. Earlier this year, John Cochrane predicted it:

Most referee reports do not identify the major point of the paper, and do not assess if the paper backs up that point. They do not notice glaring gaps of logic, basic theorems violated, econometrics advice 101 ignored. Editors are lucky if one out of three reports is vaguely useful. Clearly, this task is going to be radically impacted by AI. If I were an editor, I’d feed every paper to refine on receipt. I will surely get refine’s opinion before any referee report I write in the future.

But the reaction on X was less settled. Joshua Gans thought it belonged at submission (as Cochrane had proposed), where it can still change the paper; others asked whether we want a private toll on submission at all. (Zachary Horvitz found that renaming a file to ‘paper_final_draft_ready_for_review.pdf’ raises its score, which tells you roughly where we are.) And behind it sits a more radical (hopeful) claim, made, for example, by Adam Mastroianni: the journals are finished, and publishers storing PDFs at a forty per cent margin is unconscionable.

The pathology is real but the prognosis is wrong. Journals are about to become more important, not less, and having spent two months writing three working papers on the question, let me say why.

Since ChatGPT, new economics papers on arXiv have almost doubled, while NBER working papers, where posting is gated by affiliation, have risen about fifteen per cent. But there is no more expert attention this year than last, and there will be no more next year. Economics rations that attention by making us wait, which averages more than two years to acceptance and buys the journal no extra capacity. A fee would ration it at entry instead, and could pay the referee for the reading. So how many journals actually charge? That is what the first paper, ‘The Price of Submission’, set out to measure. In July I collected the fee and referee-payment policy of the 500 highest-ranked economics journals. (Claude agents did the collecting for a few dollars, a small demonstration of the thing the paper is about.)

One journal in five charges a fee, rising from 2 per cent of the lowest-ranked fifty journals to 68 per cent of the top fifty, at a median of $143. Now ask a different question: what price would actually clear the queue? The paper will not be pinned to a dollar – it answers in shares of the prize a placement is worth – so the figures here are my own back-of-the-envelope: about $412 at the top of the ranking today, and perhaps $1,555 in five years if capability keeps climbing, with the referee wage that balances the books rising from $206 to $799 a report. Those are mine, not the paper’s. They measure pressure on the queue rather than anyone’s price list, and the gap between the $143 charged and the $412 I get is the pressure. What I would watch is the line above which journals charge authors and pay referees, and below which they sell open-access publication instead. On the fees we can actually compare with 2019, most have gone up. And since a $1,555 fee reads differently from Stellenbosch than from Stanford, the eighty-eight journals already waiving fees for poorer countries matter.

If a machine can do the checking, what is a journal for? That is the second paper, ‘The Value of Peer Review and the Reward to Reputation’. A journal reads carefully what it can and deals with the overflow in some other way. In the past, that meant rejecting papers unread; now it can mean letting a machine read them. Both routes produce the same public stamp, but here the intuition becomes less obvious. Every paper the journal declines to certify joins the pool of uncertified papers against which that stamp is valued. So when a journal rejects good papers unread, it improves the quality of the uncertified pool and thereby weakens the value of its own certificate. A cleaner certificate can therefore be worth less than a dirtier one. At the paper’s illustrative parameters, the two rules cross exactly once, when about two submissions in five are good.

The result that matters for Refine is sharper still. Error rates on a codifiable screen end up reflecting our own behaviour. Once everyone has access to the same tool, authors will run it before submitting and fix whatever it flags. That is exactly the point the economist Luis Garicano made when the partnerships were announced: papers will already have passed through the AI referee, so it will have little new to find. At a selective journal, where few submissions clear the bar, this kind of repair roughly halves the value of a machine-screened stamp. Rejection unread, however, survives at any repair cost, and so does human judgement. Expert judgement remains the only test an author cannot rehearse in advance. The same is true of a grant panel or a hiring committee.

The third paper, ‘AI and the Research Team’, asks what happens to us. Think of a laboratory as one leader’s judgement combined with execution that can be scaled up. Better AI then pulls in two directions. It makes execution cheaper, so the laboratory grows, but it also automates codifiable tasks, so each unit of execution requires fewer people. Put those effects together and team size can turn at most once. In some fields it rises and then falls; in others it simply keeps rising. What determines the difference is how much of the work remains irreducibly human. Where none does, the team eventually shrinks to one person. Perhaps that is the least intuitive result in the three papers: the fields where machines help most may end up with the fewest scholars in the room.

Which fields? I scored 3,600 abstracts across 30 fields for how much physical execution they require, fixing the scores before looking at any outcomes. Seven of the thirty fall below the threshold. Mapping the METR capability trend onto research tasks then gives a date for the turn: 2027 is the last year in which research teams grow in algebra and number theory and in theoretical computer science, or 2028 if you allow for the slow adoption of new tools. Economics clears the threshold by just 0.001, which is close enough to a coin landing on its edge that we should be cautious. None of this is confirmed yet. In a panel of 2,279 principal investigators running to the end of 2025, the estimated effect is 0.19 with a standard error of 0.26. That is roughly what we should expect from data that end before the predicted transition begins.

The same divide runs through all three papers. AI has made producing research very cheap. It has not made checking research cheap, or judging it, or deciding which question is worth asking. Mastroianni is right that publishers extract too much for too little, but I think he draws the wrong conclusion. The scarce input in science is increasingly what a good journal can still provide: someone who has read the paper carefully and is willing to say what they make of it. There will be fewer such people, they will be paid, and their judgement will become more valuable.

Which brings me back to the paper with the blank figures. Refine.ink would have caught the problem in seconds, and the paper should have been stopped long before it reached me. But someone still had to decide that the paper should not exist. No machine made that judgement. I did.

‘The Price of Submission’, ‘The Value of Peer Review and the Reward to Reputation’ and ‘AI and the Research Team’ are on the working paper shelf. All three were written with Claude and Codex, and say so on their title pages.

For the last eight or so weeks, I have worked sixteen-hour days. I have enjoyed almost every one of them. My shoulders have been in spasm for almost the entire time.

It has taken me most of those weeks to see that those three things fit together.

Over the last six months, my workflow has undergone a radical transformation. I can now ask questions that would once have needed a research assistant, a year of their time and a grant I did not have. Alex Imas, the behavioural economist at Chicago Booth who became Google DeepMind’s Director of AGI Economics this year, knows the feeling: ‘I am not writing more papers, I’m writing very different papers, ones I would not have been able to write before. This is incredibly exciting (for me, maybe not “the world”).’

The gain is largest for those of us from the centre. Colleagues in Chicago or Cambridge have always had graduate students and a library that owns everything. At the southern tip of Africa the binding constraint was never a shortage of questions – it was capacity. A huge chunk of that constraint has now dissolved. The effect is intoxicating.

Intoxicating is the word to worry about…

Read more

Pour yourself a glass of milk. Whether your body can handle it – whether the sugar in it passes through you or leaves you doubled over by mid-afternoon – was largely settled by an economic decision your ancestors made around six thousand years ago. Somewhere in western Eurasia, people started keeping cattle for their milk as well as their meat, drawing on it day after day through the lean months. Those who could still digest that milk as adults had a little more food, a little more calcium, a slightly better chance of raising children who could do the same. A single change in the DNA near a gene called LCT began to spread. It is still spreading. Most adults on the planet cannot drink milk comfortably; if you can, you are carrying the result of that ancient shift to dairy farming in your DNA.

We do not usually think of the economy as something that changes our bodies. Economics is supposed to work on top of biology: we are born with fixed bodies and fixed appetites, and history is the story of what we do with them. Three papers published this year, all using ancient DNA, show that the biology itself has been changing – and that what changed it was the economy, meaning how people fed themselves.

Today, for the first time, we can measure human evolution as it happened, generation by generation, in the DNA of people who died thousands of years ago. Ali Akbari, David Reich and their colleagues assembled DNA from 15,836 ancient West Eurasians and looked for alleles – individual versions of a gene – whose frequency rose steadily over thousands of years. A steady, one-way rise is what natural selection looks like in the data: a version of a gene becoming more common, generation after generation, because the people who carried it survived and had more children. The old assumption was that human evolution more or less stopped once culture took over. The opposite turns out to be true. Over the past ten millennia it sped up. Their method flagged roughly 479 regions of the genome bearing the mark of strong natural selection – about twenty times as many as earlier studies could detect.

Back up first to the economy before farming. As the economist Ola Olsson put it on the Our Long Walk podcast last year, human prehistory runs in a slow pulse: in warm periods people spread out and their numbers grew, in cold periods they fell back into a few fertile refuges. When the last such cold reversal, the Younger Dryas, forced people in the Fertile Crescent to crowd around shrinking patches of wild grain, some began, half by accident, to cultivate it. The first farms were a response to scarcity.

The ability to digest lactose in adulthood appears to fit the same pattern. Milk-fat residue found in clay pots indicates that humans were likely consuming dairy before lactase persistence was widespread. Fermenting dairy – making cheese and yoghurt, for instance – reduces its lactose content, making it easier to digest. But it was in marginal regions where crops were hard to rely on – high latitudes, high altitudes and drought-prone areas – that dairy became an important ‘fallback’ food. In 2023, Stock and Wells postulated that selection for lactase persistence may have been particularly strong in northwestern Europe because the cold climate inhibited fermentation, making it more likely that raw milk was consumed in periods of famine and disease.

Yet this is just one of the many ways farming changed us. This is the heart of the story, and the figure below, drawn from a new Nature Genetics paper, lays it out plainly enough. Each block is a piece of DNA that changed because of how people lived; the deeper the colour, the stronger the change; time runs left to right, from big-game hunting to the present.

Read it from the bottom up and the argument tells itself. New foods called for new digestion: AMY1 to break down starchy grain, FADS to make scarce fatty acids from a plant-heavy diet, LCT for milk. Farming reduced dietary sources of vitamin D. For humans living under weak European skies, lighter skin improved the body’s ability to synthesise vitamin D, and so genes for lighter skin, such as SLC24A5, became common across the north until they were almost universal. Then people crowded together in villages and, later, cities, living close to their animals – and the immune system came under intense new pressure. Alleles that gave protection against tuberculosis and, later, plague rose and fell as the diseases themselves came and went. One variant, ERAP2, became more common after the Black Death. The coloured blocks cluster exactly where the economic revolutions cluster: at the Neolithic and the Bronze Age. The genome, in other words, is itself a record of economic history.

All of this evidence – the figure included – relates, of course, to West Eurasia: Europe and the Near East, where we have the best ancient DNA evidence. The particular genes are a regional story. Light skin was favoured under cloudy northern skies, and the European milk mutation is only one of several. East African cattle-herders evolved their own, independent lactase persistence; across malarial Africa it was blood variants such as sickle-cell that rose and fell. As Madeleine Bleasdale and others wrote in 2021, though Africa hosts the greatest diversity of genetic lactase persistence, a lack of ancient genomic data means it remains ‘vastly understudied’. Our species’ deep roots are African – the out-of-Africa journey began in refuges like the Cape coast, as Olsson reminds us – yet the ancient genomes we can read best so far belong to the north. The economic logic is universal; the specific genes it changed differ from place to place.

There is a second twist, and it is the one that should most interest economic historians. If the economy shaped our DNA, then that DNA is now a record economists can read back. In a new NBER working paper, Peter Huybers, Marco Tabellini, Charles Taylor and Francesco Toti measure prehistoric migration from genetics alone. When two people buried far apart share a long identical stretch of DNA – what geneticists call identity-by-descent, a segment inherited intact from a shared ancestor – someone in that family line must have made the journey between them. Counting those shared segments across seven thousand ancient individuals, they reconstruct who moved where over ten thousand years, the average genetic link spanning some 1,550 kilometres.

They find that people sorted themselves across the map according to the kind of knowledge they carried. Their model treats subsistence skill – knowing how to grow this crop in this growing season, or graze this herd through this rainy season – as a form of human capital that loses its value the further you carry it from the climate it was learned in. Migration falls off sharply with ecological distance: a standard-deviation difference in temperature, rainfall or soil cuts the flow of people by a fifth or so. And the binding constraint depends on how you make your living. Farmers move to match temperature and soil, and barely care about rain; herders match rain, and barely care about soil. When the world warms, farmers advance; when it cools, herders spread. Maria Elena Castiello and colleagues, using machine learning on 3,416 Neolithic sites, find the same logic at ground level: as the climate shifted, farmers rebalanced their crop portfolios and climbed to ever higher, cooler ground. These are households managing risk and reallocating their effort – ordinary economic behaviour – six thousand years before anyone wrote a price down.

So what is the lesson? We inherit the deep past in two forms at once. The first is knowledge and institutions – the reason, as Olsson and Christopher Paik have shown (and as chapter three of Our Long Walk to Economic Freedom recounts), that where farming and states first appeared still shapes national incomes today. The second is our own DNA. Both were produced by the same thing: the daily work of making a living. Economics usually treats the human body as fixed – the unchanging background against which its story plays out. These findings show that the background itself was changing all along. Which brings us back to that glass of milk. Whether or not you can drink it is a small, personal record of an economic revolution – evidence that the way our ancestors made their living shaped the bodies we still carry today.

Akbari, A., Perry, A., Barton, A. R., … Mallick, S. & Reich, D. (2026). Ancient DNA reveals pervasive directional selection across West Eurasia. Nature, 654; MemarMoshrefi, D., Johnson, O. L. & Huber, C. D. (2026). Insights into human adaptation from ancient DNA. Nature Genetics; Huybers, P., Tabellini, M., Taylor, C. A. & Toti, F. (2026). Climate and Prehistoric Migration. NBER Working Paper 35371; Castiello, M. E., Russo, E., Martínez-Grau, H., Jesus, A., Prats, G. & Antolín, F. (2025). Understanding the spread of agriculture in the Western Mediterranean with machine learning. Nature Communications; Stock, J. T. & Wells, J. C. K. (2023). Dairying and the evolution and consequences of lactase persistence in humans. Animal Frontiers, 13(3), 7–13; Bleasdale, M., et al. (2021). Ancient proteins provide evidence of dairy consumption in eastern Africa. Nature Communications, 12, 632; Our Long Walk podcast, ‘Why did we stop roaming’ (with Ola Olsson): https://www.ourlongwalk.com/p/why-did-we-stop-roaming

By the end of 2024, roughly one in ten American adults had used a new class of weight-loss drug. The medicines are GLP-1 receptor agonists – you will know them as Ozempic, Wegovy and Mounjaro – and they were first developed to treat diabetes. But a technology built for one purpose rarely stays within it. At higher doses, clinical trials show, these drugs strip 12 to 18 per cent off body weight, far more than diet or exercise reliably manage. Women took to them fastest: about one in sixteen had used one to lose weight, and among women who had ever been obese, roughly one in six.

Why should an economist care? Because economists have spent decades documenting that heavier people – women especially – face weight-based discrimination in wages, work and marriage. A drug built for diabetes, now used to lose weight, could therefore reach into employment, marriage and household income – places its inventors never intended it to go…

Read more

On the night of Sunday 7 March 1875, every household in the Cape Colony had to account for itself. The enumerators counted 720 984 people – 369 628 men and 351 356 women. Most were asleep in houses and huts; 4 666 were in tents and wagons, and 1 673 aboard ships lying in the bays. The clerks then spent the better part of two years adding it up.

Last Thursday, Statistics South Africa published its mid-year estimates for 2026. There are now 63 522 380 of us. That is the number reported in the newspapers. The number that will matter more is 2,12.

That is South Africa’s total fertility rate – the children the average woman will bear in her lifetime. Demographers use 2,1 as the replacement rate, the level at which one generation exactly replaces the one before it, with a small margin for those who do not survive to have children of their own. Because our child mortality is still higher than a rich country’s, our true replacement level sits a fraction above 2,1. Which means South Africa may already have slipped below it.

You can see the consequence in the age structure. For five years now the number of South Africans under fifteen has been shrinking in absolute terms: this year there are 0,34 per cent fewer children in the country than last year, while the population over sixty grows at 2,66 per cent. The number of children last fell in the early 2000s, at the height of the AIDS pandemic. This time the arithmetic runs the other way: fewer are being born.

Or let’s put it another way: On 17 April 1904, six years before the Union existed, the four colonies that would become South Africa each held a census on the same day. Together they counted 5 174 827 people. Every count since has been bigger. The line survived two world wars, the Depression, and a century of governments moving people around the map like furniture.

It fell exactly once. Between mid-1918 and mid-1919 the official estimate dropped from 6 769 000 to 6 727 000 – some forty thousand people – as the Spanish influenza came through the ports and the mine compounds. One year in a hundred and twenty.

The line has its embarrassments too. In 1960 the census found roughly a million more people than the official estimates had allowed for: the statisticians had simply lost track of how many Africans there were. And between 1976 and 1994 the state carved the Transkei, Bophuthatswana, Venda and Ciskei out of the country, declared them foreign, and stopped counting their inhabitants as South African. On census day in 1991 that came to 6,75 million people – more than the whole population of the Union in 1911. To be counted is to count, and for eighteen years millions were deliberately subtracted.

The Cape tells the same story over a longer period.

In 1865 the Cape Colony counted 496 381 people. Today the Western, Northern and Eastern Cape together hold 16,1 million – more than thirty-two times as many. Much of the early leap came from annexation: British Kaffraria after 1865, Griqualand West in 1880, the Transkeian territories over the following decades. The rest is people.

Here is the part that gets lost in the headline: South Africa’s population is still growing. What has changed is the reason: South Africans are staying alive longer.

In 2002, 59,9 of every thousand babies born here died before their first birthday; today it is 23,7. Life expectancy at birth sank to 54,7 years in 2005, at the worst of the AIDS pandemic. StatsSA’s own counterfactual is the cruellest line in the release: without HIV, a South African woman in that year could have expected thirteen more years of life than she got. Then antiretrovirals arrived at scale and the line turned. Covid-19 took three years off it again in 2021 – deaths jumped by 29 per cent in twelve months – and then it recovered. Life expectancy is now 68,3 years, the highest Stats SA has ever recorded: 65,5 for men, 71,0 for women. Nearly eight million South Africans live with HIV, and this country runs the largest treatment programme in the world. That programme is a large part of why the graph still climbs.

So the growth is real, but the engine has been swapped out underneath it. Births peaked in 2008 at 1 198 764 and have fallen to 1 072 842. The birth rate, at 17 per thousand, is the lowest in the series. What keeps the total rising is that the people already here are not dying.

Fertility is not falling evenly, though. Gauteng’s fertility rate is 1,74. The Western Cape’s is 1,77. Both are well below replacement; demographically, they are living in the country’s future. Limpopo sits at 2,86 and the Eastern Cape at 2,62, and the Eastern Cape loses people to the other two in every window Stats SA measures: a net 295 374 in the current one, 171 525 of them to the Western Cape alone. So a province can be young and old at once: the Eastern Cape has the second-highest share of children in the country and the highest share of over-sixties, because the people in between have left.

The gap is easier to see on the world ladder. Gauteng and the Western Cape have rich-country fertility: both sit just above the United States and France, and below India. Limpopo, at the other end, is almost exactly Israel – 2,86 against 2,87 – and above Egypt and Botswana. The Eastern Cape and Northern Cape are a shade under Botswana. South Africa as a whole lands on 2,12, which is Indonesia to the second decimal. One country, and a spread from American to Israeli fertility inside its own borders.

The composition has changed just as completely. In the 1904 census more than one in five people counted was classified white; today it is one in fourteen. In the 2022 census the median white South African was forty-five years old and the median black South African twenty-seven – two populations at different points on the same curve, in the same country. The South Africa of the past looked very different from this one. So will the South Africa of 2060.

The decline int the fertility rate is not, however, a white story. South Africans classified as white are 7,1 per cent of the population and, being older on average, only 5,4 per cent of women of childbearing age. Take them out of the count altogether and – even on the extreme assumption that their fertility is 1,0, below Italy’s – the national rate rises from 2,12 only to 2,18. Black African women are 84 per cent of all women of childbearing age, which means the national number essentially is their number: whatever every other group does, black African fertility has to sit somewhere between about 1,95 and 2,33.

Should anything be done about it? The honest answer is that nobody knows how to raise a birth rate, even if we wanted to.

In a new Journal of Economic Perspectives article, Michael Geruso and Dean Spears count the countries whose cohort fertility has ever dropped below 1,9. There are twenty-four. The number that have since climbed back to 2,1 is zero. The annual number of births worldwide peaked in 2012, and their conclusion is bracing: an era in which a great many humans exist may prove to be a historical anomaly with an end.

Why so little traction? In the same issue, Paula Gobbi, Anne Hannusch and Pauline Rossi give the most sensible answer I have read. Income, schooling and mortality explain about sixty per cent of the differences in fertility between countries – but barely seven per cent of the change within countries over fifty years. What is missing is family institutions: marriage systems, inheritance customs, the norms that set a floor and a ceiling on how many children a household can plausibly have. In monogamous societies economics framework accounts for more than eighty per cent of the fertility decline; in polygamous ones, under a quarter. Change the institutions and the numbers move: when Namibia improved widows’ inheritance rights in 2008, the birth rate fell by almost a quarter.

There is a South African footnote in their story. When British fertility collapsed in the late nineteenth century, it collapsed in step among British migrants in Canada, the United States – and here. The habit travelled with the people.

The one intervention that does seem to shift births is not money but time. Steven Davis, Nick Bloom and co-authors, in a new NBER Working Paper, find that couples in which both partners work from home at least one day a week end up with 0,32 more children, and attribute some 291 000 American births a year to the practice – more than public spending on childcare buys. Before we get excited: South Africa has one of the highest work-from-home rates in their thirty-eight-country sample, and in countries like ours that mostly means farming and small-scale production at or near the homestead, not a laptop on the kitchen table.

The clerks of 1875 thought they were measuring a colony at the beginning of something. They were. The 720 984 people they counted became 63,5 million. Their successors are now measuring something else: the top of the curve. The population will keep growing for decades yet, on momentum and on medicine. But somewhere in this year’s estimates is, in all likelihood, the last South African generation larger than the one before it.

Figures: Union Statistics for Fifty Years (1960); the Cape of Good Hope censuses of 1865, 1875 and 1891; Official Year Book of the Union (1966); Stats SA, South African Statistics 2000, Census 2022, and Mid-year population estimates 2026. The long-run series were compiled and cross-checked for this post.

On page 53 of the 100-page atlas Google published this week sits a map that would have puzzled the cartographers of empire. It colours each country by the share of its AI conversations that are about work. The United States is pale. Western Europe is pale. Africa is the darkest continent on the map – and, for once, for the right reason.

The atlas is Google’s first serious attempt to chart what people actually do with its Gemini models: 15 million anonymised conversations from a two-week window in April, spanning 150 countries and 140 languages. Globally, 13.5 per cent of conversations are work-related. In the United States and the European Union the share falls below that. Across a broad swathe of Africa – from Mali to Mozambique – it exceeds 16 per cent. Google’s economists call it ‘a surprising inversion’ and offer an explanation: professionals in developing economies ‘could be leveraging it more intensely as a productivity multiplier to bypass traditional business constraints’. A survey by Ipsos and Google puts the same point more bluntly: 91 per cent of Nigerian AI users say they use it to assist with their work – the highest share of the emerging markets surveyed, ahead of the UAE, India and Brazil.

Below, I combine Google’s map with newly released usage data from Microsoft, OpenAI and Anthropic to work out where South Africa stands. Among the things you will learn: why one in four working-age South Africans now uses AI (a higher share than in Japan); which South African number has not moved once in OpenAI’s rankings in five quarters; which province uses Claude more intensively than Gauteng; and why nearly half of the Eastern Cape’s AI conversations are about coursework. Also: what AI assistance with household tasks is worth...

Read more

The OpenAI Foundation holds an enormous fortune. It owns just over a quarter of the company, which at today’s valuation is worth around $220 billion. Anthropic’s seven co-founders have promised to give away 80% of what they own, and the firm runs the most generous donor-matching scheme in the history of Silicon Valley. Nan Ransohoff, who runs public goods at Stripe and has done the sums more carefully than anyone, reckons all of this could release somewhere between $37 billion and $100 billion a year in new philanthropic giving. American charitable donations already run to about $600 billion a year. We are about to add a sixth of that, almost overnight, and nearly all of it earned by the same people, on the same technology, at the same moment.

This is a new gold rush.

We have had one before. In a 2023 paper in Past & Present, the historian Caitlin Harvey traces how the gold and mineral booms of 1848 to 1910 paid for a whole generation of universities – what she calls ‘the goldfield foundations’. Berkeley, Melbourne, Adelaide, Otago. And three of our own. The University of the Witwatersrand sits on the gold reef that made Johannesburg. The University of Cape Town grew up on mining money; the Rhodes Memorial still overlooks its rugby fields. Stellenbosch, where I teach, owes its independence to Jannie Marais, who made his fortune on the Kimberley diamond fields and left much of it to a university. The mines may have closed, but the universities remain.

So here is the question worth asking of this new gold rush: will it, too, build universities?

Almost certainly not. Frontier research now depends on computing power, data and salaries that sit inside a handful of firms rather than inside universities – more on that below. And these donors are in a hurry. They believe the coming decade will decide how the AI transition goes, so they are unlikely to spend it founding new institutions. If not universities, then what?

Two answers are already on the table. I think both are, at best, incomplete.

The first is Ransohoff’s own. In a widely-read essay she argues that we need a ‘Silicon Valley for public goods’: thousands of new ‘philanthropic startups’, a class of ‘philanthropic VCs’ to fund them, talent recruited out of tech and paid in equity, and a venture capitalist’s appetite for risk and speed. Her diagnosis is largely correct. The binding constraint on all this money is not the money; it is the capacity to absorb it well. GiveDirectly, one of the few organisations that can move cash at scale, has said it could spend perhaps $500 million a year, a tenth of $5 billion and a small fraction of a $50 billion potential funding stream. So the money will need organisations to receive it.

But building a new apparatus from scratch has a cost that the essay understates. It operates apart from the two institutions that development actually depends on: the state and the market. A parallel philanthropic economy, staffed by people recruited from San Francisco and answerable to the funders who pay them, is not obviously what a middle-income country most needs. So, what is the alternative?

A second option is the nostalgic one. With USAID gutted and official aid falling across the rich world, why not let AI philanthropy simply replace it, and become the foreign aid that governments no longer wish to give?

Because, as I argued on this blog a year ago, that breaks the two principles that should govern any transfer of this kind. Foreign aid must not disrupt the relationship between a government and its citizens: the moment a health system answers to a donor in California rather than to voters in Lusaka, the fiscal contract that makes states accountable weakens. And aid must work through the market, not replace it; money that distorts local prices or crowds out local firms leaves a country poorer the day it stops. Ken Opalo has made the state half of this case with more force than I can: three decades of substituting NGOs for the government has weakened state capacity across Africa, and the way to reverse it is to put elected, accountable, imperfect governments back in charge. Axel Dreher’s careful surveys point the same way. Aid does least for growth and most for whatever the donor privately wanted; and when it stops abruptly, as it just has, the cost can be measured in conflict and in lives. Putting AI money into that system would not repair it. It would only sustain something we already know can leave a recipient economy worse off.

Which brings me to my own profession. Oliver Hanney, who edits VoxDev, has written a forceful case that development economics is failing the AI moment – too reliant on the randomised controlled trial to say much about the larger questions AI raises – and, in this FT interview, he says he would like the AI labs to hire more development economists. On the first point he is right. On the second I disagree. We do not need more development economists, at least not the kind who spend five years measuring whether one app raised one test score in one district. If this money is to fund economists at all, it should fund the ones who study innovation, firms and science: the economics of how growth actually happens. Those are complements to the market and to the state. The randomista is too often a substitute for both.

So what would I do, if I had this money to give away?

I would endow chairs.

Endowed chairs are one of the clearest examples of a wider kind of gift: one that works alongside the state and the market rather than around them. And not at Harvard or Oxford; they are well provided for. And not yet at the universities too weak to use the money well. I would fund permanent chairs at the institutions ranked, say, between 100th and 600th in the world – the strong universities in the middle, with the talent but not the endowment – at the new intersections of fields: AI and economics, AI and biology, AI and ethics. I would pay the holders at the top of the global market, enough that the best Kenyan or Brazilian or Indonesian in a field could do frontier research at home rather than move to a firm. And I would give each of them a substantial budget for computing.

The striking thing is how little this would cost. By the QS rankings there are 504 universities between the 100th and the 600th in the world. Endow a $20 million chair at each and the whole thing comes to about $10 billion – still less than a twentieth of what the OpenAI Foundation alone is worth. And $20 million buys very different things in different places. Given what American industry now pays its AI researchers – of which more below – it might stretch to a single chair, with the computing power to match, at a university in the United States. In a country like South Africa the same sum endows not a chair but an institute: a senior scholar, a bench of postdocs and doctoral students, the research staff around them, and the compute they all depend on – in perpetuity, and at the frontier.

It is worth seeing where they would be (see map). They are on every continent: the middle-ranked universities of North America and Europe, but also the Indian Institutes of Technology, the great public universities of Latin America, the research universities of Southeast Asia and the Gulf. And among them sit the survivors of the last gold rush. Berkeley and Melbourne have long since climbed into the global elite and need nothing; but Otago, built on the Otago diggings, Cape Town, raised on mining money, the Witwatersrand, on the reef itself, and Stellenbosch, on Marais’s diamond fortune, all fall inside this band. The universities the last boom built are precisely the ones the next one should endow.

I am not being sentimental about universities. This is the one intervention that answers a problem we can now measure. In a recent NBER paper, Ufuk Akcigit and colleagues track 42,000 AI researchers through two decades of American tax and publication records. The top 1% of AI scientists in industry now earn about $1.5 million a year more than comparable academics – a gap that has widened fivefold since 2001 – while academic salaries have barely moved. The response is what any economist would predict: researchers are moving to large, established, well-resourced firms, and from open, published science to patents held privately. Universities are losing their best researchers because they can neither match industry salaries nor provide comparable computing power. A well-paid, well-resourced endowed chair is designed precisely to close that gap.

There is a precedent for exactly this kind of gift. Between 1883 and 1919 Andrew Carnegie funded more than 2,500 free public libraries around the world – 1,687 in the United States alone, and one at Stellenbosch University itself. For a long time these were dismissed as a rich industrialist’s monuments. Then economists examined them. In a paper just published in the Review of Economics and Statistics, Enrico Berkes and Peter Nencka compare American towns that built a Carnegie library with near-identical towns that qualified for one but, for local reasons, did not. Patenting in the library towns rose by 10 to 12 per cent over the following two decades, and the additional inventors were disproportionately women and the foreign-born – the ‘lost Einsteins’ that economies routinely fail to develop. The two groups of towns patent at the same rate for twenty years before the grant and diverge only once the library opens (see figure). Gregory Gilpin, Ezra Karger and Nencka find a similar effect today: invest in a library and the reading scores of nearby children rise. Free access to knowledge is, on this evidence, one of the highest-return public investments we have measured.

This is the clearest lesson economic history offers: prosperity comes from ideas, ideas come from access to them, and the societies that broaden that access grow faster. Endowed chairs are the equivalent, in our century, of the Carnegie library: the same investment in knowledge as a public good, applied where the scarce input is no longer the printed book but the trained researcher and the computing power they require. And, like the libraries, they satisfy both principles. They leave the government to govern and the market to operate, and they add to the stock of knowledge that both depend on.

There is a more radical version of this idea. Instead of endowing chairs at strong universities everywhere, aim the whole fund at the places that need it most, and make each gift larger: not a chair but a research institute, fifty million dollars apiece, and only at universities in the developing world that already sit in the global top thousand. Set aside China, the high-income countries, and the states whose sanctions put them beyond an American donor’s reach and the QS rankings leave 219 of them. An institute at each would cost about eleven billion dollars: close to what the chairs would, and still only a twentieth of the OpenAI Foundation’s stake. The map, though, looks quite different. The weight shifts south and east – thirty-seven institutes in India, two dozen across Malaysia, a long band through Southeast Asia and Latin America – and, for the first time in this story, Africa appears in numbers – nine universities in South Africa, five in Egypt, and one each in Ghana, Ethiopia and Tunisia – though I would make the case for a few more, perhaps adding Senegal, Nigeria, DRC, Rwanda, Kenya, Angola and Tanzania to the list.

The last gold rush understood something this one has not. Marais and Rhodes and the diggers of the Rand did not set out to abolish poverty within a decade. They built institutions that outlasted them, and, as it happened, outlasted the mines that paid for them. A century later, the reef beneath Johannesburg is nearly worked out; the University of the Witwatersrand is not.

This boom will also end, as booms do, and some of today’s valuations will look absurd in hindsight. The question is what will remain when it does. Fund an apparatus that operates apart from governments and markets, or use the money to restore an aid system we already know distorts both, and little will last. Fund the researchers who produce knowledge, and pay them enough to stay, and this gold rush may leave what the last one did: universities that outlast the boom that paid for them.

Now to convince a billionaire.

In April, on a stage in Washington, DC, Sim Tshabalala made a claim that would have sounded reckless five years ago: South Africa could lift its medium-term growth rate to between 3 and 4 per cent. The chief executive of Standard Bank, Africa’s largest bank, grounded the claim in something other than commodity cycles. ‘There’s a direct correlation’, he said, ‘between the rule of law and GDP growth.’

As you read this, I’m on my way to Johannesburg, to give a talk at Standard Bank. My topic is broad – ‘my team and I would very much like to hear your historically and comparatively informed perspective on what can and should be done to accelerate growth in SA, and on the likely effects of AI on our economy’ wrote Tshabalala. Below I share my ideas. I hope that by the end of it you will know how countries that start at South Africa’s current income level have doubled their living standards within twenty years; how fast AI capability is actually growing, measured properly, and where the two models released this month sit on that curve; why the poor got this technology faster than any before it, and why that weakens the standard argument for state intervention; and what the machines themselves think a bank is for in the economy they are building…

Read more

On Monday I typed my name at the bottom of a one-page statement titled ‘We Must Act Now’, joining Erik Brynjolfsson, Daron Acemoglu, Joseph Stiglitz, Ben Bernanke and dozens of others – among them more than a dozen Nobel laureates. The statement warns that AI may become radically more powerful over the next ten years, driving a transformation of the economy ‘larger than the Industrial Revolution, but unfolding over a vastly shorter time frame’. It calls on economists to help build the incentives, guardrails and institutions to steer the technology well.

Most of my co-signatories worry about jobs and wages. Those are real worries, and I share them. But signing from the southern tip of Africa, I found myself worrying about something stranger: what AI might do to morality itself.

Let me explain.

Where do our morals come from? The Sunday-school answer is: from above. The answer emerging from economics, psychology and evolutionary biology is: from below – from the daily business of making a living together. In two recent papers, the Harvard economist Benjamin Enke reviews a decade of research on what he calls moral boundaries. The foundational idea is that morality is functional. In his words, it ‘evolved as a form of “psychological and biological police” to enforce prosocial behavior and cooperation, including in economic production and exchange’. Groups whose members kept promises, punished cheats and helped their partners outproduced groups whose members did not; our consciences are the enforcement technology that survived.

If morality is a tool for cooperation, it should differ across societies according to who people need to cooperate with. And it does. Everyone loves their family more than a faraway stranger; the moral question is how quickly that love falls away with distance. Economists call people whose care declines slowly universalists, and those whose care is concentrated on kin and clan particularists. Neither is more moral; they simply spend the same moral budget differently. As Enke quips, universalists are great strangers to encounter, while particularists are great friends to have (choose your dinner guests accordingly).

What moves the slope? Need. In societies built on tight clans – where your harvest, your safety and your marriage all depend on kin – loyalty pays handsomely and trusting outsiders pays nothing, and the moral system matches: fierce in-group devotion, revenge, shame. Where people make their living trading with strangers, the moral system tilts towards impartial fairness, internalised guilt and trust in outsiders.

The evidence for this is remarkably wide. Enke coded the folklore of nearly 1,000 preindustrial societies and found that those with markets and money told morally universalist stories. In Ethiopia, Devesh Rustagi compared villages by their distance to historical marketplaces: villagers who live closer – and so trade more often with people they do not know – cooperate more readily with strangers in experiments. In Greenland, traditional hunters cheat anonymous strangers in the same games; their market-employed neighbours do not. And survey data from sixty countries show that cohorts who lived longer under democracy – that great machine for impersonal cooperation – are more universalist than their compatriots who did not.

The pattern is consistent: when making a living requires trusting strangers, we learn to value them. When it does not, we do not.

For most of recorded history the direction of travel was one way. Markets expanded, cities grew, institutions scaled, and the circle of people we needed – and therefore the circle of people we valued – widened from the clan to the village to the nation and beyond. The philosopher Peter Singer called it the expanding circle. An economist would call it a repricing: strangers became useful, and morality adjusted.

Now for the question that one-page statement forces on us. What follows is not a forecast. It is what happens inside a model economists already believe – Guido Tabellini’s 2008 model of how parents decide whether to raise universalist children – when you ask it one new question: what if the strangers stop being human?

In Tabellini’s model, parents pass on values by asking, in effect, two things. Who will my child mostly deal with? And will dealing with distant strangers pay off? If the answers are ‘strangers’ and ‘yes’, they raise universalists. If the answers are ‘kin’ and ‘no’, they teach loyalty and caution. Simulate the model from 1800 and it reproduces the moral history sketched above: as markets and democracies raise the returns to impersonal cooperation, universalism climbs, generation by generation.

Then let AI arrive, and the path forks on a single parameter – whether the technology raises or lowers the returns to impersonal cooperation between humans.

In one branch, AI is the new trade route. It translates languages, verifies reputations, shrinks the cost of dealing with someone you have never met, and pulls more of humanity into productive exchange with itself. The returns to trusting strangers rise, and the circle keeps widening. In the other branch, machines take over the exchange itself. Firms produce with fewer people; algorithms mediate what workers once negotiated; the stranger across the market square becomes economically irrelevant. Nothing dramatic happens at first. But in the model, parents notice that trusting outsiders no longer pays, and invest their children’s moral budget closer to home. The circle sags, one generation at a time.

Of course, a model is not the world. Morality has flywheels the simulation ignores: religions, constitutions, schools, the sheer human pleasure of company. Values move over generations, not quarters, which leaves time to steer. That is precisely why the statement says ‘now’.

But the mechanism is anything but hypothetical, and we need not wait for 2100 to observe it. Cue South Africa. A third of South Africans who want to work cannot find it; among the young, the share is closer to half. That is millions of people whom the production side of the economy, in the coldest sense of the phrase, does not need. The functional theory of morality predicts what should follow – trust retreating, circles tightening, the stranger recast as a threat – and two decades of stagnation, eroding social trust and periodic xenophobic violence read uncomfortably like the prediction. Benjamin Friedman made the general point in The Moral Consequences of Economic Growth: broadly shared growth buys tolerance and generosity; stagnation calls in the loan. On this one dimension, South Africa may already be a preview of the AI future the statement warns about.

I remain optimistic, as always. Perhaps the machines will release us from the dreary office and its spreadsheets into the social spaces where we cooperate more, not less. We’ll do Hyrox. We’ll go to yoga. We’ll play padel, sing in choirs, run parkruns, braai with the neighbours we currently only greet over the wall. We’ll have more time to watch World Cup fixtures, even if they expand the tournament to 64 teams!

We have flattened that curve before. Marketplaces did it. Constitutions did it. A dominee or two probably helped. The wide circle we now take for granted was built, transaction by transaction, because we needed each other. The machines will decide a great deal in the coming decades. Whether we keep needing each other is still ours to choose.

On the evening of 14 July 1926, in a room at the London School of Economics, a small group of scholars sat down to inaugurate the Economic History Society. Everything, in the words of the Society’s later chronicler T. C. Barker, was ‘cut and dried’. Tawney took the chair. The constitution was approved. William Ashley was elected president; Eileen Power and F. W. Tickner joint secretaries; a council drawn from Manchester, Aberystwyth, Edinburgh and Harvard. We are not told how many people attended. The prospectus had been printed in ten thousand copies, and Lipson had agonised over whether the page should carry 43 lines or 44.

That was a hundred years ago this past Tuesday.

What the founders could not have known, as they settled their orderly constitution, was that they had built their Society on the lip of the most volatile two decades in living memory. Within three years, the Great Depression. Within seven, Hitler. Within thirteen, a world war that would scatter and kill many of the scholars whose field they were founding. Eileen Power herself was dead by 1940. Cut and dried it was not.

I have been thinking about that founding this week because I have just written a comment on a new guide to the future of the field – and the future looks about as settled now as it did in 1926…

Read more

What, really, is the job of a columnist?

This week marks four years since I moved to Substack. I arrived in July 2022 with about 400 subscribers, carried over from my decade-old WordPress blog. Today there are more than six and a half thousand of you. Over the years I have experimented with length, frequency, subject and tone, and have now settled into a rhythm: a free column on Mondays and one behind the paywall on Thursdays, though I still think of it as my Friday column.

Sometimes people ask me why I write. I do not have a very good answer. I like writing. I like sharing ideas. I like the discipline of having to make sense of something in public twice a week. Other times I am asked why it has been ‘successful’. But I am not sure what success looks like. Likes? Some of my best posts attract only a handful, and social media traction is too idiosyncratic to read much into. Subscribers? Better, perhaps, but even that does not quite answer the question. I do not depend on these columns for my daily bread, not yet at least, which makes it easier to write what I like. But the question stands: what is the column actually for?

There is, of course, always a recent economics working paper with an answer…

Consider the graph above. In the first quarter of 2014, the Sunday Times sold 408,458 audited copies per issue. A decade later: 60,759. City Press lost 90 per cent of its circulation; Business Day, 85 per cent. Die Burger is now South Africa’s highest-selling daily, at 24,161 copies, about a fifth of what PE Express, a free community paper in Gqeberha, hands out per issue. The Weekend Post closed in 2024, after 77 years.

The decline of journalism is a long story, bound up with the internet, though it does not begin there. It might surprise you to learn that the modern columnist was born out of an earlier attempt to save the industry.

The 1960s were a tough time for newspapers. In America, new media like television and the tabloid press threatened advertising revenue, accelerating consolidation across the industry. New York’s newspapers were especially badly hit. In 1962, some 17,000 newspaper workers went on a 114-day strike, bringing the city’s six daily newspapers to a standstill. While the dispute centred largely on wages and contracts, the strike is remembered as a battle against automation. From the 1930s, teletypesetting, machines that set type automatically from a coded tape, and later film- and computer-based systems had been making their way into newspaper composing rooms, where they threatened to displace skilled labour. The strike was ultimately resolved, but by the end of the decade four of New York’s dailies had shuttered.

Having survived the strike, the New York Times attempted something new: it introduced the op-ed page, short for ‘opposite the editorial’, a page of outside commentary facing the paper’s own editorials. Of the new section, the editorial team wrote: ‘We hope that a contribution may be made toward stimulating new thought and provoking new discussion on public problems.’ Fine sentiments. But, as it turned out, it was an economically shrewd move too.

The Times’ version of the op-ed page combined elements of existing opinion pages, namely regular writers alongside guest authors contributing a mix of commentary, opinion and analysis. But, unlike the existing opinion pages, the Times reserved something like a quarter of this precious real estate for premium advertising.

According to Michael J. Socolow, a media historian, the op-ed model proved highly profitable, even in the recessionary climate of the early 1970s. Part of this lay in how cheaply the section could be run: attract high-profile writers, pay them no more than $150 per piece, and use existing employees to run the department. Surveys found it was among the most read pages in the paper.

So, if the column saved journalism once, could it do it again?

Today, with the advent of the internet, and its gobbling up of ad revenue, the only viable option for journalism seems to be the subscription model. News24, for instance, crossed 100,000 digital subscribers a year or so ago to become Africa’s largest subscription news publication.

The newsroom of the future, in short, will be paid for by its readers.

Which raises an awkward question: what, exactly, are modern readers paying for?

Enter the paper: ‘What do news readers want?’ by Gregory Martin, Cameron Pfiffer and Shoshana Vasserman. They obtained something close to the perfect dataset: the complete digital history of a large American metropolitan newspaper from 2020 to 2023. That is 605 million article visits and more than 55 million encounters with its paywall.

The paywall is the clever part. Like most news sites, the paper allows a few free articles in a rolling window before demanding payment. But the meter is invisible. Readers cannot see how many free reads they have left; the limits changed repeatedly; the system glitched at random. The paywall arrives unannounced, and its quasi-random arrival lets the authors measure, article by article, what readers will pay to keep reading.

What readers click on and what they pay for, it turns out, are two very different things. The site’s biggest traffic drivers were syndicated advice columns, the agony-aunt kind that runs in a hundred papers at once. Subscriptions were triggered by hard news: public health, the economy, local politics. Willingness to pay for an average local-news article was roughly double that for an average entertainment piece. And the most-clicked articles attracted the window-shoppers least likely ever to subscribe.

My favourite detail is what happens after readers subscribe. Once articles are effectively free, they drift towards entertainment and the advice columns; while rationing scarce free reads, they spend them on hard news.

This is the kind of finding readers should be able to carry to a braai, a coffee urn or a dinner table and say: you will not believe what I read the other day. People click on the light stuff, but they pay for the serious stuff. They arrive for one kind of content and renew because of another.

The implications for strategy follow. A newsroom chasing traffic should hire entertainment writers and cut local news; a newsroom chasing subscriptions should do the reverse. Since digital advertising has collapsed anyway, and AI search summaries are expected to shrink what click traffic remains, the subscription logic wins by default. The content that keeps a subscription newsroom alive is the serious, civic, expensive kind: the very journalism the click era taught editors to doubt.

So where does that leave the columnist?

Consider one more finding. The revenue-maximising introductory price for a subscription is zero. The newspaper charges $0.99, essentially to capture your card details. And no pay-per-article price could recover more than about 61 per cent of the revenue of the standard bundle. The money is in the renewal: the monthly decision of an existing subscriber to keep paying.

That changes the columnist’s job description. Hard news breaks irregularly and brings subscribers in. The columnist writes to a rhythm. The column’s job is not to be read; it is to be renewed.

Comments measure attention; cancellations measure value.

But retention is only half the brief. The other half is which columnists earn it.

Noah Smith, an economist who left Bloomberg Opinion for Substack, where he now has more than 450,000 subscribers, asks the right question: ‘Why would people pay $10 a month for a single writer, when they could pay $25 a month for the entire New York Times?’ I charge $75 a year, about R110 a month, a shade less than a full News24 subscription at R119. So, the same question Smith puts to the Times, now closer to home: why pay me rather than all of News24?

Here is Smith’s answer: legacy opinion pages misunderstand their own product. They commission opinions, short, punchy polemics, when readers want analysis: forecasts, assessments, explanations grounded in evidence. Newsrooms treat their own opinion writers, he notes, as ‘tawdry entertainers rather than as respected analysts’. His Bloomberg editors pushed him to be more subjective, more politicised. Paul Krugman has described his New York Times column as 800 words through three levels of editing, ‘intolerable’, and he was barred from using charts. Krugman now writes on Substack, too, sending out a weekly piece to more than 580,000 readers. Both Smith and Krugman likely earn more than $1 million per year from paid subscribers; not bad for a mere columnist.

But it should not be that surprising, really. Opinions are abundant. Everyone has one, social media distributes them free, and generative AI now manufactures confident polemic at zero marginal cost: the information pollution I have written about before. No reader needs to pay for outrage. Analysis remains scarce.

Jasmine Sun, a writer covering AI and Silicon Valley, takes the argument a step further. As machine-generated content floods every channel, she argues, the value of summary collapses and the value of secrets rises: ‘reporting is the act of taking private knowledge and making it public’. And trust follows the messenger: ‘it’s not about the sentence. It’s about who says it and their track record.’

Which brings us back to economists. We hold the kind of secrets Sun describes. The research frontier circulates in working papers years before it reaches journals, let alone the public; a columnist who can translate that private knowledge into public knowledge is doing reporting, from an archive most journalists cannot access. And our training is precisely the analysis Smith says readers pay for: mechanisms, trade-offs, testable predictions.

Of course, columnists alone will not save the newsroom. Martin et al.’s most sobering result is that every newsroom loses money at the margin, that is, for each extra article it runs; even the best-performing desks generate less subscription revenue than a journalist’s salary, and expanding investigative reporting would need a subsidy of roughly $11,000 per additional article per year. James Hamilton argued two decades ago that newsrooms produced more accountability journalism than readers wanted. The new evidence says readers value it far more than their clicks reveal, yet still less than it costs.

The lesson? Columns can help a newsroom survive. They have done it before. They are cheap relative to investigative reporting, and they build the habit of renewal that holds the bundle together. But that only works if editors solicit the right kind of writer. The column’s job is to earn the next month’s subscription, and what earns it is scarce analysis from a voice the reader trusts. That is the brief the economist is built for: privileged access to the research frontier, training in mechanisms and trade-offs, and a track record readers can check. Hire that writer, and judge the work by whether readers return.

A century ago, Walter Lippmann complained that the newspaper reader ‘expects the fountains of truth to bubble, but he enters into no contract, legal or moral, involving any risk, cost, or trouble to himself’. The subscription era has finally produced that contract: renewable monthly, and easy to cancel with a click.

The columnist’s job is to give the reader a reason not to cancel.

On 5 May 1967, Prof AC Cilliers, a retired professor of theoretical physics and author of a book on the limits of growth, wrote an article in Die Burger lamenting the development of Stellenbosch:

As a unique entity, Stellenbosch today must be considered fully developed – or nearly so. Any significant further growth will undoubtedly distort its already threatened image as a rural and tranquil university town even more; and will ultimately cause it to lose its unique soul or ethos. Stellenbosch was never originally laid out or intended to become a large city. Yet the way we are proceeding now, it is rapidly on its way to becoming just that.

For example, one of the major business enterprises recently purchased the smallholding east of the new extension at Unie-Park to transform it into a new residential neighbourhood – which will make traffic to and from town via Van Riebeeck Street and Merriman Avenue, streets that largely encircle the university precinct, even worse than it already is. Another large business has bought a farm adjacent to the Dalsig extension, on the Strand side of town, and plans to build several hundred houses and a few apartment blocks. He wants to invest his money wisely, and has chosen Stellenbosch as the most attractive location for doing so.

Since the end of the Second World War, Stellenbosch’s population has increased rapidly to around 25,000, of which about half are white and the other half non-white. Where will it end? At 50,000 or 100,000 or even more, once all the hills and valleys surrounding Stellenbosch are built up?

In April last year, I published my own lament. In contrast to Prof Cilliers, I argued in favour of further development, prompted by the court-ordered halt of a luxury estate on the slopes of Botmaskop. I argued that the development offered not only ecological renewal and improved safety on previously neglected slopes but significant economic and social benefits: jobs for the unemployed, rates for the municipality and spending at local restaurants and shops. I suggested that the case for Botmaskop Estate remains ‘overwhelmingly strong’.

Read more

Twenty-four centuries ago, a playwright put a professor in a basket.

In Aristophanes’ comedy The Clouds, staged in Athens in 423 BC, Socrates is hoisted up in a hanging basket while his students measure how far a flea can jump – by dipping its feet in wax to make tiny slippers. Into this ‘think-shop’ comes Strepsiades, a farmer deep in debt, who wants one thing from the university: to teach his son how to argue his way out of paying. He’s looking for useful, applied knowledge. A clear return on investment.

He does not get it, and promptly burns the place down.

What Strepsiades wants is plain: training that pays, a skill his son can take to market. Whether that is what a university is for, or whether it exists for something less immediately useful, is a question universities have never settled. What has changed is that it is now answered, every year, by a number.

That number is a ranking. Agencies such as the Center for World University Rankings (CWUR) publish a global league table in which Harvard comes first, and always does, because the scale is built so that Harvard scores 100 and every other university is measured as a fraction of it. CWUR then ranks the rest on four things: research output and citations, the prizes their staff hold, and the share of graduates who reach senior positions in large companies.

These rankings exist for a sensible reason. A student in Soweto cannot see inside a lecture hall in Stanford; a donor cannot audit a department. A ranking is a signal that makes hidden quality comparable, much as a university degree itself once did. The difficulty is narrower: what a ranking can measure is a small part of what a university is, and that part is losing its value.

A recent paper by the economists David Cutler and Edward Glaeser begins with what a university actually is, and finds the most durable institution we have built: of America’s best-funded universities, 98.6 per cent are more than fifty years old, against 57.2 per cent of the Fortune 500 firms. What accounts for that endurance is structural. A university is a self-governing community of scholars, protected by autonomy, which attracts people with an unusual appetite for discovery. That structure allowed universities to reinvent themselves repeatedly – dropping subjects, adding laboratories – as the world changed, while keeping their character.

A structure of that kind is precisely what a ranking cannot observe, so the table measures the observable instead: citation counts, prize-holders, and the later salaries of graduates. Goodhart’s Law warns of what happens next: once a measure becomes a target, it stops being a good measure. Point a university at its citation count and it will produce citations, whether or not they add anything.

The figure below sets the world rank of South Africa’s leading universities next to their underlying score, edition by edition, with Harvard for scale.

The two panels show the same institutions measured two ways, and the results diverge sharply. On the right, the scores are almost flat: Stellenbosch has stayed within a single point of 74.8 for the better part of a decade, Cape Town and Wits sit just above it, and Harvard, by design, is fixed at 100. On the left, the same universities are ranked, and the lines move a great deal.

The flatness is built into the method. CWUR counts research over a rolling ten-year window, credits alumni careers that stretch back well over a decade, and sums prizes accumulated across whole lifetimes. Scores of this kind cannot lurch from one year to the next; they accumulate slowly. A university’s position, and the direction of any drift in it, says far more about the past twenty years of its life than about the past one or two.

The single large exception on the chart proves the point. In the 2026 edition Wits climbs from 292nd in the world to 200th, yet its score rises by only 1.8 points, and all of that comes from one pillar. Research and employability were flat and the education score fell; the gain is entirely in the faculty pillar, which counts staff holding the major prizes in their fields, and on which Wits moved from unranked to 87th in the world. Cape Town and Stellenbosch register nothing on this pillar at all.

What produced the change is impossible to tell from outside. It may be a genuine appointment, a prize-winner arriving at Wits. It may just as easily be administrative: a distinguished scholar already on the staff whose affiliation was, for the first time, reported to CWUR. In that case nothing about the university changed except the paperwork. Either way, a single entry lifted a university of tens of thousands 92 places up a global table.

Aristophanes chose his title with care. Clouds change shape continuously while carrying little substance, which was his image for a certain kind of intellectual display. The yearly movement of the rankings has the same quality: a great deal of change on the surface, very little beneath it.

The flat scores, meanwhile, conceal a change that is real but too slow to register in an annual table. It concerns what a university is for.

It is tempting to stop here, fault the method, and defend the university. But on the question of purpose, the ranking may be reporting something true. A full quarter of the CWUR score rests on employability – on graduates who reach senior corporate positions – and that is, in practice, what much of modern higher education has reorganised itself to supply. Not the frontisterion, the place for thought, but the prakterion, the place for training. Universities sell degrees as routes to a salary, advertise their employment rates, and report graduate earnings. The ranking rewards career outcomes, universities publish them, and the older question – what is worth knowing – recedes. The historian Chad Wellmon, in After the University (2026), tells this as a centuries-long story: how higher learning came to justify itself in the language of productivity and economic utility, leaving unmentioned the practices of disciplined study that once defined it.

Strepsiades, then, largely won. The university that supplies a credential and asks few questions about wisdom is close to the one we now run, and the ranking reflects that choice back to us with some accuracy.

The promise underwriting that bargain was straightforward: complete the degree, and step onto a stable professional path. Artificial intelligence weakens it from both sides. It performs the measurable academic tasks – the essay, the literature review, the standard analysis – at very low cost, and it is removing many of the entry-level jobs the credential was meant to secure. This may explain an otherwise odd scene: students booing speakers who praised AI at their own graduation ceremonies in 2026. Read as simple technophobia it is hard to follow; read as the suspicion that the path they were trained for is closing, and that their training fitted a smaller idea of the university than they had been promised, it makes sense.

It is not only new graduates who sense this. The public has begun to withdraw its confidence in the degree itself. In Britain, the share of people who think university is not worth the time and money has more than doubled, from 14 per cent in 2005 to 34 per cent in 2025, while the proportion expecting graduates to end up much better off financially has fallen from a half to barely a third. In the United States the slide is steeper: three in four adults called a college education ‘very important’ in 2010; by 2025 only about a third did. No comparable series exists for South Africa, and the mood here is probably less sour – but the direction of travel is unlikely to be different. A bargain struck on financial return cannot hold once the return itself is in doubt.

Cutler and Glaeser are honest about the limits of their own method. They can value medical research because health has a unit: years of life. When they turn to the humanities they pause, asking what it is worth to understand Shakespeare’s sonnets, or the lives of early humans, and concede that they cannot say. The value does not reduce to a number. What long looked like a weakness of the humanities is now closer to a strength: the things a spreadsheet cannot price are, increasingly, the things a machine cannot reproduce.

Two South African scholars saw this coming. The classicist Johan Thom, in his 1997 inaugural lecture at Stellenbosch, took the same comedy as his starting point and then reached back further, to the quarrel between Socrates and the Sophists. The Sophists were the first professionals of higher education: they charged fees and taught the practical art of winning arguments and public office, the original employability. Socrates took no fees and taught the pursuit of the good, which had no obvious use at all. The tension between training for success and educating for wisdom, Thom showed, is as old as the university itself.

Thom’s word for what is now at risk was schole, the Greek root of ‘scholarship’, meaning leisure, or freedom from necessity: the time and space a person needs in order to think. The classics, he argued, train a particular and increasingly rare skill – reconstructing a whole from a fragment, uit die klou die leeu, the lion from its claw, the animal from a few bones. In an age saturated with machine-generated text, the ability to judge a fragment, to tell the genuine article from a plausible imitation, may be the most valuable thing a university can teach.

The Stellenbosch philosopher Anton van Niekerk, in a chapter first published in 2009 and now free on johanfourie.com, makes a complementary point. A university, he writes, is its students and their ‘will to knowledge’; the buildings and the library are only the setting. His image is the voorloper, the person who walks ahead of an ox-wagon to test whether the road ahead will hold. A university’s task is to go ahead of society and report what it finds. This, to me, is the strongest reason for a university to take up AI and use it openly: mapping what these machines can and cannot do, and reporting back to a society that will soon depend on them.

None of this appears in a ranking, and none of it can. A league table does three things at once: it magnifies differences that are largely noise, it records with fair accuracy what universities have become, and it leaves out the part that may matter most – the capacity, protected for nine centuries, to think freely.

So the rankings are worth reading, provided one remembers what they exclude. You cannot judge an oak by counting its acorns, and you cannot take the measure of a place built for thought by recording how far its fleas can jump.

On the morning of 17 June 1796, Captain Benjamin Stout dragged himself, sputtering and exhausted, out of the sea. Hercules, his ship, had run aground just off the coast of the Eastern Cape. The ship and its captain were originally from America and had been en route, delivering rice from India to England when the wreck happened. On the shore, he met and was cared for by the local African inhabitants, who helped guide Stout and sixty of his men to Cape Town. On his way home through England, Stout published a record of his experiences, The Narrative of the Loss of the Ship Hercules.

In an age of exploration, shipwreck tales were popular and Stout’s book was well received. But where most other shipwreck authors told their tales to spread the word of God’s mercy or to claim fame and fortune, Stout did something rather strange. He dedicated his tale to American president John Adams. And not only that, he implored Adams to colonise the area that would become the Eastern Cape of South Africa, writing, “Such are the capabilities of this delightful country; and, as I apprehend, nothing is wanting to perfect the whole but the exertions of a wise and liberal government, I have recommended the establishment of a colony from America in these parts to your consideration.”

To understand how Stout could imagine such a proposal, we need to understand the world in which he operated…

Read more

In the autumn of 1965, something was missing from the tomato fields of California. For two decades, close to half a million Mexican men had crossed the border each year to pick the crops that Americans ate. They were called braceros – hired hands, brought in under a government programme that had started during the war. Then, on the last day of 1964, Washington closed the programme down. The braceros went home.

Politicians had promised this would be good news for American workers. Take away the foreigners, the argument ran, and farmers would have no choice but to offer better wages to the locals who stayed behind. It was sold, in the language of the day, as an ‘active labour market policy’: remove the migrants, lift up the natives. The logic felt like common sense. Fewer workers chasing the same jobs should mean higher pay.

It did not work out that way.

Three economists – Michael Clemens, Ethan Lewis and Hannah Postel – went back to the archives and measured what actually happened after the braceros left. If kicking out foreign workers really did raise wages for the people who remained, we should see it most clearly here, in the states that had leaned hardest on Mexican labour. Their own model predicted that farm wages in those states ought to jump by around 12 per cent.

They jumped by almost nothing. Wages barely moved. Employment of domestic farm workers barely moved. The promised gain to American workers never came.

Where did the adjustment go instead? Into machines and into crops. In tomatoes, growers bought the Blackwelder mechanical harvester and let it do the stooping. In cotton and sugar beet, the same story – more machinery, fewer hands. Where a crop could not be mechanised, farmers often switched to one that could, or grew less of it. Wages stood still. The work itself was engineered away.

This is the first lesson, and it holds well beyond California: an economy that loses workers adapts. The gains its planners promised to the people left behind tend to flow instead to whatever does the job more cheaply – often a machine.

Back to the present, and a little closer to home.

South Africa has seen another wave of anger at foreigners – marches, threats, men turned away from clinic doors and spaza shops, the now-familiar demand that ‘illegal’ foreigners be rounded up and removed. The cause, we are told, is obvious. Foreigners are taking South African jobs. Foreigners are driving crime. Get rid of them and ordinary South Africans will breathe easier.

These are confident claims, but the evidence behind them is thinner than the confidence suggests. We have little solid proof that foreign workers are the reason South Africans cannot find jobs, or that removing them would open those jobs to anyone else. What economics can speak to with more assurance is the other side of the ledger: the measurable consequences of acting on the anger.

Begin with our own history, because South Africa has run a version of the bracero experiment before. For much of the twentieth century, apartheid policy shaped the supply and price of black labour – through influx control, job reservation, and, from the 1970s, rising wage pressure as black workers organised. As that labour grew scarcer and dearer in the formal economy, employers in mining, farming and manufacturing did what the Californian growers did: they mechanised. Machines replaced unskilled workers, and the economy settled onto a path that created steadily fewer jobs for the least skilled. The structural unemployment we live with today is partly its inheritance.

Migrants are, of course, not only producers. A migrant is a worker who competes for jobs – but also a customer who helps create them, someone who buys bread, pays rent, takes a taxi, sends children to a crèche. A 2026 study from Norway, by Sigurd Galaasen and colleagues, measured this second channel directly, tracking the electronic payments of an entire country. Workers more exposed to migrants’ spending saw their own incomes rise, strongly and persistently. By spending what they earn, migrants help pull native wages up.

This is what economists mean when they say migrants and locals are usually complements, not substitutes: each makes the other more productive. Some of that is the spending we have just seen. Much of it is what migrants bring to the work itself – education, a trade, business experience, and the ideas, capital and networks that come with them. A Congolese electrician or an Ethiopian shopkeeper builds something that was not there before, and hires people to run it. A second 2026 study, by Angel Aguiar and Stephen Devadoss, put a number on what cutting that link costs: deporting just 30 per cent of America’s undocumented workers, they estimate, would shrink United States output by more than 128 billion dollars, with the deepest losses in agriculture, food processing and construction.

It is true that perhaps this mutual benefit is weaker in a country like ours, with millions of unskilled workers already searching for jobs, than in a rich, labour-scarce economy like Norway; the effect may well be smaller here. And what is probably also true is that the gains from immigration and its costs fall on different South Africans. Those of us who never meet a migrant at the hiring queue – who employ them, at most, as domestic workers, or merely read essays like this one – enjoy the upside while seldom feeling the competition. The people who feel it are not us, and their anger is not simply prejudice.

Yet the case runs deeper than any wage calculation. Many of these migrants have lived here for decades – their children in South African schools, their spouses South African, often speaking the languages we speak. They have built lives and businesses inside an economy where they are active participants, and a deportation drive would tear at all of it.

It would tear, too, at a future we cannot yet see. Every threat and every removal tells the migrants who have not arrived – the ones who might have opened a business and taken on a few employees – that they are not welcome. The firms that fear keeps away are firms we never get to count.

And come they would, despite everything, because South Africa, for all its troubles, is still the richest economy for thousands of kilometres – the place where a life can be built. That is precisely why this is harder for us than for Canada or Australia, who draw their migrants from across oceans, with no long, unpoliced border onto a poorer world. We cannot wish our geography away.

So what should we do? If the problem is really that too many people are here illegally, the economist’s answer is almost embarrassingly plain: make them legal. South Africa has so many undocumented migrants largely because Home Affairs spent years as one of the state’s worst-run departments, where a lawful visa could take longer to obtain than a child takes to be born. The department is improving, but too slowly. Fix the queue and much of the ‘illegality’ simply dissolves.

Readers may hear an argument for open borders here. But that is not what I’m arguing for. A country is entitled to know who crosses into it, and our long, porous frontier is a real problem that deserves a real answer: a properly policed border, so that future migrants come through the front door and fewer are ever forced into the shadows. Legal status serves that same end. A person on the books can be taxed, counted and, where someone proves dangerous, traced and removed – provided we also build the capable police service we still lack.

The country does not need fewer people. It needs more of them here legally – starting businesses, paying taxes and employing the South Africans who badly need the work. Shipping thousands back – worse, leaving it to crowds at the clinic door and the spaza counter to decide who belongs – would leave the South Africans who stay poorer, and it would be inhumane besides.

Sending foreign workers home did not raise a single American farm wage in 1965, when replacing a picker meant buying an expensive new machine. Today, when a task can be handed to a machine – or a line of code – faster and more cheaply than ever, the gains from any deportation are even likelier to flow to capital than to the workers left behind. There is little reason to expect it would do more for us.

In a recent essay in the Financial Times, the president of Argentina set out to justify a new law by way of economic history. Javier Milei – an economist before he was a politician – traced the modern economy to a single document: the charter of the Dutch East India Company, signed on 20 March 1602. Machines made the Industrial Revolution possible, he argued, but Dutch corporate law made it happen, because only when the law placed a ceiling on risk did capital deploy with genuine force. The machine and the legal entity were, in his phrase, ‘the double helix of modern prosperity’.

And because artificial intelligence will do for the human brain what the steam engine did for human muscle, his government has now submitted legislation to give AI the same legal architecture: a new corporate category, the ‘non-human corporation’, operated by AI agents, protected by limited liability, taxed lightly and free to choose its own governance law. Human shareholders ‘may participate, but are not required’.

It is easy to list what could go wrong. Nobody knows how companies operated by autonomous AI agents will behave – not Milei, not the engineers building the agents, and certainly not the courts that will have to untangle the first disasters. But that is precisely what makes the experiment brave. Argentina is positioning itself as the place where the fruits of this new technology, and its dangers, will materialise first. Whatever AI-run companies turn out to be, the world will find out in Buenos Aires.

From the southern tip of Africa, the contrast stings. In April, the South African government had to withdraw its draft national AI policy after journalists discovered that several of its academic references did not exist – invented by the very technology the policy was meant to govern. One government is using economic history to design its future. The other could not check its own footnotes.

But is Milei’s history right? On the main point, yes. But the fullest record I know of what limited liability actually does to an economy sits not in Amsterdam but in Cape Town: fifteen volumes of company registrations in the Cape Joint Stock Archive, recording, name by name and street address by street address, who owned the companies of the late nineteenth-century Cape Colony. That archive shows what Milei gets right. It also shows what his essay leaves out…

Read more

By the third century of the Common Era, Roman emperors had been stripping silver out of their coins for two hundred years. The denarius that once held three grams of silver held barely a smear by the time Diocletian came to the throne. And yet Romans kept paying with it. They kept accepting it. Long after the silver had largely vanished, ordinary people still trusted the stamp on the face of it. Why?

That has been a puzzle for more than a century. The archaeologist Kim Bowes, on a recent Conversations with Tyler episode, said she now finds it the single strangest feature of the late Roman economy. People knew the coins were debased. They still used them. Even when inflation reached the high triple digits, they kept paying in coin.

A new working paper by Johannes Boehm and Thomas Chaney – titled, plainly, Trade and the End of Antiquity – suggests an answer. The connective tissue of pre-modern commerce was not metal but belief.

Boehm and Chaney have done something remarkable. They have assembled a database of nearly half a million ancient coins, found in more than five thousand hoards buried across Europe, North Africa and the Middle East between AD 325 and AD 950. For each coin they record where it was minted, when it was minted, where it ended up, and roughly when its owner buried it. They then fit those half-million observations to a dynamic model of trade and money, and use the model to recover regional trade flows and consumption over more than six centuries.

They find that the Mediterranean’s relative decline began earlier than the standard story allows. Henri Pirenne, in a famous thesis, dated the rupture to the seventh century, when the Arab conquests sliced the old Roman world in two. Boehm and Chaney push the start date back at least a century, into the 500s, when the Western Roman state was already fraying. The Arab conquests of the seventh century then did something different. They redrew trade. North–south flows across the Mediterranean collapsed at the new frontier between Islam and Christianity. South–south flows, within the Caliphate, flourished. And by the late ninth century, the wealthiest stretch of the ancient Western world had become the Atlantic rim: al-Andalus, Frankish Gaul, the lands that would later become the engine of European commerce.

To put it into perspective: the centre of economic gravity in the Western world shifted northwest seven hundred years before Columbus.

But what does it mean to say a coin ‘flowed’? A bronze follis is inert. Someone trusted it enough to take it in payment, and someone else trusted that person enough to take it in turn. A coin, in this sense, is a token of trust. The economic historian Deirdre McCloskey reminds us that the English word credit comes from the Latin creditus, meaning ‘believed’. To extend credit is to extend belief. Aquinas listed faith as the sixth of seven virtues – above the four pagan virtues of prudence, temperance, justice and courage, and just below love and hope. Romans called it fides, and they treated it as the rule by which alliances were kept and oaths honoured. Carthage, McCloskey notes, was a commercial republic that distributed state revenues to its citizens. Rome ran on fides. Rome won.

When Bowes describes Romans accepting near-worthless silver coins, she is describing fides in action. The denarius worked because people believed the next person would take it. And the next person did. The Roman state was, as she puts it elsewhere in the interview, ‘a mask for an empire of friends and family’ – a network of personal trust, anchored at one end by the mint’s stamp.

This is what Boehm and Chaney’s data measure. Their estimates of mint output across regions and centuries – the figure above – map where that trust held and where it broke. The faith migrated. Mints in Carthage and Constantinople remained highly productive through the fifth century. Italian mints faded earlier than anyone had thought. The Sasanians sustained a steady silver output until the Arab armies overran them. After the conquests, a new mint network emerged across the Caliphate – Damascus, Cordoba, eventually Baghdad – while the Byzantines pulled production back behind their walls. By the late eighth century the Carolingian mints joined in, a sharp Frankish spike in the data.

Which brings us to a question Boehm and Chaney’s coin data cannot answer. What about a state that minted no coins at all?

Consider the kingdom of medieval Ethiopia.

In a new working paper, Mattia Bertazzini and Andrés Martignano use a regression discontinuity design at the fourteenth-century boundary of the Solomonic state to show that long exposure to centralised Ethiopian institutions raises literacy and urbanisation today, more than five hundred years on. The mechanism, they argue, is property rights – the gult system of feudal grants, under which the king parcelled out the right to collect tribute and the holder owed military service in return – combined with an early structural transformation out of subsistence farming.

What is striking about this state, for our purposes, is what it did not have. There was no professional civil service. There was no fixed royal capital; the Negus toured the realm with his court. And, the authors observe, ‘there was no minting of coins.’ Medieval Ethiopia administered a substantial territory across a difficult plateau for three centuries without striking a single piece of currency.

So what carried credit, in the literal sense, when there were no coins to carry it? The Coptic Church carried it. Royal chronicles in Ge’ez script carried it. The grants of gult land, recorded in monastic archives, carried it. The Negus’s annual circuit of judgement carried it. In Ethiopia, belief was registered in script and ritual. As the historian Isaac Samuel has documented, Ge’ez was one of a number of endogenous African scripts – Meroitic in ancient Kush, the Arabic-derived Ajami along Saharan trade routes – that emerged in dialogue with foreign writing systems but served local purposes: recording grants, asserting authority, keeping count.

Bertazzini and Martignano show that the institutions built on that belief are still producing measurable differences in the 1994 and 2007 censuses – eight percentage points more literate, nine percentage points more urban, on the inside of a line drawn in 1340.

The lesson? The End of Antiquity is, in the end, a story about belief migrating – to new mints in the Caliphate, to new courts and chronicles on the Ethiopian plateau, and eventually to the Atlantic rim that would, much later, come to dominate the world. Coins and chronicles did the same job. They asked one stranger to trust another. Where that trust held, economies functioned. Where it broke, empires followed.

McCloskey would say bourgeois prosperity is the eventual heir to all this. She is right. The long walk to modern wealth begins fifteen hundred years earlier, when a Roman shopkeeper looked at a silver-thin denarius and decided, against the evidence, to trust it.

Take a look at the main square of almost any old European town. Somewhere above it – on a church tower or a town hall – there is a clock. It has hung there for centuries, striking the hour for everyone within earshot: the merchant, the apprentice, the magistrate, the beggar. Except for the most exquisite ones, we barely notice it now. Yet that public, shared, freely available timepiece is one of the foundations of the modern economy.

Now travel back to eleventh-century China. Around 1088 the official Su Song built a wondrous astronomical clock tower at Kaifeng: twelve metres tall, water-driven, with an escapement-like mechanism that turned a celestial globe while small figures emerged to mark the hours. It was centuries ahead of anything in Europe. And it changed almost nothing. As the economic historian David Landes told the story, timekeeping in China was an imperial secret, bound up with astrology and the emperor’s heavenly mandate, shut away inside the palace. When invaders sacked Kaifeng in 1127 the clock was carried off, no one could rebuild it, and the knowledge died with its handful of court specialists.

The puzzle Landes left us is why the same machine could be a curiosity in one place and a turning point in another.

Read more

In 1955, 970,000 black children were enrolled in South African schools. By 1967 the figure had reached 2.24 million. By 1973 it was 3.31 million – a 131 per cent jump in twelve years. And it happened even though school was not compulsory for a single one of those children, and even though almost everyone, parents included, knew the schools were failing them.

The schools were chronically underfunded and overcrowded. Most teachers had no tertiary training. Most pupils never made it from primary school to high school, let alone to matric (the final school-leaving examination). And still the numbers enrolling grew every year. The historian Oliver Aiken calls this the paradox of Bantu Education: a system designed to give black children less, swamped by black families demanding more.

Why would a parent enrol a child in a school they knew was failing? Part of the answer is simple: any education was worth more than none. A literate child could find work a non-literate one could not. Families were making an investment – in the ordinary economic sense of the word – and they understood what they were doing.

We remember the uprising of 16 June 1976 as a revolt against Afrikaans, and as the moment Black Consciousness arrived in the streets. Both are true. But they explain why the students marched, not why they marched that year. For that, it helps to look at the economy.

The timing of Soweto appears to have been shaped, to a surprising degree, by the South African business cycle: by a boom that allowed white voters to ignore black schooling, a downturn that prompted business to help pay for it, and a recession that withdrew the funding at a particularly bad moment.

Start with the boom. Through the 1960s the economy grew quickly – six to seven per cent a year for much of the decade. Employers’ demand for skilled work was met comfortably by white workers, whose higher salaries they could afford to pay. What they wanted from black workers was narrower: literate, low-skilled, low-wage labour. Bantu Education was built to supply close to that, and little more. As Hermann Giliomee notes, white South Africans of the time largely did not support large-scale spending on black schooling. There was little economic reason, from their point of view, to train skills that employers did not yet intend to buy.

So the system was kept small by design. Spending on Bantu Education was capped at R13 million; anything beyond that had to come from taxes levied directly on Africans themselves. Black parents, unlike white parents, paid for school fees and textbooks out of their own pockets. By 1971–2, the state spent roughly eighteen rand educating a white child for every one rand it spent on a black child. In the nominally independent Homelands – the rural territories to which the apartheid state assigned black South Africans – schooling was funded worse still.

Into that gap stepped the parents. In 1967, Tswelelang Higher Primary in Meadowlands set out to build extra classrooms by levying a R2 tax on local residents and running a bottle-cap collection drive. The Department refused to match a cent of what they raised. When the Urban Bantu Council asked the liberal-leaning Johannesburg City Council to help lift the monthly schooling levy from 18 to 38 cents, the government dragged its feet, conceded two cents, and left the council R55,000 in debt by 1970.

By 1971 Soweto had twenty secondary schools – roughly one for every 80,000 urban African families. To put that in perspective: a township of more than a million people had the secondary-school capacity of a small country town. When residents pressed again for the higher levy, the deputy minister, Piet Koornhof, told them any extra money would be spent not in Soweto but in the Homelands, where black secondary schooling was supposed to happen anyway.

Then the cycle turned. The downturn of 1969 achieved something a decade of liberal argument had not: it changed the minds of business leaders. For years, liberals – here, those who wanted apartheid’s economic restrictions loosened – had warned that Bantu Education would eventually leave the economy short of skilled black workers. In a slump, with white labour looking expensive and scarce, that warning began to look practical rather than theoretical.

Companies that had ignored black schooling now moved to fund it. In August 1971 Reckitt and Colman pledged R100,000 for bursaries; Barlow Rand set up a foundation with some R700,000; in February 1972 Harry Oppenheimer’s Anglo American gave R160,000 towards a junior secondary school in Soweto; The Star launched its TEACH fund. Because these gifts flowed through the City Council and private hands, they bypassed the R13 million ceiling.

Government shifted too. A pragmatic, verligte (reform-minded) wing of the National Party had decided, as Jonathan Hyslop puts it, to accept ‘the continued existence of the urban black working class’ – and to spend accordingly. Between 1969 and 1972 it dropped job-reservation rules, eased restrictions on black urban employment, and, most importantly, in 1972 shifted the financing of black education off African taxation and onto the state’s general revenue. The 38-cent levy was finally granted. Spending on black education rose from R55 million in 1970–1 to R97 million by 1973–4.

The effect was substantial. In four years the number of secondary schools in Soweto doubled. Secondary enrolment climbed from 12,656 pupils in 1972 to 37,656 in 1976 – an increase of more than 170 per cent.

This sets up the puzzle. Why did Soweto erupt precisely when demand for schooling was at its peak, and when, for the first time, both business and the state were moving to meet it?

Several factors coincided – and most of them were economic.

First, the timing was unfortunate. The global recession of 1973–5 reached South Africa, and the government reversed course and cut spending. As one historian puts it, the building of Bantu Education’s infrastructure ‘slowed down just at the moment that the greatest demands were being made on the system.’ The chart above shows the squeeze: growth fell from 6.7 per cent in 1974 to barely positive in 1975 and 1976, and into contraction by 1977.

Second, the same demand for skilled black labour that had opened company chequebooks now drew the best teachers away from the classroom. A factory or clerical job paid better than teaching, and teaching – long one of the few respectable careers open to educated Africans – became less attractive just as the schools needed staff most.

Third, the Department chose this moment to shorten black schooling from thirteen years to twelve, combining two cohorts of pupils into one in 1976. The result was a ‘bulge’ of younger children pressed into schools that had specialised in senior grades and were not equipped to receive them. Philip Bonner’s assessment is blunt: ‘Learning was virtually impossible and the pass rate plunged to lower levels than ever before.’

Into that overcrowded, underfunded and overstretched system, the government introduced a deeply unpopular measure. The 50–50 rule – that half of all lessons be taught in Afrikaans – had for years been applied loosely; by 1964, 95 per cent of secondary classes were taught in English. But hardliners wanted it enforced. In 1976 Vorster appointed the conservative Andries Treurnicht as deputy minister, and on 11 June Treurnicht ordered five striking Soweto schools to obey the rule. Soweto had planned, as in earlier years, to resist quietly. This time the Department forced the issue. The sense of ownership that residents had built over their schools, Aiken writes, ‘collapsed in an instant.’

None of this reduces Soweto to economics. The students who marched were refusing a language that stood for their subjection, and they carried a new confidence – Black Consciousness – that told them they need not accept it. Bantu Education was, at its core, a racist project designed to keep black South Africans politically and economically confined. The business cycle does not explain that, and it is not meant to.

What economics helps explain is the timing. It suggests why the system was so close to breaking in June 1976, rather than five years earlier or five years later: a school system swollen by demand, briefly funded, then squeezed again in a recession – all within half a decade. The Afrikaans decree was the immediate trigger. The economic conditions help explain why the system it met was already under such strain.