More Light

In 1865 a 29-year-old economist named William Stanley Jevons published a book about coal, and buried in it was one of the most counter-intuitive observations anyone has ever made about progress. Everyone assumed that as steam engines got more efficient — as they squeezed more work out of every lump of coal — Britain would burn less coal. Jevons said: the opposite. Make coal-power cheaper to use, and people will find a thousand new uses for it, and the country will burn far more. He was right, spectacularly, and the idea has carried his name ever since: the Jevons paradox. Efficiency doesn't shrink our appetite for a thing. It feeds it.

It is almost always quoted as a warning — usually about energy, and I've used it that way myself, in an essay about what AI costs the planet. But lately I've been turning it over the other way round, because the defining fact of this decade is that we are making one particular thing radically cheaper — thinking itself — and Jevons has something enormous and hopeful to say about that. So let me make the optimist's case, properly, with the data. Not the giddy kind of optimism. The kind you can defend.

Start with light, because light is the clearest case there is

Before coal, before computers, the purest illustration of Jevons is something you're using right now without a thought: artificial light. For almost all of human history, light was breathtakingly expensive. A labourer in ancient Babylon might work the better part of a day to afford an hour of dim lamplight. So people simply lived in the dark — they rose with the sun and lost half of every day to blackness.

Then, century by century, light got cheaper — candle, kerosene, filament, fluorescent, LED — until the economist William Nordhaus, who spent years reconstructing its price, could show that light had become tens of thousands of times cheaper in terms of the work it costs you. And here is the Jevons twist, the whole point: we did not use the savings to sit in more darkness. We used a hundred thousand times more light. We lit streets and skylines and phone screens and fairy lights strung up for no reason but joy. We banished the dark that had owned half of every human life, and we barely noticed it happen.

Now hold that picture and change one word. For three thousand years the scarce, rationed, expensive thing was light. In our decade, the scarce thing that is suddenly getting cheap is thinking — analysis, writing, coding, translation, tutoring, design, the whole category of cognitive work. Jevons, and three thousand years of lamps, tell you exactly what happens next. We are not going to do less thinking. We are going to illuminate every problem that has been sitting in the dark because nobody could afford to shine a light on it.

It has happened every single time

If that sounds like a leap, it shouldn't, because we have run this experiment over and over, and it comes out the same way every time. Whenever we make a capability drastically cheaper, total use goes up, and — this is the part the fear always misses — new kinds of work appear that nobody could have named beforehand.

Look hard at the two middle ones, the bank tellers and the accountants, because they are the closest thing we have to a rehearsal for AI, and the economist James Bessen dug up the data. The ATM automated the one thing a bank teller did all day: count out cash. The obvious prediction was mass redundancy. What actually happened is that ATMs made each branch cheaper to run, so banks opened far more branches, so the number of tellers rose for decades — and the job shifted from counting notes to helping people, selling, solving. Same story with the spreadsheet: it vaporised the manual arithmetic that armies of clerks used to do, and the number of accountants and financial analysts went up, because once analysis is cheap, a business wants a great deal more of it.

The mistake underneath the fear even has a name economists have been using for a century: the lump of labour fallacy — the belief that there is a fixed lump of work in the world, so that a machine doing some of it must leave less for us. It has been wrong every single time. In 1900, roughly forty percent of Americans worked in agriculture; today it's under two percent, and we did not end up with thirty-eight percent of the country unemployed. They moved to jobs that didn't exist when their grandparents were born. Work is not a fixed pile you draw down. It's a fire you feed.

We have had this exact fear before

Here's the part I find genuinely calming: the fear isn't new either. Every time a technology has arrived to automate something people did by hand or head, the same three-beat story has played out — a sincere panic, a real jolt of disruption, and then adaptation into something larger. It's so consistent that, once you've watched it a few times, you can almost set your watch by it.

Read those in a row and the shape becomes almost comic. Plato worried — in writing — that writing would rot our memories. The Luddites smashed the looms. Someone, somewhere, genuinely lay awake over the future of the buggy-whip trade. And every single time, the fear was sincere, the disruption was real for some, and the world came out the far side with more — more books, more music, more work, more of the very thing everyone was sure was ending.

Let me be careful here, though, because the lazy version of this argument is smug, and the smug version is wrong. The fear is never entirely silly. The weavers really were hurt; that dip in the middle of the curve is a real person's real decade, and telling them "don't worry, the aggregate works out" would have been an insult. The honest claim isn't relax, it's always fine. It's that the dip is never the whole story — and that reading only the disruption, and stopping there, has been the single most reliable way to be wrong about technology for two and a half thousand years. AI may rhyme faster and louder than the rest. But it would be a very odd moment for the oldest pattern in the book to suddenly break.

The economics, in one picture

Why does cheapness explode the quantity for some things and not others? One idea does all the work here, and it's worth having in plain sight.

Economists call it elasticity. Drop the price of salt and nothing much happens — you already have all the salt you want; demand is inelastic. But drop the price of something with near-limitless uses — light, computing — and the quantity used detonates, because there was a vast queue of uses waiting for the price to fall. That's elastic demand, and it is the whole game.

So the only question that matters for the AI era is: how elastic is the demand for thinking? And the answer is that it may be the most elastic thing that has ever existed — because the supply of problems is effectively infinite. Think of everything that never got solved simply because the expertise cost too much. The corner shop that could never afford a lawyer, an analyst, or a designer. The kid with no tutor. The clinic in a small town with no specialist. The small idea that was never worth building software for. There is a bottomless backlog of problems the world has always had but never met, because meeting them required expensive human cognition — and that is precisely the thing whose price is now falling through the floor. Cheap thinking doesn't run out of things to think about. It runs toward a mountain of them we've been walking past for centuries.

A divided world, and why that's the opportunity

None of this lands on a flat, uniform planet. It lands on a world sharply split in two — and the split is exactly what makes the reshuffle so interesting.

On one side sit the ageing economies — Japan, much of Europe, Korea, now China — running low on workers, with ever more retirees leaning on ever fewer hands. For them AI is not a job-thief; it's a desperately needed extra pair of hands, a way to keep the clinics and factories and care homes running when the people simply aren't there. On the other side sit the young ones — India, with a median age around twenty-eight and well over a million engineers graduating every year; and Africa, younger still, with the largest unmet needs on Earth. For them, cheap cognition is rocket fuel: a nation of young builders who can now build products, not just staff other people's back offices; and a first-ever AI tutor, health worker, and legal helper for hundreds of millions who never had access to any of it.

And the deep, almost poetic point is that these two halves are made for each other. One side has work and not enough people; the other has people and not enough opportunity. Cheap intelligence, plus remote work, is a way to route one to the other — the biggest reshuffle of who-does-what since the shipping container, and a live example of the oldest idea in trade, comparative advantage, running at the speed of a video call. The young graduate in Pune or Lagos and the ageing firm in Osaka or Milan are, increasingly, each other's answer.

The honest part, because the optimism is worthless without it

I promised the defensible kind of hope, so here is the ledger's other side, briefly and squarely. Jevons giveth and Jevons taketh: the same paradox that promises abundance also means efficiency can drive more energy use, which is the real environmental worry I wrote about in The weight of a thought. And "the aggregate works out" is cold comfort to the specific person whose specific rung of the ladder gets automated this year — the entry-level coder, the junior analyst, the call-centre agent. History's transitions have often been brutal for a generation even when they were glorious for the century; the Industrial Revolution delivered decades of flat wages — economists call it Engels' pause — before the broad gains finally arrived. Abundance, in other words, is the likely sum, but nothing guarantees it's shared, or that the transition is gentle. Those are choices — about training, about who captures the gains, about rebuilding the bottom of the ladder faster than the middle gets automated. The paradox tells you the pie will grow. It says nothing about who gets a slice. That part is on us.

More light

But grant us those choices and look at where the arrow points, because it is genuinely thrilling. A world where thinking is nearly free is a world where every problem finally gets a look. Where the small business gets its analyst and the village gets its diagnostician and the curious kid anywhere gets the patient tutor that used to belong only to the rich. Where a founder of one can wield the leverage of a team of fifty, and the sheer number of things worth attempting explodes — which is the whole builder's opportunity of this moment. And where, as the machines take the rote cognition, the scarce and valuable thing becomes the part they can't do: judgement, taste, care, trust, the human being in the loop who decides what's worth doing at all.

That is what the history of light has been quietly promising the whole time. We never used cheaper light to sit in more darkness. We used it to see. The AI era, read through a 160-year-old paradox about coal, is not the story of running out of work. It's the story of finally being able to afford all the work that always mattered — the problems we've walked past in the dark for want of a lamp. Jevons saw the shape of it in a coal pit in 1865. It turns out to be the most optimistic sentence in economics, if you're willing to read it forward: make the good thing cheap, and you don't get less of it.

You get more light.

Sources & further reading: W. S. Jevons, The Coal Question (1865), for the original paradox; William Nordhaus, "Do Real-Output and Real-Wage Measures Capture Reality? The History of Lighting Suggests Not" (1996), for the collapsing price of light; James Bessen, Learning by Doing (2015) and his work on ATMs and bank tellers, and on spreadsheets and accountants; the "lump of labour fallacy" as a long-standing term in economics; on the global transition, Richard Baldwin, The Globotics Upheaval (2019) on remote work and telemigration, and standard demographic data (UN population estimates, national statistics) for median ages and graduate numbers; David Ricardo on comparative advantage; and, on the caution side, the economic history of "Engels' pause" during the Industrial Revolution (Robert Allen and others). Figures are widely-cited estimates and orders of magnitude; where a number carries real uncertainty, treat it as a direction of travel, not a decimal.