Patterns of progress

The single most important fact about human progress is how recent all of it is. Compress 300,000 years of modern humans into one calendar year: agriculture starts on 29 December, writing on 30 December, the printing press at about 22:00 on 31 December, and the industrial revolution in the last six minutes before midnight. Electricity, antibiotics, computers and the internet all happen in the final two minutes.

Look at the life expectancy row in particular. It is flat for eleven thousand nine hundred years and then more than doubles in two hundred. That row is the whole argument.

A caution: pre-modern figures are reconstructions from fragmentary evidence — skeletal remains, tax rolls, parish records — and estimates vary by a lot. Population figures for 10,000 BC span roughly 2 to 10 million depending on the source. The shape is not in dispute; the precise values are.

The flat-then-vertical shape is also why almost every intuition people have about progress is calibrated on an unrepresentative sliver. Everyone alive has only ever experienced the vertical part, which makes it feel like the normal condition. Historically it is the anomaly.

Why did twelve thousand years of genuine invention produce no improvement in living standards? Because until about 1800 the gains were converted almost entirely into population rather than into wealth.

A better plough raises yields. Higher yields mean more surviving children. More children mean more mouths, until food per person returns to roughly where it was — at a larger population. The society is bigger, more sophisticated, more numerous, and its median member is no better off. This is the Malthusian mechanism, and the historical record fits it uncomfortably well: China, Europe and the Middle East all show centuries of technological improvement with roughly flat wages.

The clearest evidence is the reverse case. After the Black Death killed perhaps a third of Europe, real wages for surviving labourers rose sharply and stayed high for over a century — not because anything improved, but because the ratio of land and capital to people improved. Prosperity per head was a function of scarcity of heads.

The escape had to break the link between output and births, and that is exactly what happened after 1800: output began compounding faster than population could absorb it, and then — see Pattern 8 — birth rates themselves collapsed. Both halves were required. This is why the industrial revolution is a genuine discontinuity rather than merely a fast patch of ordinary progress.

The trap explains something else that is otherwise puzzling: why so many sophisticated pre-modern societies felt no urgency about labour-saving machinery. When labour is abundant and cheap, and the surplus from saving it accrues to a landowner rather than the worker, the incentive to mechanise is weak. Technology is not adopted because it exists. It is adopted when it pays.

Every era of human history is legible as an energy regime, and the transitions between them are the deepest structure in the record.

• Muscle — human and animal. Caps everything at what a body can do, which is a few hundred watts sustained.
• Water and wind — the first non-biological power, and the first that is site-specific. Mills concentrate industry at rivers.
• Coal — the break. Energy stops being a function of land area and becomes a function of extraction.
• Oil and gas — energy becomes portable and dense enough for engines that move.
• Electricity — not a source but a carrier, and that is why it changed more than any single source did: it made energy divisible and transmissible.

Now the part that is consistently misread. These are usually described as transitions, as if one fuel replaced another. In absolute terms, they did not. Vaclav Smil's central observation is that historical energy transitions have been additive: the world burns more wood today than it did in 1850, more coal than it did in 1950. Each new source captured the growth rather than displacing the incumbent. Shares changed; absolute quantities almost never fell.

Two implications worth carrying forward. First, energy transitions are slow — half a century is the historical norm, because the capital stock that burns the old fuel has a long life and gets used until it wears out. Second, no previous transition has ever required the old source to actually decline in absolute terms. The current one does. That makes it structurally unlike its three predecessors, and the historical base rate is a poor guide to it.

Progress is not evenly distributed across inventions. A small number of them — economists call them general-purpose technologies — restructure everything downstream, and the rest are refinements.

They share three properties: they apply almost everywhere rather than to one industry; they keep improving for a long time after introduction; and they make other people's innovations possible, so the returns compound outside the original invention.

Notice what most of them have in common: they collapse the cost of something to near zero. Printing did it for copies, electricity for delivered power, computing for arithmetic, networks for transmission. When a cost that everything depends on falls by orders of magnitude, activities that were previously absurd become routine — and that, not the invention itself, is where the growth comes from.

Notice also the spacing. The gaps between them shrink — roughly 7,000 years, then 5,000, then 3,000, then 300, then 100, then 60. That compression is real and it is the strongest argument that progress accelerates. Pattern 9 is the strongest argument that it may have stopped.

This is the most useful pattern for anyone trying to read the present, and the least intuitive. A transformative technology does not show up in productivity statistics when it is invented. It shows up roughly forty years later.

Electricity is the canonical case. Practical generation and distribution existed by the 1880s. American manufacturing productivity did not accelerate until the 1920s. The delay was not technical — electric motors worked fine. It was that factories had been built around a steam engine driving overhead line shafts, and simply replacing the engine with a big electric motor changed nothing. The gain arrived only when a generation of new factories was laid out from scratch around small individual motors on each machine: single-storey, spread out, organised around the flow of work instead of around a driveshaft. That required new buildings, new managers, and the retirement of the people who knew the old way.

"You can see the computer age everywhere but in the productivity statistics." — Robert Solow, 1987

Solow wrote that roughly forty years after commercial computing began. Measured US productivity growth then accelerated in the late 1990s — on schedule, if you believe the pattern. The economic historian Paul David made exactly this argument in 1990, using the dynamo to predict the computer's delayed payoff.

Why the lag is so consistent: the technology is the easy part. What takes a generation is the complementary stuff — new physical capital, new firm structures, new skills, new regulation, and the death or retirement of the incumbent way of doing things. Max Planck's remark about science advancing one funeral at a time applies at least as strongly to industry.

The forecasting implication is sharp and cuts both ways. When a new technology appears and the productivity numbers do not move, that is consistent with it being transformative — the historical base rate says wait. And when enthusiasts promise it will show up next quarter, the base rate says it will not.

No society has ever stayed at the frontier. The location of the most advanced technology, the most productive economy and the most active intellectual life migrates, reliably, over centuries.

Mesopotamia and Egypt, then Greece, then Rome, then the Islamic world through the 8th–13th centuries, then Song China — which had cast iron, gunpowder, printing, paper money and a compass centuries before Europe — then northern Italy, then the Netherlands, then Britain, then the United States.

Two facts sit awkwardly together. Around 1500, and arguably as late as 1750, China and India were roughly comparable to western Europe in living standards and in several respects ahead in technology. India and China together are estimated to have produced something close to half of world economic output. Within 150 years Europe had diverged so sharply that the gap looked permanent and was widely explained as innate. It was not permanent.

The reversal is happening now and it is the largest economic event of the past half century. Convergence is not a prediction; it is a measurement. What is contested is why the divergence happened at all — coal geography, colonial extraction, institutions, property rights, competitive fragmentation of European states, scientific culture. Every one of these is a serious argument, none is decisive, and anyone stating the answer confidently is overselling.

The pattern that is not contested: leadership is temporary, it has never been retained by any society for more than a few centuries, and the incumbent has never seen the handover coming. There is no reason to think the current arrangement is the exception.

The scientific revolution is usually explained as a method — hypothesis, experiment, falsification. That undersells it. The method mattered, but the deeper change was institutional: knowledge stopped being private property and became a published, contestable, cumulative commons.

Before it, technical knowledge lived in guilds that enforced secrecy, in workshops that passed it by apprenticeship, in alchemists writing deliberately in code. Knowledge was a competitive asset, so it was hoarded, so it died with its holders, so every generation re-derived it. That is the mechanism behind the losses in the record — Roman concrete, Greek fire, the Antikythera mechanism's gearing. None of them were lost because nobody could understand them. They were lost because too few people were allowed to.

The change was a stack of unglamorous institutions:

• Printing — identical copies, so two people in different cities could argue about the same page and the same diagram.
• The journal — the Philosophical Transactions from 1665 established priority through publication rather than secrecy, which made disclosure the self-interested move.
• The patent — a time-limited monopoly bought with full public disclosure. Whatever its modern pathologies, the original bargain was explicitly designed to end trade secrecy.
• The university and later the industrial laboratory — knowledge held by an institution that outlives the individual.

The measurable consequence is that knowledge stopped being lost. The long record is full of capabilities that appeared and disappeared. After roughly 1700, essentially nothing significant is lost again. That is the ratchet, and the ratchet — not any particular invention — is what makes the modern era different in kind.

It is also the pattern most obviously reversible. Trade secrets, classified research, paywalled literature, proprietary models and non-disclosure agreements are all movements back toward the pre-modern arrangement. The historical record is unambiguous about what that arrangement produces.

Every society that industrialises passes through the same three-stage sequence, in the same order, without exception so far.

1. High births, high deaths. Population roughly flat. This is all of human history until about 1750.
2. High births, falling deaths. Sanitation, food supply and medicine cut mortality — especially infant mortality — decades before birth rates respond. Population explodes. This is the stage that produced the 20th century.
3. Low births, low deaths. Fertility falls, eventually below replacement. Population stabilises, then declines and ages.

The lag between stage two and stage three is what produced the population surge, and its size is striking: world population roughly quadrupled in the 20th century alone.

What is happening now has no precedent. Global fertility has fallen from about 5 births per woman in 1950 to roughly 2.2 today, against a replacement rate of about 2.1. A majority of the world's people now live in countries below replacement. UN projections have world population peaking around 10.3 billion in the 2080s and declining thereafter; several independent models put the peak earlier and lower.

If that holds, it is the first sustained global population decline since the Black Death — and the first in history caused by choice rather than catastrophe. It runs directly counter to the Malthusian mechanism of Pattern 2: prosperity used to produce more people, and now it reliably produces fewer.

Why this matters beyond demography: more people has historically meant more ideas. If the rate of innovation scales with the number of people working on problems — and the evidence for that is reasonably good — then a shrinking, ageing population is a headwind on progress itself, not merely on pensions. That interaction, between Pattern 8 and Pattern 9, is arguably the most consequential open question about the next century.

The most important recent finding about progress, and the one that most complicates the acceleration story.

Bloom, Jones, Van Reenen and Webb documented in 2020 that across many domains, sustaining a constant rate of improvement now requires vastly more research effort than it used to. Maintaining Moore's law took roughly eighteen times more researchers in the 2010s than in the early 1970s. The same shape appears in agricultural yields, in medical research measured against mortality, and in the aggregate: research productivity falls by something like 5% per year.

Progress has not stopped. What has changed is its price. Each successive advance costs more inputs than the last, and only a rapidly growing research workforce has kept the output rate steady. Sit that next to Pattern 8 and the tension is obvious: the input that has been compensating is the one about to stop growing.

Corroborating signals: total factor productivity growth in developed economies has been slower since about 1970 than in the half century before. Clinical trials cost more per approved drug. The share of papers and patents that are genuinely disruptive rather than incremental appears to have declined for decades. And some things have gone backwards: there is no supersonic passenger service, and it took half a century to return people to the vicinity of the Moon.

The counter-argument is serious. Measured productivity may simply be bad at capturing value that is free at the point of use. Some frontiers have moved extraordinarily fast — genome sequencing has fallen in cost far faster than Moore's law, solar modules have fallen roughly 90% in a decade, and an mRNA vaccine was designed within days of a published sequence. It is entirely possible that the old general-purpose technologies have matured onto their plateaus (Pattern 4's S-curve) while a new one is just starting its climb — in which case the measured slowdown is exactly what the forty-year lag of Pattern 5 predicts you would see just before an acceleration.

Both readings fit the current data. That ambiguity is genuine, it is not resolvable from where we are standing, and anyone who tells you otherwise is selling a position.

A technology's benefits are immediate, concentrated and obvious. Its costs are delayed, diffuse and statistical. The record shows an alarmingly consistent gap between the two — roughly forty to seventy years from widespread adoption to effective control.

Two details in that table deserve attention. Svante Arrhenius calculated the warming effect of atmospheric carbon dioxide in 1896 — the physics was not a late discovery, it was a very early one that took a century to be treated as decision-relevant. And the Montreal Protocol is the outlier in the other direction: from hypothesis to binding global agreement in thirteen years. It is the strongest evidence that the lag is not a law of nature. It is a choice about institutions, and it was made once, quickly, when the harm was legible, the substitutes existed and the industry was concentrated enough to negotiate with.

The generalisable form: the lag is longest where the benefit is immediate and the harm is slow, statistical, and borne by people other than the beneficiaries — including people not yet born, who are represented in no negotiation. That is not a technological property. It is a governance property, and it is the pattern with the most direct bearing on what happens next.