Futures

Strip everything else away and the century turns on the interaction between two of the regularities in the record.

Pattern 8 says the human population is about to stop growing and start shrinking, for the first time in history for reasons other than catastrophe. Global fertility is already at roughly replacement and still falling.

Pattern 9 says ideas are getting harder to find — that constant progress now requires a continuously growing research workforce, because each advance costs more than the last.

Those two are difficult to hold at once. The mechanism that has been paying for continued progress is more people thinking; the input about to reverse is more people. If the relationship between population and ideas is as strong as the evidence suggests, the second half of this century faces a headwind that no previous era has faced, and it arrives regardless of how good the technology gets.

There is exactly one thing that clearly breaks the collision, and it is the reason so much attention is on it: a technology that increases the number of effective researchers without increasing the number of humans. That is the actual load-bearing question about machine intelligence — not whether it can write or draw, but whether it can do the thing that a growing population used to do.

If it can, Pattern 9 is a footnote. If it cannot, the slowdown is structural and the rest of this page is a description of a long plateau.

This is the most confident statement on the page, because most of the people who determine it are already alive.

Fertility has fallen below replacement across most of the world, including nearly all of East Asia, Europe, and increasingly Latin America and South Asia. Once a country's fertility falls, it has essentially never returned to replacement level — no policy intervention anywhere has reliably reversed it, and the countries that have tried hardest have achieved modest, temporary effects at high cost.

What follows nearly mechanically:

• Ageing before shrinking. Even before totals fall, the ratio of workers to dependants deteriorates. Every pension and healthcare system in the developed world is built on an assumption that is already false.
• Migration becomes the central political question. It is the only fast lever on a national workforce, which guarantees it stays contested.
• Growth models break. Nearly all economic reasoning since 1800 assumes a growing population. The intuitions built on that period do not transfer.
• Depopulation is uneven. Some regions empty while others grow, and infrastructure built for a larger population becomes a maintenance liability — a pattern already visible in rural Japan, Italy, and parts of eastern Europe.

The counter-consideration: fewer people is not straightforwardly bad. Per-capita resource availability improves, environmental pressure eases, and the Malthusian logic of Pattern 2 runs in reverse. The pessimism is about the rate of change and about innovation, not about the level.

Pattern 3 gives an unusually clean base rate: energy transitions take roughly fifty to sixty years to reach half of supply, and historically they have been additive — the old fuel kept growing in absolute terms while losing share.

The current transition differs in three ways that all matter.

• It must be subtractive. No previous transition required the incumbent to actually decline. This one does, which removes the mechanism — capturing growth rather than displacing incumbents — that made the previous three politically easy.
• It is deliberate rather than economic. Coal beat wood and oil beat coal because they were better and cheaper for the user. This one began as policy. That is changing: solar and wind are now the cheapest new generation in much of the world, which moves it onto the historical footing where transitions actually succeed.
• It has a deadline, which nothing in the historical record did.

What the cost curves say. Solar modules follow a steep and remarkably persistent learning curve — roughly 20% cost reduction per doubling of cumulative production, sustained for four decades. Batteries have followed a similar path. Learning curves of that kind have been more reliable predictors than expert forecasts, which have underestimated solar deployment almost every year for twenty years. That is a real and under-appreciated point in favour of speed.

What the base rate says against it. Generation is the easy part. The hard parts are storage, transmission, industrial heat, cement, steel, shipping, aviation and fertiliser — and the fact that the existing capital stock has a thirty-to-fifty year life and gets used until it wears out. Turnover, not invention, is the binding constraint. That is precisely the Pattern 5 lag, in physical form.

Reasonable expectation: electricity decarbonises substantially faster than most forecasts assumed; industry, heat and aviation take considerably longer; the transition is roughly on the historical fifty-year clock rather than the ten-year one, with the outcome hinging on how much of the remaining capital stock is retired early rather than run to end of life.

Two coherent readings of the present exist, they use much of the same evidence, and the difference between them is most of what matters about the next fifty years.

Both are consistent with the current data. The honest position is that they are not distinguishable from here, and that they are not mutually exclusive — a plateau in mature technologies and the start of a new S-curve produce very similar-looking statistics for a decade or two.

What would actually settle it is in the next card.

Since the two readings cannot be separated by argument, the useful thing is to name in advance what evidence would move you. These are the indicators that discriminate, rather than the ones that merely feel significant.

The most decisive single one is the second row. Pattern 9 is the mechanism behind the slowdown thesis; if research productivity stops falling, the thesis loses its engine. If it keeps falling while research spending climbs, no amount of visible technological excitement changes the arithmetic.

Pattern 10 is the most regular finding in the whole record: roughly forty to seventy years between a technology's adoption and effective control of its harms. It has held for leaded petrol, asbestos, tobacco, CFCs and carbon dioxide.

Applied forward, it predicts that the harms of technologies now being deployed will be addressed somewhere around the 2060s–2080s. That is not a claim about any specific technology being harmful. It is a claim about institutional response times, and those have been remarkably stable.

Why the lag is structural rather than accidental:

• Benefits are immediate, concentrated and have an owner. Harms are delayed, diffuse and statistical.
• The beneficiaries are organised and present. The harmed are dispersed, and sometimes not yet born.
• Proof arrives after deployment, because the harm is what generates the evidence.
• By the time evidence is unambiguous, the technology has constituencies — employment, capital, dependency — that resist.

The Montreal Protocol proves it is beatable. Thirteen years from hypothesis to binding global agreement, and it worked. The conditions that made it possible are worth naming precisely, because they are the checklist: the harm was legible and attributable, substitutes existed at tolerable cost, the producing industry was concentrated enough to negotiate with, and the affected parties included everyone. Where those four hold, fast action is possible. Where they do not, expect the historical lag.

Run that checklist against the current frontier and it is not encouraging: harms are diffuse and hard to attribute, substitutes are contested, and the affected populations are unevenly distributed. That is the profile of a slow response, not a fast one.

A useful discipline is to ask what the hard limits are, as distinct from the current ones.

Energy. On the Kardashev scale as Carl Sagan made it continuous — K = (log₁₀ P − 6) ÷ 10, with P the power a civilization uses in watts — humanity stands at about 0.73: some 178,000 TWh of primary energy in 2025, an average of 2×10¹³ W. Type I is 10¹⁶ W, about 500 times today, and each 0.1 on the scale is ten times more power. Where civilization stands tracks the figure each year beside the measures that say whether the energy is doing any good. The binding constraint arrives well before Type I and is thermodynamic rather than economic: waste heat. Any energy used on a planet's surface ends up as heat that must radiate away, so there is a ceiling on planetary energy use that is entirely independent of how it is generated. It is far above current use, but it is not infinite, and it is the one limit that no efficiency improvement can move.

Materials. Ore grades have declined for a century — modern mining works rock that would have been waste in 1900 — and this has been offset by better technology, which is a race rather than a solution. The genuine long-run answer is circularity: at present the economy is overwhelmingly linear, and recycling rates for most specialty metals are extremely low. Nothing about that is technically hard; it is an accounting and design problem, which is precisely why it has not been solved.

Information. There may be a ceiling on how much any society can know and coordinate. Pattern 9 is consistent with a knowledge frontier that gets harder to advance simply because there is more of it to learn before you can add to it — the time needed to reach the frontier in any field keeps growing, which shortens the productive career at the edge.

Coordination. On the record, this is the binding constraint far more often than physics. Most civilizational failures in Pattern 6's list of handovers were not resource exhaustion — they were the failure to organise resources that still existed.

Not predictions, and not equally likely. These are the shapes consistent with the regularities in the record — with the historical base rate noted for each.

The track record of predicting the future is bad enough to be worth studying as its own pattern, because the failures are systematic rather than random.

• Technology is over-predicted, society is under-predicted. Mid-century futurists got atomic power, space travel and automation roughly right as capabilities and missed the contraceptive pill, the collapse of fertility, the service economy and the fact that computers would mostly be used for talking to other people.
• Timelines are compressed for the exciting and stretched for the boring. Sanitation, refrigeration and container shipping changed more lives than anything predicted on a magazine cover.
• Extrapolation misses the S-curve. Every exponential is an S-curve that has not bent yet. Predictions made mid-curve are almost always too aggressive; predictions made at the plateau are almost always too conservative.
• The binding constraint moves. Nuclear power's constraint turned out to be political, not physical. Space's constraint turned out to be economic, not engineering. Forecasts fail because they assume today's constraint is permanent.
• Nobody predicts institutional change, which is where most of the variance actually lives.

Applied to this page: the highest-confidence claim here is the demographic one, because those people already exist. The energy claim rests on cost curves that have held for forty years, which is decent but not proof. Everything about acceleration versus slowdown is genuinely undecided, and the scenarios are shapes rather than probabilities.

The most likely error in this document is the same one that has been made every time before: correctly describing the machinery and completely missing what people decide to do with it.