The Future Is Older Than You Think: Forgotten Predictions and the History of Tomorrow
In 2007 a man published a book about the future. He held no chair in futurology, ran no consultancy, and had no institutional reason to be believed. He looked at the world, concluded that he could see where it was heading, and printed his conclusions with his name and the year on the cover, which is more than most forecasters have the nerve to do.
Eighteen years later the book sits on his son’s desk.
Some of what it promised arrived. Some of it went emphatically the other way. And a surprising share of it is still in motion, which turns out to be the most interesting category of all.
The Future Is Older Than You Think takes one dated, checkable, thoroughly human forecast and treats it the way an engineer treats a failure report — not hunting for the fool, but for the mechanism. Then it goes looking for the same mechanism across two centuries of confident prophecy, from Malthus and the electric utopias to the metals that were supposed to run out in 1987 and the office that has been going paperless since before most of us were born.
What emerges is not a catalogue of human stupidity. It is a pattern with a shape, an explanation for why forecasting fails in the specific ways it does, and an uncomfortable account of why the predictions that succeed are the ones nobody remembers.
The author ends by doing the only honest thing available to anyone who writes a book like this: putting his own predictions on the record, with dates attached, for somebody else to grade.
Keywords: forgotten predictions, history of the future, why forecasts fail, futurology and its failures, systems thinking and feedback, technological change and society, intellectual honesty
Contents
Chapter One — The Book on the Desk. 15
Chapter Two — Why Nobody Keeps Score. 24
Chapter Three — The Arrow and the Pendulum.. 31
Chapter Four — Three Verdicts, Not Two. 37
Chapter Five — The Man Who Wrote Down Why He Would Be Wrong. 42
Chapter Six — Furniture: The Predictions You Live Inside. 46
Chapter Seven — The Comedy of the Flying Car 50
Chapter Eight — Scarcity Summons Its Own Cure. 54
Chapter Nine — Malthus Never Left the Building. 58
Chapter Ten — The Metals That Refused to Run Out 62
Chapter Eleven — Peak Everything. 66
Chapter Twelve — The Paperless Office Enters Its Fiftieth Year 70
Chapter Thirteen — The City That Would Not Die. 74
Chapter Fourteen — The Barrier That Held the Thing Together 78
Chapter Fifteen — The Safety Valve That Turned Out to Be a Pipe. 82
Chapter Sixteen — Efficiency Makes Us Use More. 86
Chapter Seventeen — The Machine That Listens at Three in the Morning. 90
Chapter Eighteen — Right Result, Wrong Machinery. 94
Chapter Nineteen — The End of Money, Again. 98
Chapter Twenty — The Abolition of Work. 102
Chapter Twenty-One — The Abolition of the Family. 106
Chapter Twenty-Two — The Abolition of Death. 110
Chapter Twenty-Three — The State Was Supposed to Wither 114
Chapter Twenty-Four — Every Utopia Is a Portrait of Its Own Decade. 118
Chapter Twenty-Five — The Paris Postcards of 1900. 122
Chapter Twenty-Six — Doomsday Pays Better Than Calm.. 126
Chapter Twenty-Seven — My Warning Prevented It, and Other Escape Hatches. 130
Chapter Twenty-Eight — When the Pendulum Does Not Come Back. 134
Chapter Twenty-Nine — What Eighteen Years Can and Cannot Settle. 138
Chapter Thirty — How to Read Any Forecast in Four Questions. 142
Chapter Thirty-One — Predictions, With Dates. 146
Chapter Thirty-Two — Predictions Against the Obvious. 150
Chapter Thirty-Three — A Book on Somebody’s Desk in 2044. 154
Case Study One — Concorde and the Supersonic Century. 166
Case Study Two — Too Cheap to Meter 170
Case Study Three — The Aswan High Dam.. 173
Case Study Four — The Cane Toads of Queensland. 176
Case Study Five — The Sparrows. 179
Case Study Six — Prohibition. 182
Case Study Seven — The Bridge That Wobbled. 185
Case Study Eight — DDT and the Peregrine. 188
Case Study Nine — The Man Who Measured the Air 191
Case Study Ten — The Delta Works. 194
Case Study Eleven — The Sea That Was Spent 197
Case Study Twelve — New Coke. 200
Case Study Thirteen — Iridium.. 203
Case Study Fourteen — The Segway. 206
Case Study Fifteen — The Tag That Was Going to Replace the Barcode. 209
Case Study Sixteen — The Water Under Punjab. 212
Case Study Seventeen — The Fifth Generation. 215
Case Study Eighteen — The Newton. 218
Case Study Nineteen — Wash Your Hands. 221
Case Study Twenty — One Child. 224
Preface
There is a book on my desk that was printed in 2007, and it is about the future.
This is an awkward object to own. Books about tomorrow are meant to be read while tomorrow is still safely ahead of them, which is the only condition under which they are any fun. Nobody wants to be handed a prophecy and a stopwatch in the same gesture. But eighteen years have passed since this one came off the press, and eighteen years is long enough for a book about the future to have become, in several places, a book about last Tuesday.
The second awkward feature is that my father wrote it.
A reader is entitled to know what kind of transaction is being proposed here, so let me be plain. This is not a tribute. It is also not a settling of accounts, which would be both uglier and duller. It is that I happen to have on hand something almost impossible to obtain in the study of prediction, which is a forecast that cannot run away.
Consider what normally happens to a prediction. Somebody makes one. It is published, discussed, sometimes alarming, occasionally lucrative. Then time passes and one of two things occurs. If the prediction was wrong, everyone forgets it, including its author, who has by then moved on to a fresh prediction unencumbered by the previous one. If it was right, it stops being a prediction at all and becomes an ordinary fact about the world, and the author gets no credit whatsoever, because nobody hands out medals for having foreseen that there would eventually be a great many telephones.
Forecasters are therefore among the least accountable professionals alive. Not through conspiracy — through the ordinary mercy of human memory. The record simply is not kept. There is no ledger, no scoreboard, no institution whose job it is to circle back in twenty years and ask how it went. Every generation of prophets inherits a field that has been thoughtfully swept clean of its own past by the reliable mechanism of nobody bothering to write anything down.
A dated book, printed and never revised, is a ledger. That is why this one is on my desk, and it is also why I did not choose a more celebrated specimen. Famous forecasters have reputation management. My father has a paperback.
There is also the matter of the year, which is luckier than it looks. Two thousand and seven sits on the lip of several cliffs at once. The smartphone existed but almost nobody had one. Social networks were a curiosity for students rather than the plumbing of public life. The financial system was about to demonstrate that a great many clever people had been extending a line. Artificial conversation was a novelty you could visit on a website and giggle at. A forecast written from that ledge is not a forecast written from safety; it was made just before several of the relevant landscapes rearranged themselves, which means its errors are informative rather than merely quaint.
Now, the temptation, which I want to name so that I can refuse it in public.
The temptation is scoring. It would be easy, and mildly entertaining for perhaps thirty pages, to walk through the thing with a red pen handing out marks. Right about this. Wrong about that. Hopelessly wrong about the other. The reader would spend a pleasant afternoon feeling superior to a man in 2007, which is not a demanding sport, since the man in 2007 did not know about a number of things the reader learned by the straightforward method of living through them.
I am going to do a certain amount of this, because you cannot skip the evidence, but only long enough to reach the part that matters. Because when the hits and the misses are laid side by side and stared at for a while, something impolite happens to them. They stop looking like a list of lucky guesses and unlucky ones. They start looking like the same failure occurring over and over, in both directions at once.
Here is that failure, stated as simply as I can manage.
Prediction imagines the future as an arrow. Something is happening; it will go on happening; extend the line and read off the answer. This is such a natural way for a mind to work that it is genuinely difficult to catch yourself doing it. But almost none of the systems that matter — economies, cities, appetites, populations, cultures, technologies — behave like arrows. They behave like pendulums. Every pendulum has a restoring force, and a restoring force is invisible while the swing is small. You discover it exists only when the swing grows large enough to summon it, which is to say, precisely at the moment the forecaster has finished drawing his line and gone home.
This is not a claim that nothing ever changes. Things change enormously; that is the whole reason forecasting is worth attempting. It is a claim about the shape of the change. Movement in one direction generates the conditions for movement back, and the forecaster who sees only the outbound half publishes a book about a destination that is in fact a turning point.
What makes my father a useful specimen rather than merely a convenient one is that he named this problem himself, in his own opening pages, before making a single prediction. He proposed something close to a law: that the trends which end up mattering are the ones nobody is currently watching. He wrote down the exact reason his forecasts would fail, and then went ahead and made them for twenty-two chapters. In the whole literature of failure I have not found a cleaner demonstration that understanding a mechanism grants no immunity from it. Engineers know this. It is why the reports keep getting written.
So this book has a thesis, and the thesis is sitting in the title, three times over.
The future is older than you think, first, because the predictions themselves are secondhand. Very little of what anyone says about tomorrow is new. The end of the nation-state has been imminent for a century and a half. The abolition of work, of money, of the family, of death — these are not insights, they are heirlooms, passed from one confident generation to the next and lightly reupholstered each time in whatever technology happens to be exciting. Strip the vocabulary from a Victorian utopia and a digital one and you are frequently looking at the same document.
Second, the future is older than you think because the world keeps returning to where it has been. Not exactly — a pendulum in the real world is damped, and it never comes back to quite the same point — but returning nonetheless. What looked like a one-way journey into a new condition turns out to have been the outward half of a swing, and the return half is where most forecasts quietly go to die. The past is not behind us in a straight line. It is the thing we are oscillating around.
Third, and most annoying, the future is older than you think because the successful predictions are already inside your house and you have stopped noticing them. Correct forecasts have a miserable career path. They come true, and in coming true they become furniture, and furniture receives no citations. Nobody murmurs that it was foretold when they consider the fact that they can see a relative’s face while speaking to them from another continent. The consequence is severe and almost never acknowledged: the archive of prediction is structurally biased toward failure. Our shared conviction that forecasters are always wrong is partly a fact about our filing system rather than a fact about forecasters.
I should say what this book is not, because two neighboring genres are easy to fall into and both are cheap. It is not a debunking book, the kind that lines up dead experts for target practice and concludes that nobody knows anything, which is a conclusion that costs its author nothing and helps its reader not at all. Nor is it one of those relentlessly sunny volumes that answers every warning with a chart showing that things got better anyway. Both of those are themselves forecasts wearing the costume of history — the first predicting that expertise will always disappoint, the second that the trend line of improvement will hold. I would rather look at the machinery.
And you have a stake in this, which is my last reason for bothering. You are surrounded at this moment by claims about the next twenty years, most of them delivered with total confidence by people who will never be graded on them, and some of them are shaping what you study, where you live, and what you are frightened of at two in the morning. Learning to recognize the shape of a forecast that is about to bend is not an academic pleasure. It is closer to learning which ice will hold.
A word about method, since I am asking you to trust a scoreboard.
I have tried to grade this book the way one ought to grade an opponent, which means looking for contradictions before looking for confirmations. It is dismally easy to read an old text charitably and find prophecy everywhere, because old texts are vague and the world is large and the human capacity to see faces in clouds does not switch off when the clouds are made of prose. So the rule I set myself was this: a prediction counts as a hit only if it specifies something that could have gone otherwise, and if the thing that actually happened matches not just the mood but the substance. Anything that requires generous interpretation is not a hit. Anything that came true by a route the author explicitly ruled out is not a hit either, and there is at least one splendid example of that in what follows.
The same discipline has to apply in the other direction. A prediction is not a miss simply because it has not happened yet, and this is where a great deal of the fun in this book lives.
Which brings me to the honest limitation of the whole exercise.
Eighteen years is not a verdict. It is one swing. Anyone who tells you that a forecast made in 2007 has now been definitively settled is doing to the recent past exactly what the forecaster did to the future — running a confident line through insufficient data and calling the result knowledge. So this book works with three verdicts rather than two. Some things happened. Some things went the other way. And a large and entirely respectable category of things is still in motion and will be decided, if it is decided at all, by people who are currently arguing with their teachers.
That third column is not a dodge. It is where the argument earns its keep, because the pendulum model makes claims about it. If the mechanism I am describing is real, then some of what currently looks like a failed prediction is a prediction in the middle of its return swing, and some of what currently looks vindicated is about to reverse. Those are not comfortable things to say. They are, however, checkable, which is the point.
And so to the last chapters, which I could not have written any other way.
It would be indecent to spend a book examining a man who was brave enough to put a date on his opinions and then decline to do the same. So the final chapters contain my own predictions. They are specific, they carry years, and at least one of them argues against something that presently looks obvious to nearly everybody, on the grounds that a method never turned on its own author is not a method but a hobby.
They will not all survive. That is arithmetic rather than modesty. I ask only that whoever eventually sits down with this book and a red pen extend me the courtesy I am attempting to extend here, which is to hunt for the mechanism instead of the fool. The people who came before us were not standing around waiting for us to show up with better ideas. They had ideas. What they lacked was our hindsight, and what they had instead was a deadline — which is precisely our situation, and will be yours.
The book is on the desk. Let us open it.
Chapter One — The Book on the Desk
The cover shows clocks and an eye.
Not a clock and an eye, arranged with any particular thought, but a stock photograph of several clock faces and a human eye, floating in the way that things floated on book covers in the middle of the last decade, purchased from an image library under a royalty-free license whose number is printed with touching diligence on the copyright page. Eight digits. Somebody typed them in. Whoever laid out this book wanted you to know that the clocks and the eye had been paid for.
I mention the cover because it tells you more about the year than the text does. Two thousand and seven was a moment when the visual language of the future was still assembled from parts — a clock for time, an eye for perception, a slight blur for profundity — and the parts were available for a modest fee. There is nothing embarrassing about this. It is simply the last generation of futurism that still had to buy its imagery from somebody else.
Inside is a book of twenty-two chapters, printed in Florida by a small press, written by a man living in a small town in Ontario, and it is about nothing less than everything that will happen to the human species. This is not a joke at its expense. The ambition is the point, and the ambition is what makes it useful.
Here is what it covers, in roughly the order it covers it. The internal architecture of a personality. The rise of invented online selves and what happens when they stop needing a human behind them. Whether a machine can perform psychotherapy. Whether we should be medicating ourselves into contentment, and whether the objection to doing so survives contact with the history of coca wine. The nature of imagination once imagination becomes a shared public space rather than a private one. The prospect of escaping the body. A design for a state run without paper money, without most of its bureaucracy, and without the assumption that everyone must be employed. Two chapters on immortality, one enthusiastic and one nervous. A chapter arguing that the erotic has been extinguished by its own abundance. A survey of the world’s civilizations and where each is heading. A chapter on terrorism, a chapter on barbarism, a chapter asking whether the modern world has quietly been won by the devil, a program for the future of philosophy, a list of the author’s rules for living, and a closing image of humanity as a ship that needs a course more urgently than it needs a good hull.
That range is the first thing worth noticing, because it is completely unlike the way we forecast now. Contemporary prediction is a specialist trade. The energy person predicts energy. The demographer predicts demography. The technologist predicts technology and then, in the final chapter, gestures at society in a way that his publisher insisted upon. Nobody sits down and attempts the whole board at once, partly because it is embarrassing to be wrong in public across so many fields, and partly because there is no career in it.
But the whole board is where the interesting failures live. A specialist forecast fails quietly and locally. A forecast that covers everything fails in ways that reveal the joints between subjects, and the joints are precisely where the trouble is. Almost every prediction in this book that went wrong went wrong at a joint — the place where a technology met a household, or a household met a price, or a price met a habit.
The second thing worth noticing is who wrote it, and here I have to be careful, because a son writing about his father has a natural pull toward two kinds of dishonesty and I would like to avoid both.
So, only the checkable facts. The man who wrote this book was not an academic and made no claim to be. His path to that desk had gone through the Orthodox priesthood, then the directorship of a clinical research center, then business. By 2007 he was running a company that trained people for work in pharmaceutical research, and — this matters — he had moved that entire operation onto the internet years before anyone was calling it a trend. He says so in the book, in the middle of the chapter on redesigning the state, as evidence that the redesign is practical. The office in the town near his home had gradually emptied out. One instructor worked from Toronto, another from Calgary. Students in Singapore took the courses at hours convenient to Singapore. The cost of the whole thing had collapsed.
Understand what he had actually done. He had run a small, real, working prototype of remote knowledge work, and it had performed exactly as he described. He then generalized from it to the entire species, and predicted that the cities would empty out.
This is the single most instructive sentence I can write about forecasting, so I am going to leave it alone for now and come back to it several times over the course of this book. The prototype was real. The mechanism was real. The extrapolation was wrong. And it was not wrong because he was foolish; it was wrong because a working example is the most persuasive and most dangerous thing a forecaster can possess.
The third thing about this book is its temperature, which is warm. It is not a doom volume. Chapter Seven opens by observing that nobody writes cheerful utopias anymore, everything is machine rebellions and asteroids, and announces that this one is going to be pleasant for a change. It then lists the seven problems of humanity, beginning with the fact that dying is undesirable and living is frightening, and proceeds to solve them. The tone is that of a man enjoying himself enormously on a winter evening.
Its method is worth describing, because the method determines the errors. This is not a book that argues from data. There are numbers in it — tax rates, export figures, suicide statistics, the price of a Chinese television — and they are generally accurate for their moment, but they arrive as illustration rather than as evidence. The actual engine of the argument is philosophical. Rousseau is quoted in French, twice, with translation supplied. Baudelaire arrives in the middle of a chapter about the death of the erotic, in the original, because the author evidently could not bear to paraphrase him. Kant is worked through at some length. Maslow’s hierarchy of needs is used as the structural spine of the whole enterprise, applied first to individuals and then, in the final chapter, to entire civilizations.
That is a nineteenth-century way of building an argument about the twenty-first century, and it has consequences in both directions. It is why the book is good on motivation — on what people are actually trying to get, which turns out to be the most stable thing in any forecast — and why it is poor on friction. A framework built out of needs and philosophy has no natural place to put a landlord, a zoning board, a supply chain, or a habit that people simply cannot be bothered to break.
There is also a detail on the title page that dates the book as precisely as the cover does. The author’s name is printed in an anglicized form. A Russian-language writer publishing philosophy in English through a small American press in 2007 shortened himself for the market, which is a small, unremarkable, entirely period-appropriate act, and which I mention only because it is the kind of thing that vanishes from the record within one generation. Somebody made a judgment about what an English-speaking bookshop would tolerate. Ten years later the judgment would have gone the other way.
And there is one section whose value has changed completely since it was printed. In the chapters on virtual personality and virtual psychology, the author reproduces his own conversations with the conversational programs that existed at the time — a chat system called George, its animated companion Joan, the venerable ELIZA, a novelty program presenting itself as God, and a transcript, taken from a magazine, of two such programs talking to each other. He includes them as evidence for an argument. They now function as something else entirely, which is a primary source. These are verbatim records of what it felt like to talk to a machine in 2007, written down by somebody who took the experience seriously at a time when almost nobody did, and preserved by the accident of having been printed.
Read them now and the sensation is peculiar. The programs are obviously, hopelessly limited. They contradict themselves within three exchanges. They have no memory. And yet the author, sitting there in Ontario, notices the thing that mattered: that the machine’s willingness to receive whatever you say, without judgment and without fatigue, is a feature rather than a deficiency, and that people will eventually prefer it. He is talking to a toy and correctly identifying what the toy is a prototype of. That is not a common skill, and it coexists in the same volume with the confident announcement that most schools will close.
I want that on the record because the standard modern posture toward old forecasts is contempt, and contempt makes you stupid. It is very easy to be clever about a man who thought we would all be growing steaks in vats and studying at home by now. It is considerably harder to notice that he also thought people would confess things to a machine that they would never confess to a human being, which is a stranger and truer observation than anything most professional futurists of that decade managed, and which nobody has ever congratulated him for.
There is one more feature of the object that qualifies it as a specimen, and it is the humblest one. Nobody has touched it since. There is no revised edition, no updated afterword, no quiet correction in the second printing. The prophecies stand as printed. Whatever else you can say about a small press in Florida, it does not go back and fix your predictions for you.
So the book on my desk is a fixed point, an unmodified record of what a well-read, unaffiliated, technically competent person believed about the twenty-first century in the last year before the smartphone became universal. It is not a great book. It was never going to change the world. But it has one property that the great books about the future almost never have, which is that it can be graded.
Let me set out, before we start grading, what I think makes any old forecast worth the trouble of grading. There are four requirements and this book meets all of them, which is rarer than it sounds.
It has to be dated and unrevised, so that nobody can quietly move the goalposts. It has to be specific enough that some outcomes would have counted as failures — a book which predicts only that things will be different is unfalsifiable and therefore uninteresting. It has to cover enough ground that the pattern of hits and misses can be compared across domains, since a single prediction tells you nothing about mechanism. And its author has to have been sincere, which is the requirement people forget. A forecast produced to sell a product or frighten a legislature is evidence about the producer, not about forecasting. This one was written by a man with no commercial interest in any of it, which is exactly why his errors are worth studying: they are the errors that thinking produces on its own, without help.
One last observation before we open it properly. The book contains almost no dates. Things will happen, are coming, will soon arrive, will eventually be accepted. Only twice does an actual year appear as a target, and both times the year belongs to somebody the author is quoting. This is not evasion; I do not think it occurred to him that a date was owed. But it means that grading him requires a judgment call on every single item, and it is the first thing I intend to do differently when my own turn comes at the end of this book.
Chapter Two — Why Nobody Keeps Score
Consider two people whose job is to tell you about the future.
The first stands in front of a map and says there is a seventy percent chance of rain tomorrow. The second sits in a conference chair and says that within a decade the nature of work will be transformed. Both are forecasters. Only one of them is any good, and the reason is not intelligence, education, or the difficulty of the subject. It is that the first one gets a report card and the second one does not.
Weather forecasting is one of the genuine quiet triumphs of the last century. A five-day forecast today is about as reliable as a one-day forecast was in the nineteen-eighties. This did not happen because meteorologists became cleverer than other scientists. It happened because a weather forecast is checked against reality by the following afternoon, every single day, forever, and the results are recorded and compared. When you are scored constantly and publicly, you improve or you are replaced by someone who does. When you say that the nature of work will be transformed within a decade, nothing whatsoever happens in ten years, because nobody has kept the tape, and because the claim was constructed — usually without conscious dishonesty — so that almost any outcome would count.
The most thorough attempt to actually keep the tape was made by the psychologist Philip Tetlock, who spent two decades collecting specific, dated, checkable predictions from several hundred experts in politics and economics and then, with monastic patience, waiting to see what happened. Twenty-odd thousand forecasts later, the result was published in 2005, two years before the book on my desk. The headline finding, which has been repeated so often that it has lost its sting, was that the average expert performed at roughly the level of an intelligent guess. The more interesting finding was underneath it.
Tetlock borrowed a distinction from an old essay about hedgehogs and foxes. The hedgehog knows one big thing and drives every question through it; the fox knows many small things and holds them loosely. The hedgehogs performed worse. And here is the part that matters for this book: hedgehogs performed worse not despite their confidence but through it. Having one powerful organizing idea is exactly what allows a person to extend a trend line without hesitation, because the single idea explains why the line will hold. The fox is worse at explanation and better at prediction, which is an uncomfortable arrangement for anyone who has to make a living talking.
Notice also what Tetlock had to do to obtain his data. He could not simply gather up predictions that already existed, because they did not exist in a usable form. He had to go and commission them, pin the experts down to specific outcomes and specific windows, and then store the answers himself for twenty years. That is what it costs to build a ledger in a field that has never had one.
Why has no one else paid that cost? Because nobody’s interests are served by it.
Look at where the money and the attention actually sit. The forecaster is paid on delivery, for the speech or the book or the strategy document, and the payment clears years before the verdict. The client wants a decision made this quarter, not an audit conducted in 2043. The publisher wants a book that sells now. The journalist wants a striking claim today. And the audience, which is the only party with an actual stake in accuracy, has no mechanism for recall and, more to the point, no appetite for it. Nobody wants to attend the twenty-year reunion of a prediction.
There have been attempts to build the missing institution. The most charming lives at the Long Now Foundation, which set up an arrangement in the early two-thousands where anyone could stake a public, dated, financial bet on a specific claim about the future, with the losing side’s money going to charity and the whole argument preserved in writing. The most famous entry has Ray Kurzweil wagering that a computer would pass a properly conducted Turing test by 2029, against Mitch Kapor’s conviction that it would not, with the terms of the test hammered out in exhausting detail precisely because both men understood that vagueness is the escape hatch.
That word — vagueness — deserves a paragraph of its own, because it is the load-bearing wall of the entire industry. Read any professional forecast closely and count the verbs. Things are poised to, set to, likely to, on the verge of. Trends are accelerating, though never to any stated value. Changes are coming, though no month is named. Something will be transformed, and transformation is a word with no failure condition attached, because any state of affairs can be described as transformed relative to some earlier one.
None of this is usually deliberate deceit. It is the natural shape a sentence takes when the person writing it is genuinely uncertain and also genuinely needs to be published. The uncertainty is honest; the grammar that absorbs it is what makes the claim unscoreable. And once a discipline has settled on that grammar, the accurate and the useless become indistinguishable in the text, which means the accurate forecaster receives no advantage. This is the mechanism by which a field can contain excellent practitioners and still not improve.
There is a related trick, so common that I no longer think of it as a trick: predicting something that has already started. If you announce in a given year that a technology already in the hands of ten million people is going to be important, you will be proved right, and nothing whatsoever has been contributed. A great deal of what passes for successful futurism on the conference circuit is this — reporting from slightly further up the same road, dressed as reporting from around the corner.
But blaming the forecasters lets the rest of us off, and I do not think we deserve it, because there is a demand-side explanation that fits the evidence better than any account of supplier misbehavior.
Most people, most of the time, do not consume forecasts in order to learn what will happen. They consume forecasts in order to settle how to feel. A prediction of collapse and a prediction of abundance are, for the overwhelming majority of readers, not competing pieces of information about the world but competing moods on offer, and the choice between them is made the way one chooses music. This is why the same person can hold, without discomfort, a general conviction that experts are always wrong and a specific terror about something an expert said last week. The general conviction is a position; the specific terror is a mood; they are stored in different places.
If that is what the audience is buying, then accuracy is not the product and no market pressure will ever produce it. Fund managers, to take the nearest comparison, are scored ruthlessly and publicly, quarter after quarter, and the majority still fail to beat a simple index over any long stretch. Scoring, it turns out, is necessary for improvement but nowhere near sufficient. What forecasting has is the worse arrangement of the two: no scoring at all, and an audience that would not read the scores.
What is striking about that arrangement is how few people have used it. A structure exists that would permit any confident person to make their confidence checkable, at trivial cost, with the added benefit of publicity. The uptake is a rounding error against the volume of prediction produced every week. This is not because forecasters are cowards. It is because a dated bet converts an asset into a liability. An unspecified prophecy pays forever; a specified one has an expiry date after which it either pays nothing or costs you.
Prediction markets are the other partial answer, and they work reasonably well within their range, which is the range of things that resolve cleanly and soon. They can tell you about an election. They are close to useless for the questions this book is about, because you cannot write a contract on whether the family will survive as an institution, and if you could, nobody would live long enough to collect.
The consequence of all this absence is the thing I actually want you to take from this chapter. It is not that forecasters are unreliable. It is that the unreliability cannot be corrected, because correction requires memory and there is no organ for it. Every field that has improved its accuracy — medicine, aviation, weather, structural engineering — improved by building an institution whose entire purpose was to remember failures in detail and refuse to let them be forgotten. Aviation has crash investigators. Medicine has, eventually and with resistance, trials and registries. Forecasting has anniversaries that nobody holds.
Which is why the predictions repeat. A field with no memory cannot notice that it is saying the same thing for the fifth time. That is the whole of the mystery of why the paperless office keeps arriving, and we will get to the paperless office, which is a comedy in five acts.
And it is why a single unrevised book on a desk is worth more than its quality deserves. It is a fragment of the missing archive.
Chapter Three — The Arrow and the Pendulum
On top of certain old steam engines there is a device that looks like a piece of jewelry: two heavy brass balls on hinged arms, spinning around a vertical shaft.
It is called a governor, and James Watt did not invent it — millers had been using something like it — but he attached it to the steam engine, and in doing so he built one of the first machines in history that could be trusted to look after itself. As the engine speeds up, the balls swing outward. Swinging outward pulls a linkage. The linkage closes the steam valve. The engine slows. The balls drop, the valve opens, the engine speeds up again.
Now imagine an observer who arrives, watches the engine for four seconds while it is accelerating, and leaves. He will report, accurately, that the engine is getting faster. He will project, reasonably, that it will continue to get faster. He will conclude, catastrophically, that it is going to tear itself apart. He has observed a real trend, drawn a valid line through it, and missed the entire point of the machine, which is that the acceleration is what triggers the thing that ends the acceleration.
That observer is the forecaster, and the whole of this book is about what he did not see.
The governor is worth dwelling on because it makes visible something that is almost always invisible. The restoring force is a physical object, made of brass, bolted on top, gleaming. In the systems that forecasters actually deal with, the equivalent object is a price, or a habit, or a status anxiety, or the mood of somebody’s mother-in-law, and it does not gleam and it is not bolted anywhere. It has to be inferred, and it is easiest to infer after it has already acted, which is to say too late.
Chemists have a phrase for the same behavior in their own field. If you take a system that has settled into a balance and lean on it — heat it, squeeze it, add more of something — the balance shifts in the direction that relieves what you did. Squeeze it and it takes up less room. Heat it and it absorbs the heat. The system does not resist you out of spite. Its response is simply composed of the same forces you disturbed, running in the direction that undoes the disturbance. Biology has its own version, which is why your body temperature is thirty-seven degrees whether you are in Cairo or Oslo, and why every attempt to lose weight by eating less encounters an organism that quietly turns down the thermostat in reply.
The pattern is so general that its absence is what needs explaining. Any arrangement that has persisted for a while has persisted because it has something in it that pulls back. Arrangements without a restoring force do not survive to become the subject of forecasts; they blow up early, unobserved, and are replaced by arrangements that have one. What we forecast about, almost by definition, is systems that have already demonstrated the capacity to return.
So why is the arrow so irresistible?
Partly because of when we look. A restoring force is weak while the disturbance is small — that is what makes it a restoring force rather than a wall — so the beginning of any movement genuinely does look like unopposed acceleration. The forecaster is not hallucinating. He is sampling during the only phase in which the arrow model is correct, and publishing before the phase ends. Every technology that ends up plateauing spends its early years on a curve that is indistinguishable from a curve that will never stop.
Partly because of who looks. The forecasters who dominate any given decade are people fluent in whatever is new, which usually means technologists, and the restoring forces are almost never technological. They are economic, social, institutional, and psychological. A person who can tell you precisely what a network will be capable of in eight years is frequently the last person in the room who will think to ask what a landlord will do about it. The expertise that lets you see the outbound swing is the same expertise that blinds you to the return.
And partly because the arrow is a better story. A pendulum has no plot. Nobody wants to hear that the thing will change substantially and then change partly back and settle somewhere awkward that resembles neither the old state nor the promised one. That is, unfortunately, what usually happens, and it is why accurate forecasts are so unsatisfying to read that they are seldom published.
Now for the variable that decides everything, and which almost no forecast mentions.
A restoring force does not act instantly. It acts late, and the length of the delay determines whether the system settles gently or thrashes. You already know this from a bad shower. You turn the tap toward hot and nothing happens, because the water in the pipe has not arrived yet, so you turn it further. Then the hot water arrives, all of it, and you leap backward and wrench the tap the other way, and thirty seconds later you are standing in cold water wondering why a species that has built telescopes cannot manage a bath. Nobody in that story is stupid. The whole disaster is produced by a delay of a few seconds between an action and its consequence.
Farmers have been living inside a larger version of this for centuries and gave it a name. Pork is expensive this year, so more piglets are raised. Pigs take time to grow. By the time they reach market there are too many, so the price collapses, so fewer are raised, so in due course there are too few and the price climbs again. Nobody is irrational at any point. Every individual decision is a sensible response to the price actually visible on the day it is made. The oscillation is manufactured entirely by the gap between the decision and its effect, and it will keep running as long as pigs take time to grow.
This is why the important question about any prediction is not merely whether a restoring force exists but how long it takes to arrive. Short delays give you a system that barely moves — you turn the tap, the temperature adjusts, nothing dramatic occurs, and no one writes a book about it. Long delays give you enormous excursions, and enormous excursions are exactly what get forecast as permanent transformations, because during the outbound phase they look like nothing else in living memory.
Housing is the standard modern demonstration. Demand rises, prices rise, and the response — building more of it — takes somewhere between five and fifteen years once you include the arguing. So the price signal that should be self-correcting instead produces a decade of apparently unstoppable ascent, during which it is entirely reasonable to publish a book explaining that the young will never own anything, followed by a correction that arrives at the worst possible moment for whoever bought the book. The mistake is not in observing the ascent. It is in mistaking a long lag for the absence of a mechanism.
Two clarifications, because I do not want to be accused of the opposite error.
First, a pendulum in the real world does not come back to where it started. It is damped, and it is being pushed, and it drifts. When the return swing arrives, the world it returns to has been permanently altered by the excursion. Offices did not empty out and then refill unchanged; the argument about who has to be in the room on Tuesday is now a permanent feature of employment, which it was not before. The claim is not that nothing happens. The claim is that the destination is almost never the endpoint of the line the forecaster drew.
Second, and more importantly: some things really are arrows. There are one-way changes in human affairs, and pretending otherwise is a fashionable stupidity of its own. Literacy has not swung back. A species that goes extinct does not return. Knowledge, once written down and widely copied, has never been comprehensively lost, whatever the romantics say about libraries. These are ratchets, and a ratchet is exactly a system in which the restoring force has been deliberately removed.
Which means the practical question is never whether the world runs in lines or in circles. It is narrower and much more answerable: for this particular claim, is there a mechanism that pulls back, and if so, at what amplitude does it engage? A forecaster who asks that question will still be wrong frequently. But he will be wrong in smaller and more interesting ways, and he will be spared the specific humiliation of having watched a machine accelerate for four seconds and concluded that it was about to explode.
Chapter Four — Three Verdicts, Not Two
For most of its history Scots law let a jury say three things rather than two.
Guilty. Not guilty. And, sitting awkwardly between them for centuries and periodically scheduled for abolition, not proven — a verdict which acquits the accused while declining to endorse his innocence. Lawyers have never entirely agreed on what it means. Juries have used it anyway, doggedly, generation after generation, because it corresponds to something that actually happens in the world and that the other two verdicts cannot express: the state of having heard the evidence and finding it insufficient in either direction.
Anyone grading old predictions needs that verdict, and hardly anyone uses it.
The convention is binary. A forecast either came true or it did not, and books about forecasting are accordingly organized into triumphs and pratfalls, which is convenient for the author and misleading for everyone else. The trouble is that a prediction is a claim about a process, and processes are frequently still running at the moment somebody decides to grade them. Declaring a verdict on an unfinished process is not rigor. It is the same impatience that produced the bad prediction in the first place, pointed in the opposite direction.
So here are the three columns I am going to use, and what each one costs to enter.
The first is the plain hit. The predicted thing arrived, in something like the predicted form, and can be pointed at. From the book on my desk: people would say things to machines that they would not say to human beings, because the machine cannot be embarrassed on your behalf and cannot despise you. That has happened on a scale nobody in 2007 could have costed. It is a hit, and I will spend a whole chapter on why it feels like less of one than it is.
The second is the reversal, which is stronger than a mere miss. A miss is when nothing happens. A reversal is when the world moves in the predicted direction, generates its own opposition, and ends up somewhere the forecaster explicitly ruled out. Schools were supposed to close and children were supposed to learn at home over the internet, more cheaply and more safely. In 2020 the entire planet ran that experiment involuntarily, which is a stroke of luck no grader deserves, and the result was not adoption but the fastest and most emphatic snap-back in the recent history of any institution. Parents who had never previously defended the school as an idea discovered within about six weeks that they were passionately in favor of it. That is not a failed prediction. That is a prediction that was tested and answered.
The third column is not proven, and it is the one that needs guarding, because it is also the standard refuge of every prophet who has ever been embarrassed.
The classic study of that refuge was done in the nineteen-fifties, when a small group of psychologists joined a group who believed the world would end on a specific date and that they alone would be collected by a flying saucer. The date arrived. Nothing happened. The psychologists were interested in what would follow, and what followed has been the template ever since. The group did not disband. Within hours a revised account had emerged in which the catastrophe had been averted by the group’s own faithfulness, and the members — who had previously been secretive — began, for the first time, actively recruiting.
The pattern generalizes far beyond doomsday cults, and gently enough that respectable people do it without noticing. A date passes. The claim is restated without the date. The failure is reinterpreted as a delay, and the delay is reinterpreted as evidence that the underlying insight was too far ahead of its time to be judged by ordinary schedules. Every one of those moves feels reasonable while you are making it. Together they constitute a machine for converting any refuted claim into a pending one, and a pending column with no admission requirements is worthless, because everything can get in.
So the third verdict comes with a fee, and the fee is a date.
To place a prediction in the not-proven column, I have to be able to say what would settle it and roughly when. If I cannot name a plausible outcome that would move it into the failure column by a stated year, then it is not pending; it is unfalsifiable, and unfalsifiable claims get filed separately, under decoration. Take the abolition of paper money, which the book on my desk treats as an imminent administrative convenience. That is genuinely unsettled. Cash has fallen dramatically as a share of transactions across most rich countries, and the value of banknotes actually in circulation has in several of them gone up rather than down, which is exactly the sort of double movement this book exists to explain. So: pending — and pending until, let us say, 2040, by which point a major economy will either have retired physical currency or it will not, and the claim can be closed.
There is a fourth judgment that keeps arising and does not deserve a column of its own, so I will name it here and use it as needed: right thing, wrong size. The forecaster identifies a real phenomenon and then multiplies it by ten. Remote work is the obvious case. Everything the author said about the mechanics of distributed knowledge work was correct, and he had run it himself; what he got wrong was the fraction of humanity to whom it applies, which is not most people and never was, because most jobs involve an object or a body that has to be physically present. Partial credit is the honest grade, and refusing to award it is how you end up with the two most popular and least useful positions available: that the old forecasters saw everything, or that they saw nothing.
Chapter Five — The Man Who Wrote Down Why He Would Be Wrong
On the second page of his foreword, before any prediction has been made, the author states that predicting the future by extending current tendencies almost always fails.
He goes further, and proposes it as something like a law, which he names: the inadequacy of futurological extrapolation. The rule he states is that the tendencies which turn out to dominate the future are typically ones that are absent or unnoticed in the present. He notes that this is exactly how evolution behaves, with the magnificently successful lineages dying out and the marginal ones inheriting the earth, and that paleontologists and futurists are consequently in the habit of asking the same question afterward, which is who could possibly have foreseen this.
He then announces that he will therefore not be making predictions at all. The book, he explains, will discuss goals and the strategies for reaching them rather than forecasting events, since forecasts on the distant future are futile and mostly forgotten.
Twenty-two chapters follow. They contain, by my count, several dozen distinct claims about what will happen, many of them arrived at by extending a current tendency in a straight line. The cities will empty. The schools will close. Paper money will be abolished. The body will be optional. The nation-state will dissolve. The erotic will not return.
I have thought about this for some time and I want to be precise about what it demonstrates, because the cheap reading — that the man was a hypocrite — is both unkind and wrong. He was not concealing anything. The refutation and the offense sit fifteen pages apart in the same volume, available to any reader, unedited. Nobody hides a thing by printing it.
What it demonstrates is that knowing a rule is not the same as being bound by one, and that the gap between those two states is where most human failure lives.
Psychologists have a tidy name for one part of this. Ask people whether their judgment is distorted by the usual well-documented biases and they will readily agree that human judgment in general is; ask whether their own judgment is, right now, in this instance, and they will report that it seems fine. The effect survives even when the person being asked has just been taught the bias in question. In other words, understanding a systematic error appears to be stored somewhere that does not communicate with the part of the mind that is currently committing it. The knowledge is real, and it is in the wrong room.
Medicine offers the least comfortable demonstration, and the cleanest. That washing your hands between patients prevents infection is not contested by any physician anywhere. There is no faction. There is no debate. And observed compliance with hand hygiene in hospitals has, for decades, run at levels that would be considered a scandal in any other safety-critical trade, which is why the corrective measures deployed have not been educational at all. They have been architectural: dispensers mounted where the hand goes anyway, audits, someone whose job is to watch, and in some places a light that changes color. Nobody tried to make doctors believe in germs. They already did. What was missing was not belief but a mechanism.
Surgery went the same way and arrived at the same answer. The intervention that made the biggest measurable difference to operating-room deaths in the last twenty years was not a drug or a device but a printed list of things to confirm out loud before cutting, which works precisely because it does not rely on anyone’s intention to remember.
That is the distinction I want to install now, because it will be doing heavy lifting for the rest of this book. There are rules a person declares, and there are rules that actually bind. A declared rule is a sentence. A binding rule is a sentence plus something that happens if you break it. The author of the book on my desk had a declared rule of the highest quality — genuinely one of the better statements of the problem I have read anywhere, and it was published two years before the most cited academic study of forecasting failure — and precisely nothing attached to it. No date, no wager, no named disconfirming event, no colleague appointed to object, no cost of any kind for proceeding as though the rule did not exist. Under those conditions the rule cannot do work. It functions as an ornament placed at the entrance of the building, which is a fine location for an ornament and useless as a lock.
It is worth asking what would have bound him, since the answer applies to everyone reading this. A date would have. A stated year converts a mood into an obligation and cannot be quietly renegotiated later. A wager would have, for the same reason and with interest. A named disconfirming event would have — the discipline of writing down, at the moment of prediction, what observation would make you abandon it, which almost nobody does because it feels like sabotaging your own argument, which is exactly what it is and exactly why it works.
And an opponent would have. Not a critic afterward, when the book is printed and the author’s identity is fused to its contents, but somebody sitting across the table beforehand whose job is to construct the strongest case against each claim before it is committed. Institutions that take error seriously build that role in deliberately, because they have learned that a person cannot reliably play it against himself.
Which is, I suppose, the honest description of what I am doing at this desk, eighteen years too late, with no authority whatsoever except the accident of having been born afterward.
Chapter Six — Furniture: The Predictions You Live Inside
Look around whatever room you are sitting in and count the prophecies.
The rectangle in your pocket that contains, in a practical sense, every book ever written, every map, and the ability to speak to any person on the planet — that was promised for well over a century and is currently being used, in all probability, to look at a photograph of somebody’s lunch. The device on the wall that keeps the room at a fixed temperature regardless of the weather outside was a marvel within living memory. The bank card is the visible remnant of a decades-old campaign to abolish coins. If you have spoken to a relative’s face on a screen this month, you have used the single most predicted technology in the history of prediction, and it is unlikely you thought anything about it at all.
That last one has a particularly humiliating history. The idea of the visual telephone was in circulation before the telephone itself had finished arriving; cartoonists were drawing it in the eighteen-seventies. In 1964 it was demonstrated as a working commercial product at the New York World’s Fair, complete with booths and a tariff, and it was heavily promoted for years afterward as the coming thing. It failed comprehensively. It was expensive, it was awkward, and it turned out that a great many people making telephone calls did not want to be looked at, which is a piece of information about human beings that no amount of engineering could have supplied.
And then, decades later, having been declared a definitive example of futurist overreach, it arrived anyway — free, universal, and attached to nobody’s business plan. It is now so thoroughly ordinary that its arrival was never announced. There was no morning on which the world observed that the videophone had finally come. It simply seeped in, and by the time anyone might have held a ceremony it would have seemed eccentric to bother.
This is the career path of a successful prediction, and it explains something about our collective sense of the past that I think is genuinely distorting. The predictions we remember are the ones that failed, because failure retains its shape. A flying car remains permanently a flying car, forever absent, forever quotable, useful as a joke for as long as anyone lives. A fulfilled prediction, by contrast, dissolves. It stops being a claim about the future and becomes a description of the present, and descriptions of the present are the least memorable sentences a language can produce.
The dissolution is helped along by a habit we have of retroactively downgrading whatever arrives. There is a rueful principle in computing, usually credited to Larry Tesler, that artificial intelligence is defined as whatever machines cannot do yet. Playing chess at grandmaster level was, for forty years, the standard illustration of genuine machine intelligence; the day after it happened it was reclassified as brute calculation and dropped from the syllabus. Recognizing speech, translating text, holding a conversation, diagnosing from an image — each was a landmark until it worked, at which point it became a feature. The goalposts are not moved dishonestly. They are moved because the mystery is what we were actually pointing at, and the mystery does not survive the solution.
Which brings me back to the one clean, uncelebrated hit in the book on my desk.
In the chapter on virtual psychology, the author argues that a person will disclose things to a machine that they would withhold from a human therapist, and he gives the reason plainly: with a human being you are always aware of a person who could judge you, could be disgusted, could remember. The machine has no moral commitments and no capacity for contempt, and this makes it, for certain purposes, better. He predicts that daily conversation with such a program could become as ordinary as personal hygiene.
He was not the first to notice the effect. The man who built the original conversational program in the nineteen-sixties was appalled by it — he watched people who knew perfectly well that they were addressing a few pages of pattern-matching become absorbed, confide, and in one much-repeated instance ask him to leave the room so that the conversation could continue in private. He spent much of the rest of his life arguing that this was a warning rather than an opportunity.
What the book on my desk did was to look at the same phenomenon and decline to be appalled. It treated the willingness to confide in a machine as a genuine feature of human beings rather than a pathology, and forecast that we would build on it. Eighteen years later that is one of the largest things happening to people, and no one is going to trace the credit, because nobody traces credit for furniture.
I would like to leave you with the practical consequence rather than the melancholy one, because there is a use for all this beyond fairness to the dead.
If successful predictions become invisible, then the invisibility is a signal, and you can run it backward. When you find yourself using something without noticing it, ask when it was first promised and by whom, and you will frequently discover a long file of people who were laughed at, followed by a short period in which the thing quietly worked, followed by nothing. That file is the real record of what forecasting can do, and it is systematically missing from our estimate of forecasting’s reliability. We have preserved the comedy and thrown out the evidence.
Chapter Seven — The Comedy of the Flying Car
In a hangar in Washington State in the late nineteen-forties, a man named Moulton Taylor built a small aircraft whose wings and tail could be detached, folded, and towed behind it on the road.
He called it the Aerocar. It flew. It drove. It was granted certification by the American aviation authorities in the mid-fifties, which is a genuinely difficult thing to obtain and which almost no subsequent project of the kind has managed. A manufacturer expressed interest on the condition that a few hundred orders be secured in advance. The orders did not come. Six were built. One of the six is in a museum, which is where the flying car has lived ever since — not in the future, but in a display case, being photographed by people who use it as evidence that nobody can predict anything.
It is worth being precise about why the flying car became the standard joke, because the standard joke is wrong about its own subject.
The usual account is that the technology proved too hard. This is false. The technology was demonstrably achieved by one man with a small workshop before the Suez Crisis. What defeated it was that a road vehicle and an aircraft are optimized against incompatible constraints, and any object that satisfies both is worse at each than a specialist would be. An aircraft wants to be light and wants long fragile surfaces; a car wants to survive being reversed into a bollard by somebody holding a coffee. Combine them and you get a mediocre car that is also a mediocre plane, at the price of two.
But even that is not the real obstacle, and the real obstacle is the one this book keeps running into: the constraint was never in the machine. It was in the operator. A car can be entrusted to a distracted seventeen-year-old because the failure mode of inattention is a scrape. The failure mode of inattention at four hundred meters is a funeral, and it is somebody else’s funeral if you come down on a house. Which means the flying car does not require an engineering advance so much as a licensing regime, an air traffic system for millions of simultaneous amateur pilots, an insurance market willing to price the whole arrangement, and a public prepared to accept the resulting death rate. None of these things is technological, all of them are expensive, and none of them was in anybody’s picture in 1949, because the people doing the picturing were engineers and the missing pieces were not engineering.
Now here is the part that the joke conceals, and it is the reason I am starting this stretch of the book with a failure rather than a success.
The flying car was never actually about flight. It was about a bundle of things people wanted: to leave when they liked, go where they liked, avoid the traffic, and not maintain the vehicle for the ninety-six percent of the time it sat idle. Every single item in that bundle has been delivered. Not by aircraft, but by a navigation system that tells any driver in the world the fastest route in real time, by summoning a vehicle with somebody else’s maintenance schedule, and by removing entire categories of journey through the crude expedient of not having to go. The function arrived on schedule and in a completely unexpected body.
We do not credit this, because forecasts are graded on form and delivered in function. The prophecy specified an object; the world supplied a capability; and because the object never materialized, the whole episode is filed under comedy. This is a grading error, and it is not a small one — it is the same error, running in the opposite direction, as the one that makes fulfilled predictions invisible.
The jetpack, its companion in the museum of ridicule, is a different case and it is worth separating them, because the distinction organizes everything that follows in this book. A rocket belt was successfully demonstrated in the early nineteen-sixties and appeared in film and at public spectacles. It flew for approximately twenty seconds. It flies for approximately twenty seconds today, and it will fly for approximately twenty seconds in a hundred years, because the limitation is the amount of energy that can be stored in a quantity of fuel a person can carry on their back, and no cleverness relaxes that. The jetpack is not waiting for an entrepreneur. It is waiting for different chemistry.
So there are at least three distinct ways for a prediction of this kind to die, and confusing them is how people conclude that forecasting is hopeless.
A prediction can fail against physics, in which case it will keep failing forever and no amount of investment will help. It can fail against economics, in which case it is not dead but dormant, and will revive precisely when some cost falls below some threshold — which is why things that were absurd for forty years become ubiquitous in four. And it can fail against human beings, which is the most common and the least anticipated, because the forecaster is generally an enthusiast and enthusiasts systematically misjudge how much other people want the thing. The people who designed the visual telephone did not imagine anyone would object to being seen. The people who designed the flying car did not imagine anyone would object to their neighbor having one.
Sort any old forecast into those three bins before laughing at it and something uncomfortable happens: the physics bin turns out to be nearly empty. Most of the celebrated failures of futurism were not impossible. They were merely unwanted, or unaffordable at the time, or blocked by an institution nobody had thought to include in the drawing.
Chapter Eight — Scarcity Summons Its Own Cure
For a few decades in the middle of the nineteenth century, the most strategically important substance on earth was bird droppings.
On a scatter of rainless islands off the coast of Peru, seabirds had been depositing guano for a very long time, and the accumulated deposits turned out to be an extraordinary fertilizer at a moment when European and American soils were exhausted and populations were rising. The trade became enormous. It made and financed a national government. Laborers, many of them indentured Chinese, were worked to death digging it. In 1856 the United States passed an act allowing its citizens to claim, on behalf of the nation, any unoccupied island anywhere in the world that happened to have guano on it, which remains one of the more remarkable pieces of legislation ever enacted by a serious country.
And then the deposits began to run out, exactly as everyone could see they would, and the anxiety was entirely rational. Agriculture depends on fixed nitrogen. Nitrogen is four-fifths of the air and completely useless in that form to plants. The known supplies were a finite pile of ancient droppings and some Chilean mineral deposits, both being consumed at speed. A competent forecaster in 1890, counting the pile and dividing by the rate of use, would have projected a crisis in food supply, and he would have been counting correctly.
In the first decade of the twentieth century two German chemists worked out how to pull nitrogen out of the air using pressure, heat and a catalyst, and by the outbreak of the First World War the process was operating industrially. The pile stopped mattering. A substantial fraction of the nitrogen in the body of every reader of this book arrived there by that route, which is to say that most of us are alive because a resource crisis was answered by making the resource out of the sky.
This is the mechanism that has wrecked more forecasts than any other, so I want to state it carefully rather than triumphantly, because it is regularly deployed as a piece of cheerfulness and it is not one.
When something becomes scarce its price rises, and a rising price is not merely a number. It is a signal broadcast simultaneously to every person on earth who might have any use for the thing, and it says: whatever you were doing that involves this, doing less of it is now worth money, and finding an alternative is now worth money, and going to look for more of it in places that used to be too expensive to search is now worth money, and recovering it from the rubbish is now worth money. Four separate industries are funded by one price rise, and they operate in parallel, and none of them existed the day before.
Now consider the position of the forecaster confronting this. He can count the stock. The stock is a known quantity, published in tables, satisfyingly precise. He cannot count the substitutes, because the substitutes are not yet invented, and there is no register of things that do not exist. So his ledger has a fully itemized column on one side and an unavoidably blank column on the other, and the blank column is the one that historically wins. It is not that he is pessimistic by temperament. It is that his method has a hole in it shaped exactly like human ingenuity, and no honest amount of care will close it.
A related and less obvious error concerns the word reserves, which is responsible for a great deal of confusion in public argument. Reserves are not the amount of a substance in the ground. They are the amount currently identified and worth extracting at present prices with present technology, which means reserves are an economic inventory wearing a geological costume. This is why the reserves of almost everything have a maddening habit of being larger after decades of extraction than they were before. Nobody is manufacturing the metal. The category is expanding, because when the price rises the boundary of what counts as a reserve moves outward, and deposits that were merely rock become resources.
Here, though, is the boundary of the whole argument, and I would rather draw it myself than have it drawn for me.
The mechanism runs on price signals, and it therefore works only where a price exists and reaches the person making the decision. It works beautifully for tin and copper and nitrogen, which are owned, traded, and priced. It does not work at all for anything held in common and used for free. Nobody is bidding up the price of the atmosphere. No signal reaches a fishing fleet telling it that the last spawning stock is worth more intact than landed, because the fleet does not own the fish and the boat that shows restraint is simply leaving them for a competitor. An aquifer under four countries has no price. Topsoil has no price until it has already gone.
So the history of resource panics is not the reassuring story that things always work out. It is a much more specific finding: things work out where there is an owner and a price, and they do not work out where there is neither. The nineteenth century solved fertilizer and did not solve the Grand Banks. Every future panic can be sorted by that criterion on the day it arrives, and the sorting will tell you more than the tonnages will.
Chapter Nine — Malthus Never Left the Building
The most consequential book about the future ever published in English appeared anonymously in 1798 and its author spent the rest of his life revising it because he was not satisfied.
Thomas Malthus argued that population, left alone, multiplies — each generation producing more than replaced it, compounding — while the food supply can be increased only by adding fields and effort, which accumulates rather than compounds. Two quantities growing in two different ways must eventually diverge, and when they do, the correction is not administrative. It is famine, disease, and war, arriving not as misfortunes but as the accounting.
It is important to understand how good this argument was. It was not a mood. It was a mechanism, stated cleanly, with an inescapable conclusion, and it was supported by every scrap of evidence available to anyone in 1798. For the whole of recorded history up to the moment Malthus sat down, human populations had risen in good times and been cut back in bad ones, and nobody anywhere had achieved a durable general improvement in the condition of ordinary people. Darwin read Malthus and found in him the engine of natural selection, which is not the tribute paid to sloppy thinking.
He was also more careful than his reputation allows. Later editions added the possibility that people might restrain themselves — marrying later, having fewer children by choice — and he treated this as a genuine escape route rather than a formality. The caricature of Malthus as a man who simply announced doom is unfair, and it is unfair in the specific way this book keeps identifying: the qualification was in the original and did not survive transmission.
What defeated him was not that he reasoned badly. It was that he was writing at the precise hinge of the only period in human history when both of his premises stopped holding at once.
Food output did not accumulate; it compounded, through crop rotation and then mechanization and then synthesized fertilizer and then plant breeding, and it has outrun population every decade since he died. That is the half of the story everyone knows. The other half is stranger and far more important for this book, because it is a reversal rather than a rescue.
Malthus assumed that prosperity increases births. Every observation he had said so: give people more food and they marry earlier and have more children. And then, beginning in France earlier than anywhere else and spreading unevenly across the world over the following century and a half, the relationship inverted. As countries became wealthier, better educated, more urban, and as it stopped being ordinary for children to die, people began having fewer of them. Not because they were exhorted to. Not through any policy that reliably worked. They simply did, in country after country, in wildly different cultures, under every conceivable political system, with a consistency that remains one of the most robust findings in the social sciences and one of the least explained.
Look at what that means structurally. Malthus identified a real force pushing in one direction and did not know that prosperity itself would generate an opposing force. The correction did not come from outside the system as a punishment. It was manufactured by the very success that was supposed to be impossible. That is the machine at the center of this book, operating at the largest scale available, and it took about a hundred and fifty years to become visible, which is why nobody saw it coming.
And the swing did not stop at the middle. Consider where the anxiety now sits. Within the lifetime of people currently drawing pensions, the respectable emergency has moved from too many people to too few — governments paying families to have children and failing, workforces contracting, pension arithmetic that does not close, and a genuine and unresolved argument about whether any society has found a way to reverse the fall once it starts. The same institutions that funded population control programs now fund natalist ones. The energy is identical. The sign is opposite. One human lifetime.
I do not offer that as a reason to relax, and I want to be careful here, because the Malthus story is routinely used as a general licence to ignore any warning about limits. It licenses nothing of the kind. Malthus was wrong about the world of the following two centuries; he was an accurate description of the ten thousand years before him, and there is no law of nature guaranteeing that the exemption is permanent. Regional famines have continued to happen, and where they have happened the cause has usually been political rather than agronomic, which is its own uncomfortable finding.
What the episode actually teaches is narrower and more useful. When you meet a forecast built on two trends diverging, the question to ask is not whether the arithmetic is right — it usually is — but whether either trend is capable of altering the other. Malthus treated population and food as independent quantities that merely collided. They were not independent. Each was quietly reshaping the other the whole time, and the reshaping is where the future actually happened.
Chapter Ten — The Metals That Refused to Run Out
There is a table in a book published in 1972 that has done more damage to the reputation of computer modeling than any other page ever printed.
The book was commissioned by an international group of businessmen and scientists and produced by a team at an American institute of technology, and it used one of the first world-scale computer models to examine what would happen if industrial growth continued. The table listed commodities alongside the number of years each would last. Gold had until roughly 1981. Mercury, the middle of that decade. Tin by 1987, zinc by 1990, petroleum by 1992, natural gas by 1993.
None of it happened. Gold is still mined. Tin is on the market. The world has been extracting petroleum for three decades past its stated expiry and the identified reserves are larger than they were when the table was printed. The forecast is now a standing joke, deployed whenever anyone wishes to suggest that modeling is a superstition with a keyboard.
And here I have to do something that spoils the joke, because this book has no business enjoying a cheap victory.
The authors did not intend that table as a set of predictions, and they have spent fifty years saying so. It was an illustration of a simple and deliberately crude calculation — take the known reserves, divide by current consumption, observe how quickly a finite number disappears when the denominator is growing. It was a demonstration of why static thinking about exponential use is alarming, presented as a stepping stone toward the actual model, which was concerned with the interaction of population, industry, food, pollution and resources over two centuries and which did not claim that tin would vanish in 1987. Independent comparisons of the model’s central scenario against subsequent real-world data have found it holding up considerably better than the table’s reputation suggests, which is not a fact that appears in the joke.
So the honest verdict is awkward and I am going to state it in full. The illustrative table was crude and was never a forecast. The public and a fair number of governments received it as a forecast anyway, acted on it, and abandoned the entire enterprise when it failed. Both of those things are true, and the second one is not the authors’ fault in any moral sense — and it is entirely their problem in every practical sense, because a warning that cannot survive transmission has not been issued.
That is a lesson about forecasting that has nothing to do with accuracy, and I think it may be the most important one in the book. A prediction is not a statement. It is an object that will be handled by journalists, politicians, opponents and eventually schoolchildren, and it will be reduced at every stage. Confidence intervals do not survive a headline. Qualifications do not survive a summary. Illustrative examples do not survive at all, because an example is concrete and memorable and the surrounding caution is neither. If you put a specific year in a table, that year is what will exist in twenty years’ time, whatever the caption said.
The most instructive sequel arrived in 1980, when an economist and a biologist made an actual bet — a rare event, as we have seen. The biologist held that resource scarcity would drive prices up; the economist held that human response would drive them down. Five metals were chosen by the biologist’s side, a thousand dollars was staked, and the settlement date was set ten years out. All five metals fell in price. The economist won and the episode entered folklore as a decisive verdict.
It was nothing of the kind, and the reason is worth more than the anecdote. Later analyses have shown that the outcome depended heavily on the decade chosen. Run the same wager over most other ten-year windows of the twentieth century and the result flips more often than not. What was settled in 1990 was one particular decade, in which a global downturn suppressed demand and mining capacity happened to be ample. A single bet with a single ten-year window is a coin toss dressed as an experiment, and both men, being clever, knew it, and it made no difference to how the result was used.
What I take from the metals is not that resources are infinite. They obviously are not. It is that a stock divided by a rate is not a prediction, because neither the stock nor the rate is a fixed quantity — the stock expands as price makes marginal deposits worth digging, and the rate falls as price makes waste worth avoiding. The calculation treats as constants the two things most certain to move, and it does so because they are the two things easiest to look up.
Chapter Eleven — Peak Everything
In 1956 a geophysicist working for an oil company stood up in San Antonio and told an audience of petroleum engineers that American oil production would peak between 1965 and 1970 and decline thereafter.
His employer, according to the account he gave afterward, tried to get him to soften it. The industry regarded the claim as absurd; the country was awash. M. King Hubbert’s reasoning was not economic and barely mentioned prices. He observed that extraction from any finite field follows a roughly symmetrical rise and fall — you find the big easy deposits first, production climbs, then it becomes progressively harder — and that summing enough fields gives a single curve for a whole country.
American production peaked in 1970.
I want to sit with that for a moment, because this book has a duty to record successes as loudly as failures, and this one is exceptionally clean. Hubbert made a specific, dated, falsifiable, deeply unpopular prediction about a fourteen-year horizon and was vindicated almost to the year. There are not many of those. He deserves the credit and he is generally denied it, because of what his followers did next.
What they did next was extend the method to the planet, and then, crucially, to time without end. The world would peak around the turn of the century or shortly after; thereafter production would fall permanently; and since the modern economy runs on liquid fuel, the decline would be accompanied by the collapse of industrial civilization, an outcome discussed in a substantial literature in the early two-thousands with dates attached and, in some quarters, with recommendations about where to buy land.
From about 2008 onward the United States raised its production so sharply, by getting oil out of dense rock that had previously been considered uneconomic, that it passed its 1970 peak and went on to produce more than any country ever had. The curve was not a law of geology. It was a law of geology multiplied by a price, and when the price stayed high for long enough, techniques that had existed for decades but had never been worth the trouble became worth the trouble.
So the doctrine failed in exactly the way this book keeps describing: the outbound trend was real, the mechanism that would answer it was invisible because it lived in a different discipline, and the forecast mistook a long delay for the absence of a response. But notice what Hubbert himself got right and his successors lost. He was making a claim about a specific set of fields under specific conditions over a stated interval. They were making a claim about the species.
The peak habit is much older than oil and it recurs with a regularity that ought to be a warning in itself. A distinguished economist published a book in the eighteen-sixties arguing that Britain’s coal would become progressively more expensive to extract and that national supremacy would decline with it, which caused a parliamentary commission and a great deal of anxiety and did not happen — Britain’s coal problem, when it eventually arrived a century later, was that nobody wanted the coal. Since then we have had peak helium, peak phosphorus, peak sand, peak lithium, peak water, peak fish, and, in the more excitable stretches of the last two decades, peak everything simultaneously, as a book title.
Some of these are serious. Phosphorus genuinely has no substitute — you cannot synthesize an element — and the deposits are concentrated in politically awkward places. Groundwater in several major agricultural regions is being drawn down faster than it recharges, and an aquifer, as I have said, has no price and therefore no self-correcting mechanism. Fisheries have collapsed and stayed collapsed. The correct response to the failure of peak oil is emphatically not the conclusion that limits are a myth. It is the narrower conclusion that a peak in supply is a statement about cost, and cost responds to human beings.
Which sets up the inversion that has quietly happened while nobody adjusted their vocabulary. The serious conversation about oil is no longer about when we run out. It is about when demand stops growing — whether the world stops wanting the stuff before the world stops having it. The anxiety has flipped from geology to appetite within a single generation, and the same institutions that once modeled scarcity now model surplus. That is a pendulum, observed in real time, in a field where everyone involved is numerate and well funded.
If I had to compress the whole of this chapter into an instruction, it would be this: when somebody shows you a peak, ask whether they are describing a physical limit or a price. A physical limit is a wall. A price is a door with a fee on it, and the fee changes.
Chapter Twelve — The Paperless Office Enters Its Fiftieth Year
In 1975 an American business magazine ran a feature on the office of the future in which an executive explained that paper was on the way out.
It was a reasonable thing to say. Documents were about to become electronic; electronic documents do not need to be printed; therefore, less paper. Every step of that argument is sound and the conclusion was wrong for approximately the next thirty years, during which time the consumption of office paper in the developed world rose steadily, and rose fastest in exactly those organizations that had computerized most thoroughly.
The mechanism is almost funny once you see it. A document that exists only in one filing cabinet is difficult to produce and therefore rare. A document that can be generated in a moment, revised eleven times, circulated to forty people and printed by each of them for reading on the train is not rare at all. Computers did not remove the reason for paper; they removed the friction from the production of things that might get printed, and then the laser printer removed the friction from printing them. Two convenience improvements in series produced a mountain.
Researchers who studied this properly at the turn of the century found the effect at close range: introduce electronic mail into an organization and paper consumption in that organization goes up substantially, because messages that would never have been written are now written, and a portion of them get printed. The people doing the printing were not being irrational or nostalgic. Paper has properties that screens of that era did not — you can spread six pages out at once, annotate them, hand one to somebody, and read them in a meeting without appearing to be doing something else.
Now, the part that almost nobody says out loud, because by the time it happened the joke had become load-bearing.
The paperless office arrived. Global demand for printing and writing paper turned over in the first decade of this century and has been falling ever since, in some markets steeply. Offices that used to run on printed memoranda now do not. The filing room has become a meeting room. Anyone under twenty-five entering a professional job today will handle a small fraction of the paper their predecessor handled in 1995, and the trend is not reversing.
So the 1975 prediction was correct. It was correct about the direction, correct about the cause, and correct about the eventual outcome. It was merely thirty-five years early, and the thirty-five years contained a period of the opposite behavior severe enough that the prediction became a synonym for foolishness. By the time it came true, the phrase paperless office could not be used in a sentence without irony, and the people who were finally living in one had inherited the joke without the claim.
I find this the single most useful case in the book, because it isolates the variable that matters and that almost no forecast supplies. Direction is the easy part. Direction can often be reasoned out from first principles by anyone paying attention, which is why so many old predictions read as eerily prescient once you delete the dates. Timing is the hard part, and timing is the entire practical content of a forecast. A prediction that is right about direction and wrong about timing by three decades is not partially useful. For anybody who has to act on it — invest, train, build, choose a career — it is worse than no prediction at all, because it produces confident action at precisely the wrong moment.
What sets the timing is almost never the technology’s capability. It is the replacement of a whole surrounding arrangement: the screen good enough to read on for eight hours, the second screen so you can have two documents at once, the tablet you can hand to somebody, the search that makes a digital pile more navigable than a physical one, the signature that is legally valid without ink, and the retirement of the generation that had a filing system in its head. None of those is the original insight. All of them had to arrive before the original insight could be redeemed, and each had its own schedule, and the schedule of the slowest one governs everything.
Which suggests a question to ask of any forecast you meet, and I offer it as the practical residue of this chapter. Not is this plausible — most of them are plausible, that is the trouble — but what is the slowest thing that has to change first, and who is currently in charge of it? The answer is usually a person, an institution, or a habit, and it is usually not in the room where the forecast was made.
Chapter Thirteen — The City That Would Not Die
The city has now been killed by the telegraph, the telephone, the automobile, the television, the fax machine, the internet, and a virus, and it keeps turning up for work.
Each obituary was written in good faith and each had a real mechanism behind it. If a message can travel instantly, why must the merchants stand in the same square? If a man can drive, why live where he works? If the office can be assembled out of screens, why maintain the expensive building, and why pay to sleep near it? By 1980 the argument had acquired a name and a cottage — the electronic cottage, in which the family worked from home and the industrial-era commute became a curiosity. By the late nineties it had a slogan about the death of distance. And in 2007 it produced the version in the book on my desk, which is the most concrete of the lot: move the population out of the expensive cities onto empty land, build cheap insulated houses with their own power, pay people a decent allowance, teach the children over the internet, and let the whole apparatus of commuting and rent and traffic simply stop.
In 2020 the experiment was run on everybody at once, without consent, which is a piece of evidence no forecaster is normally granted. For a period of roughly eighteen months the cities did empty. Rents in the expensive centers fell for the first time in a generation. Commentary was written declaring the matter settled.
By 2023 the rents in most of those centers were at record highs.
The mechanism people miss is that the city was never primarily a place to be near a desk. A desk is the one thing in a city that genuinely can be moved. What cannot be moved is the reason the desks were gathered in the first place, which is matching — the ability of a person with an unusual skill to find the small number of employers who need exactly that, and of an employer with an unusual requirement to find the handful of people who can meet it. A thin market forces everyone toward the average job and the average worker. A thick one lets both sides specialize, and specialization is where the money is, which is why wages in dense places are persistently higher for the same work and why people keep paying absurd rents for the privilege of being confused about why they live there.
Matching is not only economic, and this is the part the 2007 forecast could not see, because it was written by a settled man with a family. Cities are also where people find friends and partners. A person of twenty-two does not move to an expensive city for the career alone; they move because that is where the other twenty-two-year-olds are, and no amount of bandwidth substitutes for the possibility of meeting somebody you were not looking for. The screen is excellent at maintaining relationships that already exist and nearly useless at generating new ones, which is precisely backward from what the forecast required.
There is a beautiful piece of evidence for the durability of all this, and it concerns time rather than space. If you look at how long people have spent travelling to and from their daily business — in medieval walking towns, in nineteenth-century streetcar suburbs, in modern car cities, across wildly different countries and centuries — the average comes out at roughly an hour a day, over and over, with remarkable stubbornness. Faster transport has never bought people shorter journeys. It has bought them longer ones at the same cost in time, which is how the city expands rather than dissolving. The budget is not measured in miles. It is measured in the amount of daily life a human being is willing to spend in transit, and that number appears to be a property of the species rather than of the technology.
Which means the correct prediction, had anyone made it, would have been neither death nor survival but redistribution. Offices did loosen. Attendance three days a week is now ordinary where five was compulsory. Some smaller cities within reach of larger ones grew sharply, because a commute you make twice a week can be twice as long. Downtown districts built entirely around lunchtime sandwiches for office workers have suffered badly and some will not recover. The city did not die; a particular internal arrangement of it did, and the pieces are still being reallocated.
Partial credit, then, and I want to award it precisely. The book on my desk was right that the technical requirement for distributed work would be met, and right about the direction of travel, and right that it would be resisted mostly by habit. It was wrong in identifying the city with the office. It treated the metropolis as a container for jobs, when a metropolis is a machine for encounters, and the encounters were never the thing that could be delivered down a wire.
Chapter Fourteen — The Barrier That Held the Thing Together
Of all the predictions in the book on my desk, the one that has been most completely fulfilled at the technical level and most completely refuted at the human level concerns translation.
The argument runs as follows. Cultural barriers are dissolving, because if a custom or a phrase is unclear to you it can be explained in a fraction of a second. Language barriers no longer exist, because websites with automatic translators allow anyone to converse freely. Therefore the concept of national unity will become obsolete, and people who cannot tell whether their correspondent is African or Anglo-Saxon will find racism difficult to sustain.
The first two clauses came true beyond anything the author imagined. Translation between the major languages is now free, instant, available on a telephone, and good enough that people conduct business and romance through it. A traveller can point a camera at a menu. The barrier did not merely lower; for most practical purposes it fell over.
And the conclusion has failed so thoroughly that it is difficult to discuss politely. The two decades in which the language barrier disappeared were the two decades in which national and ethnic identity became more salient in public life across most of the world, not less. Nobody with a functioning newspaper subscription believes otherwise.
What went wrong is a mistake about what a barrier does, and it is one of the most useful errors in this book, because it generalizes to a dozen other cases.
The forecast assumed that separation causes difference. Remove the wall, contact occurs, contact produces familiarity, familiarity produces merger. But separation frequently maintains difference the other way around: by preventing the friction that makes people conscious of what they are. Two villages that never meet have no opinion of each other. Two villages that meet daily develop one, and it is not automatically a good one, because now there is something at stake — a job, a road, a school place, an argument about whose grandfather was there first.
Social psychologists have known the shape of this since the middle of the last century. Contact between groups does reduce hostility, reliably — under conditions. The groups need roughly equal standing in the situation, a shared goal, a reason to cooperate rather than compete, and some authority that supports the arrangement. Take those conditions away and contact does not reduce hostility at all. It can sharpen it. The internet, considered as a contact machine, supplies volume and removes every one of the four conditions, since it is precisely a place with no equal standing, no shared goal, no requirement to cooperate, and no supporting authority.
History had already run this experiment on a grand scale with the previous great barrier-lowering technology. When printing arrived, informed people expected Christendom to become more unified, since everyone could now read the same texts. What actually followed was schism, more than a century of religious warfare, and — this is the piece that matters most — the manufacture of the modern nations themselves. Printers had to choose a dialect to print in, and the chosen dialect became a language, and the people who read it became conscious of themselves as a public with a name. Print did not dissolve the tribes. It created larger and much more dangerous ones, by giving millions of strangers a way to imagine that they belonged together.
The Victorians made the identical prediction about the telegraph and said it with the same confidence: instantaneous communication between nations would render war impossible, since misunderstanding was war’s true cause. They then produced the bloodiest century on record, using the telegraph to coordinate it.
None of this means the barrier should have stayed up, and I want to be careful, because this argument is easily borrowed by people I would not care to be borrowed by. A great deal of what fell when translation arrived has stayed down and stayed good. Scientific collaboration across languages is a straightforward gain. A family scattered across four countries can now hear each other’s voices, which is a thing my great-grandparents’ generation simply lost. Small literatures reach readers they could never have reached.
The distinction that survives the evidence is this. Lowering a barrier reliably improves relations between people who wanted to reach each other and were prevented. It does nothing good, and often something bad, between people who did not want to meet and are now unable to avoid it. The forecast in the book on my desk assumed the whole world consisted of the first kind. Everyone who has ever built a communication technology has assumed this, because they are, by temperament, the first kind.
Chapter Fifteen — The Safety Valve That Turned Out to Be a Pipe
There is a model of the human mind that arrived in the nineteenth century, when the most impressive machine anyone had seen was a steam engine, and it has never entirely left.
In this model a person contains pressure. Drives accumulate — aggression, desire — and if they are not released they build until something bursts, which is why civilization requires channels for discharge, and why the discharge can be redirected into work or art or sport. The vocabulary is hydraulic throughout: pressure, release, outlet, sublimation, blocked, pent up. It was an entirely reasonable metaphor in 1895 and it produced a hundred years of confident advice.
The book on my desk runs the whole argument and extends it to the internet. Aggression and sexual drive are natural inherited needs; prohibitions cannot remove them; but the ability to discharge them in cyberspace will reduce their appearance in reality, however paradoxical this sounds. Give the violent a violent game. Let the maniacs and the murderers and, in a sentence I have enjoyed more than I should, the politicians satisfy their appetites in a virtual world where the victims are not real. Improve the quality of the simulation and human temperament will be recalibrated toward peace.
The hydraulic model has been tested about as thoroughly as any idea in psychology, and it does not survive. The purest test is the oldest folk remedy: hit something when you are angry and you will feel better. Experiments in which angry people are invited to strike a punching bag find, consistently, that they subsequently behave more aggressively toward the person who angered them, not less. Rehearsing anger does not drain it. It practices it, and it teaches the person that anger is a thing one acts on.
The specific question of whether violent games make people violent has been fought over for three decades and remains genuinely contested; I am not going to pretend otherwise, and anybody who tells you the matter is closed in either direction is selling something. But the prediction under examination here is not the contested one. The prediction was that a virtual outlet would visibly reduce real aggression, and that is the claim that failed. The outlet was built, at enormous scale, and no corresponding drainage appeared anywhere.
What appeared instead was something the hydraulic model has no vocabulary for. Aggression online did not substitute for aggression offline; it developed into its own activity with its own rewards, its own communities, its own status ladders, and — the decisive feature — its own audience. A man shouting in a field is discharging something. A man shouting to forty thousand people who agree with him and applaud is doing a job, and he will come back tomorrow to do more of it, because it worked.
That is the mechanism the forecast missed, and it is the opposite of a valve. A valve reduces pressure by letting a fixed quantity out. What was actually built was a distribution network, and a distribution network does not deplete its contents. It finds the people scattered too thinly to have ever formed a group, connects them, and thereby manufactures a quantity of coordinated feeling that did not previously exist in any container at all. The number of people with a given grievance in a city of a million might be two hundred, who will never meet. Connect the same two hundred across a continent and you have an organization.
It is worth asking when substitution does work, because it sometimes does, and the honest answer sharpens the point. A substitute succeeds when it delivers the actual good the person was after. Methadone works because the body cannot tell the difference in the way that matters. Regulated gambling displaces illegal gambling because a bet is a bet. Substitution fails when what is offered is a rehearsal of the good rather than the good itself — when the person wanted to be feared and is given the simulation of being feared, or wanted to be desired and is given an image. A rehearsal does not satisfy an appetite. It trains one.
So the general form of the error, which is worth naming since we will meet it again, is this: treating a human drive as a quantity to be routed rather than a practice to be reinforced. Anyone designing an outlet for something should ask which of the two they are actually building, and the test is simple enough. Does the person leave having had enough, or does the person leave having got better at it?
Chapter Sixteen — Efficiency Makes Us Use More
A young English economist published a book in 1865 pointing out something that his contemporaries found perverse and that most people still find perverse today.
Improved steam engines used dramatically less coal to do the same work. Everyone assumed this would extend Britain’s coal supply. William Stanley Jevons observed that the opposite had occurred. Cheaper power meant power was worth using in places where it had not been worth using; new industries became viable; the engine spread into applications nobody had considered. National coal consumption did not fall with efficiency. It climbed with it, and it climbed faster.
The finding is so counterintuitive that it has had to be rediscovered roughly once a generation ever since, usually by somebody who has just watched a conservation policy fail.
The clearest modern demonstration is light. Over the past two centuries the amount of illumination obtainable for an hour’s wages has increased by a factor so large that writing it down is not useful — candle to whale oil to gas to filament to fluorescent to the diode, each step an enormous gain. If the demand for light were a fixed quantity, we would now spend almost nothing on it. Instead we have illuminated car parks that no one crosses at three in the morning, floodlit buildings that nobody is looking at, and cities visible from orbit. Every efficiency gain was absorbed by wanting more light, in more places, for longer, until light became something we throw away.
The household version is the one that ought to be taught in schools. The twentieth century equipped homes with washing machines, vacuum cleaners, refrigerators, and hot running water, all of them genuine and enormous labor savers, and the historian who went looking for the corresponding collapse in domestic work found that the hours had not fallen. What had happened was that the standard rose to absorb the saving. When laundry meant a day of boiling and wringing, a shirt was worn for a week. When laundry became a button, wearing a shirt twice became slovenly. The machine did not reduce the work; it raised the definition of clean, and then required the work anyway.
Notice the shape of that, because it is not quite the same as Jevons and the difference matters. Jevons describes a price falling and demand rising to meet it. The household case describes something subtler: an expectation adjusting until the old burden is restored at a higher standard of living. Both are restoring forces. Only one of them can be seen in any set of accounts.
Now the necessary correction, because this argument has become a favorite of people who wish to conclude that efficiency is pointless, and that conclusion does not follow.
The rebound is usually partial. When cars become more fuel-efficient, people do drive more, but they do not drive so much more that fuel consumption rises — the extra driving typically claws back some fraction of the saving, not all of it and rarely most of it. The cases where the rebound exceeds the saving entirely are real, but they are the exception, and they tend to involve a technology so transformed by the efficiency gain that it enters entirely new markets, which is what happened with steam and with light. For a household appliance or a national vehicle fleet the honest expectation is that you will keep a good deal of what you gained and lose some of it to your own enthusiasm.
The practical lesson is therefore not that efficiency fails, but that the engineering estimate of a saving is a ceiling rather than a forecast. An engineer calculating that a new furnace uses thirty percent less fuel has calculated correctly and has not predicted anything, because the calculation holds the householder’s behavior fixed and the householder is the one variable guaranteed to move. Somebody who can now afford to heat the spare room will heat the spare room.
This has a direct bearing on the book on my desk, whose economic chapters rest almost entirely on savings of this type. Virtualize the state and taxation becomes cheap. Move education onto the network and it costs practically nothing. Automate production and the energy required collapses. Each of those calculations is a ceiling computed with the human beings held still. What actually happened to educational technology is instructive: the marginal cost of delivering a lecture fell to nearly zero, exactly as predicted, and the cost of obtaining a degree in most of the developed world went up.
Chapter Seventeen — The Machine That Listens at Three in the Morning
Somewhere in Ontario, at some point before 2007, a man sat down at a computer, opened a web page running a program written four decades earlier, and typed that he was sad.
The program asked whether he was often sad. He said yes. It asked whether he was sure. He said no. It observed that he was being rather negative. He typed that he did not want to live. It asked why he said that. He said because nobody loved him. It expressed doubt. He conceded that his mother loved him. It asked who else in the family loved him. He said it seemed like nobody. He then added that perhaps his cat loved him too, and the program asked why he said that the cat loved him.
He was conducting an experiment and he knew exactly what he was talking to. He notes in the text that by the third or fourth exchange one can feel that the program does not understand. And he still typed that he did not want to live, to a machine, having invented for the purpose a version of himself who was depressed, which is an act with rather more going on in it than the surrounding paragraph acknowledges.
Out of that afternoon he produced the cleanest correct prediction in his book. Replacing the psychologist with a computer, he wrote, can have a therapeutic effect, because in a human being we always see somebody who could judge us and be privately disgusted by our weaknesses. There are things a patient will tell a machine that they will not tell a person, and they will say them sooner, because the machine has no moral commitments and none of the reluctance that any human practitioner eventually shows. He predicted that daily conversation of this kind might become an ordinary routine, like brushing one’s teeth.
Everything in that paragraph has turned out to be true, and the reasoning was right as well as the conclusion, which is rarer. When researchers later ran the controlled version — the same interview conducted by a human and by something visibly artificial — people disclosed more to the machine, showed less fear of being judged, and displayed more of the small physical signs of genuine distress rather than the managed version people present to other people. The effect is not a failure of the human interviewer’s skill. It is a structural property of being observed by something that cannot think less of you.
The man who built the original program was horrified by all of this and spent much of his later life arguing against his own creation, and his objection deserves better than the summary it usually receives. It was not that people were being fooled — he understood they were not, exactly. It was that certain tasks should not be delegated to machinery regardless of how well the machinery performs, because the value of the act lies in its being performed by someone for whom it costs something. A machine can compute a judgment. It cannot choose, in the sense of having something at stake in the choosing, and to hand it work that consists of caring is to redefine caring as the production of the correct output.
Two things are worth saying about that objection now that the world has decided against it in practice.
The first is that the objection assumes the alternative is a person. For a great many people at three in the morning the alternative is nobody, and an argument that a machine is an inferior substitute for a friend is not an argument at all if the friend is not available. Most of the moral force of the case evaporates on contact with the actual queue for human attention, which in most countries is measured in months.
The second is that the objection identifies something real that the enthusiasts have no answer to. A listener that cannot judge you also cannot object to you. Part of what a person gets from another person is friction — the moment when somebody who loves you says that you are being unfair to your sister, and the statement carries weight precisely because it costs them something to make it and because they will still be there next week to see whether you took it. A machine optimized for the absence of judgment has, by construction, removed the mechanism by which we are corrected. Whether that matters depends on what the person came for, and people frequently come for reassurance while needing correction.
Where is the restoring force in this one? I think it is visible but small, which is what a restoring force looks like early. There is a growing insistence on knowing whether one is speaking to a person. There is regulation arriving unevenly. And there is the beginning of a status inversion, in which human attention becomes the premium product precisely because the machine version is free — the same pattern that made hand-made goods expensive after the factory arrived. My guess, and it is only that until the final chapters where guesses are supposed to carry dates, is that we end up with an arrangement in which the machine handles the volume and human contact becomes something people pay for and boast about.
Chapter Eighteen — Right Result, Wrong Machinery
The ninth chapter of the book on my desk is called the formula for immortality and it takes about four pages, which is efficient for the subject.
The proposal is this. While you are talking to somebody over a video connection, a program records their gestures, expressions and reactions to what you say. After several hundred conversations there exists a database of that specific person’s audio-visual responses. The program is then taught to deploy them — to react to your words, to draw on the accumulated stock of their opinions, to compose new remarks from the templates of their old ones and from the day’s news. The result is a companion who listens to your news, laughs at your jokes, and holds views. The author suggests it might be marketed under a name meaning something like eternal life, and adds, correctly, that the first versions will stutter.
The outcome he describes has arrived. The route he describes is not how it happened, and the difference is the most instructive single item in this whole book.
What was actually built was not a recording of one person. It was a system trained on the written output of an enormous number of people, which acquired from that aggregate the general shape of how humans respond, and which can then be pointed at any particular voice. It is not a copy of anybody. It is a distillation of everybody, wearing whichever face is requested. The architecture is not merely different from his proposal; it is close to the opposite of it — general first, particular afterward, where he assumed particular all the way down.
Now for the detail that has occupied me more than anything else on this desk. He had the correct mechanism in his hands, two chapters earlier, and did not connect it.
In the chapter on virtual personality he describes exactly how the conversational program he had been talking to worked, and he describes it accurately: it accumulates answers from a very large number of dialogues with a very large number of people, deposits them in a general database, and selects from that common pool according to what you have just said. He even notes the number of conversations it had held, which ran to millions. He looked directly at a machine that produced plausible individual responses out of aggregated human material, wrote down how it worked, and then, forty pages later, proposed that the way to build a convincing person is to record that person.
Both components were in the same volume. The join was never made. And I want to insist that this is not a failure of intelligence, because the join is only obvious from where I am sitting, which is a chair that exists because somebody else made it.
The error has a name I would give it, which is the intuition of imitation: to reproduce a thing, copy the thing. It is the most natural assumption a mind can make and it has been wrong at almost every important juncture in the history of technology. To fly, copy a bird — and the flapping machines all failed, and flight arrived from a fixed wing, which no bird has. To make a machine understand language, write down the rules of the language — and decades of that produced very little, and the results came instead from statistics over vast quantities of text, which contain no rules anybody wrote. To make a machine play well, encode the wisdom of experts — and the machines that eventually surpassed the experts had been given hardly any expert wisdom at all.
The pattern is consistent enough to be worth stating as a rule of thumb. Forecasters predict the future by scaling up the method currently in front of them, because that is the only method available for inspection. The future usually arrives by a different method, and quite often by one that had already been tried, dismissed as crude, and left lying about.
So how do I grade this? Under the rule I set out earlier, a prediction does not count as a hit if the thing arrived by a route the author expressly ruled out. He did not rule this route out; he simply failed to see it, which is a lesser offense. So: partial credit for the outcome, and no credit for the mechanism.
And the distinction is not bookkeeping. Somebody who predicts an outcome without the mechanism has no ability to say what comes next, because they do not know what they are watching. If you believe the technology is a recording device, you expect the industry to be organized around consent and estates and the rights of the deceased. If you understand it is a distillation of a general pool, you expect the industry to be organized around who owns the pool, which is precisely the argument now being conducted in every courtroom that will hear it. He got the destination and missed the road, and the road is where all the traffic is.
Chapter Nineteen — The End of Money, Again
The most fully worked-out proposal in the book on my desk is a plan to abolish cash, and it is worth setting out because it is not vague at all.
The state takes over the banking system — buying controlling stakes over twenty-five years rather than seizing anything — and withdraws paper and metal money from circulation. All payment then runs as electronic transfer. A levy of five percent is applied when money enters an account and five percent when it leaves, so the treasury collects roughly a tenth of every transaction and every other tax is abolished. Nobody can evade it. Nobody can pay an undocumented worker off the books. The drug trade, which the author notes runs on banknotes, loses its medium. The cost of collecting taxes, which is enormous, disappears, because collection becomes an automatic property of the payment system itself.
Half of this has happened and half of it has produced the clearest reversal in the whole volume, which makes it an unusually good specimen.
Cash as a way of paying for things has collapsed across most of the developed world. In several countries the share of transactions conducted with notes and coins has fallen into the low single digits, shops have begun declining cash on the grounds that handling it is a nuisance, and an entire generation has reached adulthood without ever having queued at a bank counter. That much the forecast got right, and got right early, at a time when the assertion sounded eccentric.
And here is the thing that no one predicted: over the same period, the quantity of banknotes actually in existence went up. Not down. In the major currencies the value of notes in circulation grew steadily through the years in which people stopped using them to buy anything, and jumped sharply in 2020. A world that had stopped paying with cash was accumulating it.
The explanation reveals a mechanism this book has not yet named and which I think is the second most useful one in it, after the restoring force itself. Money is not one thing. It is at least four things bundled into a single object: a way to pay, a way to store value, a way to transact privately, and a way to transact when nothing else works. A banknote does all four. Nobody chose that arrangement; it is simply what a piece of paper happens to be good at.
When a superior technology arrives, it rarely replaces the whole bundle. It takes the function it is best at and leaves the others behind. Electronic payment took the payment function completely, because it is better at it in every respect. It did not take the store-of-value function, which is why people hold notes in a drawer against inflation and bank failure. It did not take privacy, for the obvious reason that it is the opposite of privacy. And it emphatically did not take the last function, the one that only becomes visible when the power fails, the network goes down, the bank has a bad morning, or the country is invaded.
So the object did not die. It specialized. And the pattern generalizes with almost no adjustment: the horse lost transport and kept sport and ceremony; the letter lost correspondence and kept the legal and the intimate; the physical book lost reference and kept reading. When you meet a prediction that something will disappear, the question to ask is not whether the replacement is better but which of the several jobs it is better at, because the remaining jobs will keep the old thing alive at reduced volume and higher meaning.
The reversal is sharper still, because the restoring force in this case is legislation and it is visible in the record. Cities and states that watched shops refuse banknotes began passing laws requiring them to be accepted, on the grounds that a cashless economy quietly excludes the poor, the old, the undocumented and the unbanked. Central banks in some of the most enthusiastically cashless countries in the world have since gone the other way, warning that a payment system with a single point of failure is a national vulnerability, and national preparedness advice in more than one northern European country now tells households to keep physical money at home for emergencies. Within about fifteen years the same institutions moved from encouraging the death of cash to insisting on its survival, and neither position was foolish at the time it was held.
As for the second half of the proposal — the state as sole banker, the automatic universal transaction levy — it remains politically inconceivable, and it has nonetheless become the exact subject of the current argument about state-issued digital currency, where every objection raised is one the 2007 text simply did not consider: what a government could do with a complete record of every purchase, and what it could do with money that can be programmed to expire or to be unspendable on certain things.
Verdict: pending, and I said in an earlier chapter that pending costs a date. Mine is 2040. By then a major economy will have formally retired physical currency or it will not, and this claim can be closed by somebody with a shorter memory and a longer future than mine.
Chapter Twenty — The Abolition of Work
In 1930, in the opening months of the worst economic collapse of the industrial era, John Maynard Keynes wrote a short and cheerful essay about his grandchildren.
He argued that the current misery was a temporary maladjustment and that, on any reasonable projection of compounding productivity, the economic problem — the struggle for subsistence which had occupied the entire history of the species — would be substantially solved within about a hundred years. His grandchildren would work perhaps fifteen hours a week, largely to feel useful, and their real difficulty would be what to do with themselves.
The productivity forecast was roughly correct. The working week was not. We arrive at his deadline with the arithmetic vindicated and the conclusion refuted, which makes this the most respectable failed prediction of the twentieth century.
What he missed is that leisure was not the only thing the surplus could be spent on, and it lost the auction. Hours did fall enormously — from something close to seventy a week in the industrial nineteenth century down through the decades until roughly the nineteen-seventies, at which point the decline slowed almost to a stop in most rich countries. The gains stopped being taken as time and started being taken as things: larger dwellings, second cars, holidays by aircraft, and a standard of ordinary comfort that would have been aristocratic in 1930.
Underneath that choice sits a mechanism which defeats every forecast of general contentment ever written. Some of what people want is absolute — food, warmth, a dry roof — and can be supplied to everybody at once. The rest is positional. A good school is good relative to other schools. A desirable address is desirable because not everyone can have it. Status cannot be manufactured, only redistributed, and a society growing richer in unison does not satisfy positional appetite by a single unit. Whatever fraction of human effort is directed at relative standing is therefore immune to productivity, and it absorbs surplus indefinitely, like a drain that has been installed at the bottom of the swimming pool.
The book on my desk arrives at this territory from a completely different direction and reaches a proposal that is more radical than Keynes. Its argument is that the state has adopted job creation as its measure of success, and that this produces work which exists in order to exist — the road nobody will drive on, the cashier who exists to slide items past a scanner that could read them anyway. It would be cheaper, he says, to let a substantial share of the population stay at home on an adequate allowance than to manufacture employment for them at great expense. Give them the money, let them study or fish or write poetry, and stop pretending the activity was ever the point.
It is worth recording that an anthropologist made a closely related argument, from the political left, a decade later, and it became an international bestseller. Two men with nothing in common looked at the same thing and saw the same thing, which is usually a sign that the thing is there.
The trials that have since been run on paying people unconditionally have produced a fairly consistent picture, and it is not the picture either side wanted. People do not, on the whole, stop working; the effects on employment tend to be modest. What improves, reliably and substantially, is health, stress, and the sense of being able to plan. The money buys stability more than it buys leisure, which is a genuinely interesting result and one that nobody had predicted, since the entire argument had been conducted on the assumption that the question was whether people would remain willing to work.
But the restoring force in this domain is not economic at all, and the evidence for it is nearly a century old. In the early nineteen-thirties a team of researchers spent months in an Austrian village whose single factory had closed, and what they documented was not primarily poverty. It was the disintegration of time. People walked more slowly. They lost track of the day. They stopped attending the meetings of the political organizations they had been passionate about. Borrowing from the free library fell, despite the sudden abundance of hours in which to read. Men who had planned to use the time for something reported, with evident bewilderment, that the time had gone and they did not know where.
That study is why I regard the abolition of work as the most underestimated hazard in the whole utopian tradition, and why the book on my desk, which is otherwise attentive to human needs, has a hole here. Its author organizes everything around a hierarchy of needs and then treats employment purely as an income mechanism, when the evidence is that a job supplies structure, enforced society, a reason to be somewhere, and a defensible answer to the question of what one is. Remove the income problem and those four remain, unaddressed, and they do not solve themselves. A person needs something that is expecting them.
Chapter Twenty-One — The Abolition of the Family
On the early kibbutzim the children slept in a separate building.
This was not a hardship arrangement but a principle, worked out deliberately and defended at length. If the family was the engine of private property and of inequality between men and women, then raising children communally, in their own houses with trained carers, would produce a generation free of possessiveness. Parents saw their children for a set period each afternoon. The arrangement ran, in many communities, for decades.
It is essentially gone. It did not end by decree or by ideological defeat; it was dismantled community by community, over the nineteen-seventies and eighties, mostly at the insistence of mothers, until the children slept in their parents’ houses again. The ideology remained popular considerably longer than the practice, which is generally the way round these things happen.
The abolition of the family is the oldest recurring prophecy in Western thought and it has the worst record of any of them. Plato proposed that his guardians should hold wives and children in common so that no man would know his own son and all would care equally for all. Nineteenth-century socialists announced its impending disappearance as a matter of historical inevitability. The early Soviet state pursued it with real energy — communal dining, communal laundry, divorce made trivially easy — and then reversed course within about fifteen years, restoring marriage, restricting divorce, and awarding medals to mothers of large families, which is one of the fastest and least apologetic policy reversals of the century.
The book on my desk is interesting here because it is on both sides. In one chapter the family is being deliberately destroyed by a state that has replaced moral order with an effective police force and no longer needs households to keep people in line. In another the family is about to be restored as the core of society, once the network allows both parents to be at home and the state stops needing to drag them into cities as a workforce. The author does not appear to notice that he has predicted the same institution’s destruction and its restoration ninety pages apart, and the contradiction is more informative than either prediction, because it marks the exact place where a thinker cannot get a stable reading.
What actually happened is the thing that neither prophecy allows for, which is that the institution split into its components and the components went in opposite directions.
As a legal and economic arrangement, the family did weaken, substantially and almost everywhere. Marriage rates fell, marriages happened later, cohabitation became ordinary, divorce became unremarkable, and the fastest-growing household type across the developed world is one person living alone — a form of life that was statistically negligible for most of human history and now houses a large minority of adults in some countries.
As a relationship between parents and children, it intensified. Time-use surveys are unambiguous and counterintuitive: parents in the rich world spend more hours directly engaged with their children than parents did in the nineteen-sixties, and this is true of mothers despite the enormous increase in women’s paid employment, and dramatically true of fathers. Standards of supervision that would have astonished a parent in 1965 are now unremarkable. The child who used to be sent outside until dinner is now driven to an activity.
So the same period produced fewer families and more parenting. The unit shrank and the bond thickened, which is not a contradiction but a reallocation — the family lost the functions that could be performed better by institutions and markets, and concentrated everything it had left on the one function nothing else can do. And in the background, quietly, the multigenerational household has been returning in several countries, not for ideological reasons but because housing became unaffordable, which is how most durable social change actually arrives.
The lesson is the one the previous chapter began and this one completes. Institutions rarely die. They lose jobs to competitors and specialize in what remains, and the specialization frequently makes them more intense rather than less. Anybody predicting the disappearance of a very old arrangement should first list what it is currently doing, then ask which items on the list have a serious rival, and then look hard at whatever is left over. Whatever is left over is the thing that will still be there in fifty years, doing more of itself than before.
Chapter Twenty-Two — The Abolition of Death
Two chapters of the book on my desk are about immortality, and they disagree with each other, which is the single thing I admire most in the whole volume.
The first is confident. It quotes a NASA scientist arguing that since the difficulty with biological molecules is that they decay, the sensible move is to transfer the contents of the brain onto semiconductor chips, which tolerate wide temperature ranges, require neither food nor oxygen, last for millennia, and can be copied. The biological body then becomes a burden — it freezes, it needs clothes, it damages easily — and would be better replaced with limbs of great strength that can be upgraded when a newer model appears. Divert a tenth of what we spend on medicine into electronics, the argument runs, and immortality arrives in the near future.
The second chapter is nervous, and much better. It works through what immortality would actually mean, notes that awareness of one’s own non-existence is a peculiar thing to attempt, observes that the question dissolves entirely if the ordinary understanding of time is wrong, and treats the whole matter as unresolved. Having proposed a solution, the author sat down and argued with himself in print. That is not a common act.
The promise itself is very old and has been arriving imminently for a long time. Alchemists sought the elixir. A Nobel-winning biologist at the beginning of the last century believed that intestinal bacteria caused aging and could be countered with fermented milk. A Russian revolutionary and physician was so convinced that repeated blood transfusion would rejuvenate the body that he ran an institute for it and died in 1928 from a transfusion he administered to himself. A philosophical movement in the same country had earlier proposed that the common task of humanity was the literal resurrection of every ancestor who had ever lived. Modern cryonics began in the nineteen-sixties with a book and, shortly afterward, with the first man to be frozen in the hope of later repair. And in our own decade the promise has taken the form of escape velocity: the claim that medicine will soon extend life by more than a year for each year that passes, after which nobody who is alive need ever die.
Against all of that, the record.
Life expectancy at birth roughly doubled across the developed world over the twentieth century, which is one of the great achievements of the species and is routinely misdescribed. Most of that gain came from stopping infants and children from dying and from defeating infectious disease. It was not achieved by making old people older; it was achieved by making it normal to become old at all. And the gains have slowed markedly since the nineteen-eighties, because the remaining causes of death are the difficult ones.
Meanwhile the ceiling has not moved. The oldest human being whose age is properly documented died in 1997, at a hundred and twenty-two. In the decades since, with billions more people alive, better medicine everywhere, and the wealthiest individuals in history spending freely on the problem, nobody has come close to her. The survival curve has become more rectangular — more of us reach the wall — but the wall has not visibly receded.
This is the one domain in this book where the restoring force is not economic, institutional or psychological. It appears to be biological, and it is the sort that yields to no price signal. That does not make it permanent; the biology of aging is a live research field, several interventions extend life substantially in laboratory animals, and I am not going to conclude a chapter on failed prophecy by issuing a confident prophecy of my own in the other direction. But the pattern of the last fifty years is unambiguous: every decade has produced an announcement that the problem is about to be solved, and every decade has ended with the same number at the top of the table.
There is also the matter of what is being promised, which the nervous chapter is right to press. The chip proposal assumes that a person is a pattern of information and that moving the pattern moves the person. Perhaps. But a copy of you, running elsewhere, is straightforwardly a different entity from the one now reading this sentence, and the fact that it would insist otherwise is not evidence, since it would insist otherwise either way. Every version of this technology that has actually been built produces a thing that resembles somebody and is not somebody, and the resemblance is the whole product.
Verdict: reversal on the timing, which was supposed to be soon and has been supposed to be soon since the seventeenth century, and genuinely open on the substance, with an expiry I am prepared to state. If by 2060 the verified maximum human lifespan has not exceeded a hundred and twenty-five years, the escape-velocity claim of our decade should be filed with the alchemists and not before.
Chapter Twenty-Three — The State Was Supposed to Wither
In 1997 two authors — a financial journalist and a member of the House of Lords who had edited The Times — published a book arguing that the nation-state was finished.
Their reasoning was elegant. Wealth was moving from land and factories, which sit still and can be taxed, into information, which does not. Anyone with a portable computer and a satellite connection could conduct business from anywhere, and would therefore choose the jurisdiction offering the best terms. Governments, unable to compel, would have to compete for residents as businesses compete for customers. Taxation would collapse toward whatever the most mobile citizens were willing to pay. The state, which the authors described as history’s most successful protection racket, would be reduced to a service provider. Their memorable image was of a farmer’s cattle acquiring wings.
The book on my desk quotes this argument at length and with evident delight, and builds on it a picture of the individual as a sovereign entity — a person-state, transacting globally, unbound by the accident of where they sleep.
What happened next is the most complete reversal in this book, and it is complete because the same technology that was supposed to liberate the taxpayer turned out to be far more useful to the tax collector.
Consider what a state actually needs in order to tax you. Not force — force is cheap. It needs to know what you have. For the whole of history this was the binding constraint, which is why smuggling worked, why cash worked, and why an offshore account behind a wall of national secrecy worked. Then the records went digital, and a digital record can be searched, matched, and shared between countries at almost no cost. Within about a decade the United States had legislated to require foreign banks to report on American clients, dozens of countries had adopted an automatic exchange of financial account information, Swiss banking secrecy in its traditional form had effectively ceased to exist, and more than a hundred and thirty jurisdictions had agreed in principle to a minimum rate of corporate tax specifically to end the competition the 1997 book had predicted would be irresistible.
Government spending as a share of the economy did not fall over this period in the countries concerned. It rose.
The mechanism deserves a name, and I will call it the legibility trap. A technology that appears to empower the individual against an institution will usually empower the institution more, because the institution can afford the analysis. Both parties gain a new capability; only one of them has a department. The citizen gets the ability to move money quickly; the ministry gets the ability to see every movement of every citizen’s money simultaneously and to compare it against the declared figure. The asymmetry is not in the tool. It is in who can operate it at scale.
The second attempt at the same escape ran from about 2009 and is instructive because it repeated the error in a purer form. A currency outside the reach of any state, transferable without a bank — this was, explicitly and in the founding documents, the sovereign-individual argument again. And the outcome has been that the ledger is public and permanent, which makes it, for investigative purposes, considerably more traceable than a suitcase of banknotes; that the points where the system touches ordinary money are licensed, supervised and required to identify their customers; and that several governments now hold substantial quantities of the currency, having seized it. This is not a failure of the technology. The technology works exactly as designed. It is a demonstration that a system built to be beyond institutions ends up being used by the institutions best equipped to read it.
I should not overstate the reversal, because states did lose things over these two decades, and lost them badly. They lost the ability to control what their populations read and to be the loudest voice in their own countries, and the consequences of that are still working through. What they did not lose was money and identity, and money and identity are what a state is.
So the honest summary is that the forecast identified a genuine new capability and assigned it to the wrong party. When you next meet a prediction that a technology will liberate individuals from an institution, the productive question is not whether the capability is real. It is which side of the arrangement can hire more analysts.
Chapter Twenty-Four — Every Utopia Is a Portrait of Its Own Decade
The fastest way to date an imagined future is to look at what it does not bother to imagine.
Thomas More’s island, published in 1516, has communal property, religious tolerance of a sort, six-hour working days, and a striking contempt for gold, which is used for chains and chamber pots so that nobody will covet it. It also has slaves. The most radical social imagination in Europe restructured property, labor, worship and the family, and left slavery standing, because slavery was furniture. You do not redesign the furniture; you do not see it.
Edward Bellamy’s enormously popular American utopia of 1888 organizes the entire nation as a single efficient enterprise into which every citizen is enrolled for a term of service, with production planned centrally and distribution handled by credit against a national account. This was not a nightmare to its readers; it was the most admired institutional form of the eighteen-eighties — the great railroad corporation — enlarged until it covered everything. H. G. Wells, a generation later, produced a world administered by a competent technical elite, because his decade had decided that the engineer was the model of a serious person.
The book on my desk is exactly this kind of document and its decade is unmistakable. Its skeleton is a hierarchy of psychological needs. Its remedy for despair is therapy, delivered continuously, with pharmacology in support. Its picture of freedom is a household that has withdrawn from the city onto its own patch of land with its own power supply and a good connection. Its ultimate human goal is self-actualization, a word almost nobody used before the middle of the last century and which the author applies not only to persons but to entire civilizations. Nothing in it could have been written in 1888 and nothing in it will be written in 2088.
What survives across all these documents is not the technology and not the institutions. It is the anxiety being cured. Every utopia abolishes whatever its author’s decade found unbearable, and if you want to know what a period could not stand, do not read its complaints — read its paradises and see what has been quietly removed.
More removed idleness and ostentation, because he was surrounded by a court. Bellamy removed class warfare and the terror of the business cycle, because he had watched strikes put down by soldiers. Wells removed muddle. The book on my desk removes boredom, humiliation, the sense of being unimportant, the necessity of tedious labor, and the fear of dying — which is a fairly precise inventory of what a comfortable, educated, unaffiliated man in a wealthy country found intolerable in 2007. It is a portrait of an era that had solved subsistence and had not solved meaning.
Which gives us a diagnostic instrument to turn on ourselves, and I recommend using it before reading anything published this year. Ask what the current forecasts abolish. Loneliness is being abolished, by companions that are always available and never tired. Cognitive drudgery is being abolished. The blank page is being abolished. Nobody is currently promising to abolish poverty or war with any conviction, which tells you something about the confidence of the period. Whatever a decade proposes to remove is the thing it can no longer bear, and the removal will not work in the manner advertised — but the diagnosis of the pain is almost always accurate, and it is the most reliable information any utopia contains.
There is a darker version of the same mechanism and it needs saying, because the book on my desk exhibits it. A utopian text absorbs the political temper of its moment and preserves it, and the political passages of this one have aged worse than anything else in the volume — including a long passage borrowed from a contemporary commentator on the conflicts of the Middle East, reproduced approvingly, which reads today as a period specimen of a particular kind of rhetoric and which I am not going to reproduce here. The author’s own convictions elsewhere in the book are pacifist and cosmopolitan and, I am satisfied, sincere. That is precisely what makes the lapse instructive. A person constructing a future takes the technology from imagination and the politics from the newspaper, and the newspaper is what dates fastest and travels worst.
So the practical rule for reading any future, including the ones in the final chapters of this book: strip out everything the author borrowed from the argument that was happening around him while he wrote. What remains is the actual forecast. What you have removed is the era, and the era is what makes an old book about the future so uncomfortable to read and so useful to have.
Chapter Twenty-Five — The Paris Postcards of 1900
Around the time of the great Paris exposition, a French commercial artist and several of his colleagues produced a series of cards depicting the year 2000.
They are among the most reproduced images in the history of prediction, and they are usually shown as a joke. Firemen fly to burning windows with wings strapped to their backs. A game of croquet is conducted on the seabed by people in diving helmets. A bus is towed across the water by a harnessed whale. Postmen fly. Farmers ride machines. Everything happens in frock coats.
Two of the cards are worth more than the rest of the set combined, and they fail in exactly the same way as each other.
The first shows a barber’s shop. A gentleman sits in a chair, draped, while a machine shaves him, and a second machine brushes his hair, and a technician stands at the side of the room operating a control panel and watching. The barber has been mechanized. The barber’s shop has not. The chair, the drape, the going out to be shaved by somebody, the whole social ritual of it — all preserved, with the labor replaced.
The second shows a classroom. The pupils sit in rows wearing headsets. At the front, a machine like a mangle is being cranked by an attendant, and books are being fed into it, and from the machine a bundle of wires runs to the heads of the children, so that the contents of the books can be delivered directly into their minds. The teacher has been mechanized. The classroom, the rows, the compulsory attendance, the syllabus, the ranking of pupils, the very idea that education means the transfer of the contents of approved books into passive heads — all preserved, in every detail, with only the labor replaced.
This is the single most reliable failure in the whole history of imagining the future, and it deserves a name. Call it mechanized substitution: the forecaster removes the human from an activity and leaves the activity untouched, because the activity is invisible to him. He can see the barber, who is a person doing something. He cannot see the barber’s shop, which is not a thing anybody does but an arrangement everyone is inside.
And look at what actually happened to education, because the postcard is wrong in a way that is genuinely instructive rather than merely quaint. Nobody has ever piped a book into a child’s head. What occurred instead was that the scarcity disappeared. Every lecture, every text, every explanation, in effect all of it, became available to any child with a connection, free, at any hour. That is a far larger change than the one on the card, and it left the classroom standing, because the classroom was never mainly a delivery mechanism. It was a place where children are supervised, sorted, socialized and compelled to do the work, and none of those functions was touched by abundance. The card imagined a revolution in transmission and got a revolution in access, and misread which one the school was for.
The pattern recurs in every generation of visual futurism. The kitchen of the future, in the exhibitions of the nineteen-fifties, is a kitchen with buttons: the same meal, the same housewife, the same hour, with the labor automated. The robot butler is a butler. The flying car, as we have seen, is a car. Whenever the future is drawn rather than argued, it comes out as the present with the work removed, because a picture must show something recognizable or it shows nothing at all.
The book on my desk commits the identical error in prose, and now that the shape is visible it is easy to point at. Its redesigned state still has taxation, welfare payments, immigration quotas, an armed force, a national bank and a citizenship examination. Every one of those is retained without examination; only the labor of administering them is automated. It is the barber’s shop with a control panel, drawn by a man who could imagine the abolition of death but not the abolition of the immigration file.
Which raises the question I would rather ask than answer, since answering it is what the last chapters are for. What are we currently bolting machines onto without noticing that the thing itself is the odd part? The meeting is being automated — summarized, transcribed, scheduled — and almost nobody is asking why there is a meeting. The essay is being automated, at speed, while the question of what the essay was ever for goes unexamined. The commute is being made pleasanter. In each case the labor is under assault and the arrangement is not, and in each case I would put money on the arrangement being the thing that eventually goes.
Chapter Twenty-Six — Doomsday Pays Better Than Calm
Imagine two advisers submitting reports on the same subject on the same morning, and follow their careers for a decade.
The first writes that the situation is dangerous and may deteriorate badly. The second writes that the situation is uncomfortable but is likely to resolve itself. Now consider the four possible worlds. If things go badly, the first is a prophet and the second is finished. If things go well, the first is a responsible professional who quite properly flagged a risk, and the second is a man who was right about something everybody has now forgotten was ever in doubt. Three of the four outcomes favor the pessimist and the fourth is a draw.
Nobody designed this. It is simply what falls out of the arithmetic of blame, and it operates in every field where somebody is paid to have an opinion about what happens next. There has never been a public inquiry into why a warning was excessive. There have been a great many into why one was insufficient.
Underneath the professional incentive sits a much older one in the reader. Attention is asymmetric: negative information is processed faster, remembered longer, and weighted more heavily than positive information of the same magnitude. This is not a modern corruption; it is a sensible property for an animal that only has to miss one predator. But it means that in any market where attention is the currency — and every market for forecasts is one — alarming claims arrive with a structural subsidy that accurate claims do not receive.
I want to be even-handed about who pays for this, because the standard version of this chapter is written by someone with an axe. Money buys reassurance as reliably as it buys alarm. Industries facing regulation have funded a very large quantity of comforting research about tobacco, lead, sugar and much else, and that literature was frequently more dishonest than anything produced on the alarmed side, because it was manufactured deliberately rather than merely encouraged by incentive. The point is not that pessimists are corrupt and optimists are honest. It is that both sentiments can be purchased, and the market for one is larger and cheaper to enter.
The cost of all this is rarely counted, and it is not simply the embarrassment of wrong predictions. It is that a population saturated in warnings stops responding to them.
Hospitals discovered this the hard way. Modern monitoring equipment produces an enormous number of alarms, the overwhelming majority of which turn out to signify nothing, and the documented result is that experienced clinicians learn to filter the noise — which is entirely rational and which occasionally means a real alarm is filtered too. The remedy has not been to make the alarms louder. It has been to make them rarer and better calibrated.
The same thing has happened to the public at population scale, and anyone who has watched a genuine emergency being announced into a general shrug has seen the invoice. A community that has been told the end is imminent every eighteen months for forty years is not stupid to have stopped listening; it is behaving exactly as a well-designed filter should. Which means every unnecessary alarm has a cost that lands on somebody else’s real one, and this cost is never charged to the person who raised it, because by the time the invoice arrives the connection cannot be traced.
There is one small piece of evidence for all this sitting on my desk, and I did not go looking for it. The book I have been examining opens its central chapter by observing that nobody writes optimistic utopias anymore, only catastrophes and machine rebellions, and announces that it is going to write a cheerful one for a change. It then did so, at length, competently, with real arguments. It was published by a small press and sank without trace, in a decade during which books forecasting collapse were selling in the millions.
I do not offer that as evidence that his book deserved better. Much of it, as this volume has spent two hundred pages establishing, was wrong. I offer it as a data point about the market. The most careful reasoning about the future has no advantage over the least careful. The variable that predicts sales is the emotional register, and the emotional register that sells is dread — with one exception, which is the specific brand of optimism that promises the reader personally will become rich or immortal. Nobody has ever made money from the accurate forecast, which is that things will change substantially, in a direction nobody guessed, at a pace slower than the enthusiasts claim and faster than the institutions can manage, and that most of it will be neither wonderful nor fatal.
Chapter Twenty-Seven — My Warning Prevented It, and Other Escape Hatches
In 1974 two chemists published a paper arguing that a family of industrial gases used in refrigerators and aerosol cans would drift into the upper atmosphere, break down under ultraviolet light, and destroy ozone.
The industry’s response was that this was hypothesis, not fact. It was, at that point, entirely correct in saying so. Eleven years later a British team working in Antarctica reported a seasonal collapse in ozone over the continent so severe that they initially suspected their instruments. By 1987 an international agreement had been signed to phase the chemicals out; it was ratified with unusual speed, subsequently strengthened, and complied with. Production fell away. Atmospheric concentrations peaked and began to decline. The layer is now expected to recover over the middle of this century.
So the warning was correct, the intervention happened, and the catastrophe did not. Which means that a person today can say, entirely accurately, that scientists once predicted the destruction of the ozone layer and it was not destroyed.
This is the hardest problem in the grading of predictions and I have delayed it as long as I decently can. A forecast that changes behavior cannot be assessed by its outcome, because the outcome has been contaminated by the forecast. If the warning works, the thing warned about does not occur, and the warning looks like an error. The better the warning, the more it resembles a false alarm.
Which is why every failed prophet in history reaches for this defense, and why it cannot simply be dismissed. It is sometimes true. The trouble is that it is available for free.
So it needs a test, and I think a workable one has three parts, all of which the ozone case passes and almost nothing else does. Did the preventive action actually occur, in a form somebody can point to — a treaty, a law, a documented change in what firms did? Is there a measured chain running from the action to the outcome, so that we can see the quantity fall rather than merely assuming it? And was the action taken in response to this warning, rather than for unrelated reasons that happened to help? Where all three hold, the prophet may claim credit and should. Where the first fails — where nothing was done and the disaster still did not arrive — the claim is not a defense but a confession, because it means the forecast was wrong and its author has not noticed.
The interesting middle case is the one everybody remembers laughing at. An enormous quantity of money and labor went into remediating date handling in computer systems before the year 2000, and very little went wrong, and the episode is now shorthand for hysteria. The remediation is documented and real. It is also true that some countries and organizations spent comparatively little and suffered comparatively few failures, which is a genuine anomaly that the confident version of the story on either side does not accommodate. The honest position, which almost nobody holds, is that we do not know how much of the calm was purchased and how much was there anyway, and that this is exactly what a successful prevention looks like from the inside — indistinguishable from an unnecessary one.
The prevention defense is only the most respectable of a family of escape hatches, and the rest deserve to be named in one place so they can be recognized on sight.
There is the moving date, in which the year quietly falls off the claim and is replaced by soon. There is the retreat to scenario, in which a forecast that was received as a prediction is later described as an illustration, which is sometimes fair and sometimes constructed after the fact. There is the unstated condition, where the prophet reveals, once the deadline has passed, that the claim always depended on an assumption which unfortunately did not hold. There is the reinterpretation, in which the predicted thing is redefined until something that happened will fit inside it. And there is the assertion that it is already happening and only the inattentive cannot see it, which has the elegant property of being unanswerable.
The optimist has the same toolkit and uses it just as freely: the trend is intact and merely pausing, the technology is nearly there, adoption is following the usual curve, the critics said the same about the telephone. That last one is a particularly bad argument and I want to disable it here, since I have been on the receiving end of it. That some ridiculed innovations succeeded does not raise the probability of any given ridiculed innovation succeeding, because the ridiculed failures are far more numerous and are not remembered. It is the archive problem from earlier in this book, running in the other direction.
The remedy for all of it is simple to state and almost nobody does it. When you make a prediction, write down at the same moment two things: what observation would prove you wrong, and what would have to happen for you to be entitled to claim that your warning is what prevented it. Both sentences must be written before the outcome, because afterward the mind will generate whichever one it needs. That is the standard I am going to be held to in the chapters that follow, and it is the standard I could not hold my father to, because he never wrote either sentence, and neither has almost anyone else.
Chapter Twenty-Eight — When the Pendulum Does Not Come Back
In 1977 a hospital cook in Somalia became the last person in history to catch smallpox in the ordinary way, and in 1980 the disease was declared eradicated.
It has not returned. It will not return, because it has nowhere to live: the virus had no animal reservoir, no way of persisting outside a human body, and once the chain of transmission was broken everywhere at once it simply stopped existing outside two freezers. A disease that killed something in the region of three hundred million people in the twentieth century alone was removed from the world by a coordinated effort of about a decade, and the removal is permanent.
This book has spent a great many pages arguing that things come back, and I would be a poor advertisement for my own method if I did not now spend a chapter on the cases where they do not. A model that explains everything explains nothing, and the pendulum is worthless as a tool unless I can say when it does not apply.
The general rule is straightforward once stated. A restoring force exists when the return journey is cheaper than the outward one, or at least affordable. A ratchet exists when it is not — when the way back is far more expensive than the way out, or requires a coordination that cannot be assembled, or is physically closed.
Physical closure is the cleanest kind. An extinct species does not come back; a genome is not a population, whatever the press releases suggest. An eradicated pathogen does not spontaneously reappear. A river’s aquifer, once drained past a certain point, may compact and lose the capacity to hold water again, which is why some depleted groundwater is described as mined rather than used. In each case the outward path was cheap and the return path does not exist at any price.
Knowledge is the great upward ratchet and it is the reason the modern world is not a cycle. Once something is understood, written down, and copied widely, it has never been comprehensively lost, whatever the romance about burning libraries suggests — Alexandria did not take anything with it that was not held elsewhere. This is why the medieval fear of decline has no modern equivalent among serious people: the productivity of a society can collapse, and repeatedly has, but the recipe for germ theory is not going to be mislaid.
Some social changes are ratchets too, though fewer than optimists assume. The horse has not returned to transport, because the fleet, the breeders, the stables, the farriers and the fodder economy would all have to be rebuilt from nothing against a working alternative. Child mortality has not returned in any country that reduced it, absent war. Neither has literacy retreated where it became universal.
And now the uncomfortable half, because the same logic runs downhill, and this is the part that a book about failed pessimism has an obligation to say clearly.
Anything that is a stock rather than a flow can be destroyed faster than it can be built, and that asymmetry is itself a ratchet pointing the wrong way. Trust between neighbors takes generations to accumulate and can be gone in a season. An institution that has taken a century to become competent — a statistical agency, a public health service, a functioning court — can be hollowed out in a few years by people who do not understand what they are removing, and rebuilding it does not take a few years. Expertise is a stock: a discipline that loses its practitioners loses the ability to train the next ones, and there are trades in which this has happened and the knowledge is genuinely gone.
The record of the last two decades is unkind to anyone who assumed the good ratchets were permanent. Diseases that had been effectively eliminated in wealthy countries have returned where vaccination rates fell, which is not a natural phenomenon but a coordination failure. The taboos against certain categories of weapon and certain treatments of civilians have weakened. Nuclear arsenals, which fell substantially for two decades, have stopped falling. Peace in a place where three generations had come to regard it as a permanent property of the landscape turned out to be a flow requiring maintenance rather than a stock in the vault.
So the diagnostic, when you meet a claim that some change is irreversible, is to ask what would have to be rebuilt to reverse it. If the answer is a physical object or an industry, the claim is probably sound; those are expensive and nobody will pay. If the answer is a habit, a norm, an institution or a shared belief, the claim is almost certainly false, because those can be abandoned by simply ceasing to maintain them, and ceasing to maintain something requires no effort at all. It is the cheapest action available to any society, and it is therefore the one most likely to be taken.
The sharpest instance in this whole book is the one I examined earlier. The fall of the language barrier is a genuine ratchet; translation will not be uninvented, and the capacity is now free and universal. And it did not produce a single one of the consequences the forecast attached to it. That is the most important thing a ratchet can teach us: a change can be entirely permanent and entirely fail to deliver what was promised, because the permanence was in the tool and the promise was about the people.
Chapter Twenty-Nine — What Eighteen Years Can and Cannot Settle
Here, then, is the ledger, which is what the whole first half of this book has been assembling.
In the first column, the plain hits. That people would disclose to a machine what they would withhold from a human being, and would do so precisely because the machine cannot judge them. That conversational programs would become good enough to be preferred to human company for certain purposes, and ordinary enough to be used daily. That the internet would become a single shared imaginative space rather than a network of separate ones. That the language barrier would be dissolved by automatic translation. That solar generation would become cheap enough to matter, a claim he made tentatively and undersold. And that a great deal of employment consists of activity manufactured to satisfy the requirement that people be employed.
In the second column, the reversals — not things that failed to happen, but things that were tried and answered. That schools would close and children would be taught at home over the network. That cities would empty as the reason to be in them dissolved. That providing a virtual outlet for aggression would reduce aggression. That national identity would fade once translation removed the barriers. That the erotic, having been killed by its own abundance, would not return in any form.
In the third column, the genuinely open. Whether physical money is retired. Whether the state’s administrative machinery is eventually reorganized along anything like the lines proposed. Whether income becomes detached from employment. Whether the substrate of a mind can be changed. Whether the nation-state gives way to something else. Six substantial claims that eighteen years have not touched, and I have attached a date to each of them elsewhere in this book so that they can be closed by somebody.
And two entries that do not fit a column, which I have graded separately. Distributed work, where he was right about everything except the fraction of humanity it applies to. And the artificial companion, where the outcome arrived and the mechanism he proposed was close to the opposite of the one that delivered it.
Now stand back from the ledger, because there is a pattern in it and the pattern is the finding of this book.
Every one of his hits is about individuals. What a person will confide, what a person will prefer, what a person finds humiliating, what a person will pay to avoid. Every one of his reversals is about institutions. What a school is for, what a city is for, what a nation is, what happens when you offer a substitute to a crowd. He was consistently perceptive about human beings one at a time and consistently wrong about human beings in arrangements, and he was wrong in a specific direction: he assumed that institutions are the sum of the needs they serve, so that once a need can be met otherwise, the institution has no further business existing.
That assumption is the load-bearing error of the whole volume, and its source is visible in the structure. He built everything on a hierarchy of individual needs — a fine instrument for understanding a person and a hopeless one for understanding a school, which persists for reasons that appear on nobody’s list of needs: because the buildings exist, because the staff have careers, because the timetable makes parental employment possible, because nobody can agree on what would replace it, and because it is where the children are.
Which brings me to the honest limit of the grading itself, since I promised in the preface not to pretend that eighteen years is a verdict.
Eighteen years settles anything with a short response cycle. Consumer adoption is fast; if a device or a service was going to be taken up, it has been. Prices move fast. Business practice moves reasonably fast. Anything a person can decide alone on a Tuesday has already been decided, several times, and the aggregate is visible.
Eighteen years settles nothing that requires a generation. Whether people continue to form households in the current pattern is a question that will be answered by people not yet born. Institutional form moves at the speed of the careers inside it, which is to say about forty years, because most institutions change substantially only when the cohort that built the current arrangement retires. Values move at roughly the same pace and mostly by replacement rather than persuasion. The entire third column of my ledger consists of claims of this kind, which is not a coincidence: the questions eighteen years cannot answer are the questions about what people will decide to be.
So a fair grade has to come with a statement of what would change it, and here is mine. If, by 2045, the third column has substantially resolved in his favor — cash retired, income detached from work, the administrative state reorganized around automatic collection — then the correct reading of this book will not be that he was wrong but that he was early by a generation, and that his true error was one of pace rather than of direction, which is a far lighter charge. If instead the third column resolves against him, then the pattern I have identified is confirmed twice over, and the finding stands: he saw the person clearly and the arrangement not at all.
I do not know which. That is the point of writing the sentence down.
Chapter Thirty — How to Read Any Forecast in Four Questions
You will meet a confident claim about the next ten years within about two days of finishing this book, probably sooner, probably from someone with better credentials than mine.
What follows is what I actually use. It is not a method for producing forecasts, which I am not sure is possible. It is a method for handling somebody else’s, and it takes about four minutes.
The first question is what has to be true first, and who is in charge of it. Every forecast has a prerequisite that is not the exciting part — a standard, a law, a piece of infrastructure, a professional body’s approval, a habit that has to change, a generation that has to retire. Find the slowest one and identify who controls it. If the answer is a committee, a regulator, an insurance market, or a profession protecting an examination, add a decade. If the answer is a large number of individual householders acting on their own, add two. The technical capability is almost never the constraint; it is simply the only part of the system the forecaster can see from where he is standing.
The second question is what pulls back, and how long it takes to arrive. Somewhere in every system there is a mechanism that responds to the movement and opposes it, and the forecast almost certainly does not mention it. Look for a price that will rise, a group whose position is threatened, a scarcity that will be created, a standard that will inflate to absorb the saving. Then ask how long the response takes, because the delay is what determines whether you are looking at a gentle adjustment or a decade of apparently unstoppable movement followed by a snap. Long delays are what produce confident books.
The third question is whether the obstacle is a wall or a price. A physical limit does not move: there is only so much energy in a quantity of fuel, only so much information in a signal, and no company has ever raised money and repealed a conservation law. An economic obstacle is a door with a fee on it, and fees fall, which is why things that were absurd for thirty years become universal in four. And a human obstacle — people not wanting the thing, or not wanting their neighbor to have it — is the most common of the three and the least often modeled, because forecasters are enthusiasts and enthusiasts systematically overestimate how much everybody else wants what they want.
The fourth question is what would count as failure, and by when. If no answer exists, you are not holding a forecast. You are holding a mood with a vocabulary, and it should be enjoyed as such and never acted upon. If the person will not supply a date, supply one yourself and write it in the margin, because a claim with a date attached becomes a thing you can check and a claim without one will follow you around for the rest of your life.
Two supplementary tests catch most of what the four questions miss.
One: when something is predicted to disappear, list the several distinct jobs it currently does and ask which of them the replacement is actually better at. It will take those and leave the rest, and the rest will keep the old thing alive in a smaller and more specialized form. Almost nothing dies; things get demoted.
Two: strip out whatever the author has borrowed from the argument going on around him at the moment of writing. The political convictions, the current villain, the assumption everyone in his circle shares and nobody states. What remains is the actual forecast. What you have removed is the decade, and the decade is what will make the text embarrassing to read in twenty years.
Let me demonstrate on something live, without resolving it, since resolving it is what the next two chapters are for. Take the widely held claim that a large fraction of knowledge work will be automated within a decade. First question: what has to be true first? Not capability — capability is arriving. Liability has to be true first: somebody has to be answerable when the output is wrong, and the professions where this matters most are precisely the ones with licensing bodies, insurers and courts. Those move slowly and have every interest in moving slowly. Second question: what pulls back? The price of the human alternative falls as demand for it falls, which makes hiring a person cheaper exactly when the substitute gets good, and the volume of work expands to absorb the new capacity, as it has every previous time. Third question: wall, price or people? Not a wall. Partly a price. Substantially people, since a great deal of professional work consists of someone being accountable rather than someone being correct. Fourth: what would count as failure, and by when?
That last one is mine to answer, not the forecaster’s, and I have run out of ways to postpone it.
Chapter Thirty-One — Predictions, With Dates
Everything in this book up to this point has been the easy half, and I would like that acknowledged before I start.
Grading a dead forecast requires patience and fairness and no courage whatsoever. What follows is the other kind. These are specific, they carry years, each one names the mechanism it is betting on, and each one comes with the sentence that would prove it wrong — which is the condition I imposed on my father and could not enforce, since he never wrote it.
One further discipline, imposed on myself for a reason set out earlier. None of what follows is a warning. I am not predicting any disaster, which means that if these fail I will have no prevention defense available to me. I cannot claim that my book averted anything, because nothing here asks anybody to do anything. That was deliberate. It closes the most respectable escape hatch in the trade before I have a chance to reach for it.
By 2035, no country in the Group of Seven will have withdrawn legal tender status from its physical banknotes. Payment by cash will continue to shrink toward the margins; the notes themselves will remain legally valid and will still be printed. The mechanism is resilience: every serious attempt to remove them now runs into a national security argument about single points of failure, and that argument has been winning since about 2022. Falsified if any G7 state formally demonetizes its currency and does not reverse within two years.
By 2040, average daily travel time to work in large metropolitan areas of the rich world will be within ten minutes of its 2025 value. Not shorter. The pattern will have changed — fewer days, longer individual journeys — while the daily budget holds. Falsified if a sustained fall of more than twenty minutes appears in the national time-use statistics of three or more comparable countries.
By 2035, fully remote work will remain a minority arrangement, employing fewer than one in four prime-age workers in every G7 economy. Hybrid will be the settled norm and will be boring by then. Falsified if any of those countries reports more than a third of its prime-age workforce working entirely from home for three consecutive years.
By 2033, headcount in the licensed professions — law, medicine, accountancy, engineering sign-off — will not have fallen by more than fifteen percent from its 2025 level in any G7 country, regardless of what the technology can do by then. The mechanism is liability, as set out in the previous chapter: these are trades in which a substantial part of the service consists of somebody being answerable, and answerability has not yet been successfully assigned to a system. Falsified if any of those four professions in any of those countries contracts by a third or more.
By 2032, at least one major consumer service in a G7 country will market a human-only tier at a premium price and advertise the absence of automation as the selling point, and it will not be a novelty — it will be a category. The mechanism is the status inversion that follows every automation: when the machine version becomes free, the handmade version becomes expensive and boastful. Falsified if no such tier exists at scale by that year.
By 2040, no high-income country will have returned to replacement-level fertility and held it for three consecutive years as a result of government policy. Payments, tax reliefs and campaigns will continue and will continue not to work, because the thing being subsidized is not the expense. Falsified by a single clear counterexample sustained for three years.
By 2035, compulsory attendance at physical schools will remain the legal norm in every country of the OECD. Some proportion of instruction will be delivered remotely; the requirement to be in the building will not be abolished anywhere, because the building is doing four jobs and only one of them is instruction. Falsified if any OECD state removes the attendance requirement outright.
By 2035, there will still be no widely used public institution that keeps score on public forecasts. Somebody will have tried; it will not have lasted, because no constituency benefits from it. Falsified if a national government, a major broadcaster or a major newspaper establishes a public forecasting ledger and sustains it for five consecutive years. This is the prediction I would most like to be wrong about.
Two entries have already been made elsewhere in this book and I will not restate the reasoning: the maximum verified human lifespan, entered in the chapter on immortality, and the status of physical money as a store of value, entered in the chapter on cash. They carry their dates where they stand.
A final note on how these should be read, since a list like this invites a misunderstanding I would rather prevent. I do not expect them all to survive. If eight out of ten hold, I have not demonstrated skill — a decade is short, most things persist over a decade, and predicting continuity is the cheapest trick in the trade. The genuine test is which ones fail and whether they fail for the reasons I have given. A prediction that fails through a mechanism I described is worth more to whoever grades this than one that succeeds by luck, and I would rather be caught having reasoned correctly to the wrong answer than the reverse.
Chapter Thirty-Two — Predictions Against the Obvious
The previous chapter was mostly a bet on continuity, which is safe, and I said so.
This one is not. What follows are claims that run against the current consensus, made using the machinery of this book, and I expect to lose some of them badly. A method never turned against the prevailing view is not a method; it is a way of agreeing with people in a more elaborate font.
First. Measured loneliness will not fall in the countries with the highest adoption of artificial companions, and by 2035 will be no lower than it was in 2025 in any of them. This is the one I hold most firmly and it follows directly from the mechanism set out earlier: a substitute satisfies an appetite when it delivers the actual good and trains the appetite when it delivers a rehearsal. Being attended to by something that cannot leave, cannot be disappointed in you and has nothing at stake is a rehearsal of being known. It will be enormously popular and it will not do the thing it is being adopted for. Falsified by a sustained fall in loneliness on standard population measures in any high-adoption country.
Second. Sales of physical books for sustained reading will not collapse; by 2035 they will be within twenty percent of their 2025 level in the English-speaking markets, and possibly higher. The consensus that print is in terminal decline has been wrong for fifteen years already and is wrong for the reason given in the chapter on cash: the screen took reference, search and disposability, and did not take immersion. Print has been demoted to the job it is best at, which is the job people are willing to pay for. Falsified by a fall of more than a third.
Third. Credential requirements in the licensed professions will tighten rather than loosen by 2035. The obvious expectation is the opposite — that when a machine can do the technical work, the gate protecting the human should become less relevant. I expect the reverse, because when verification becomes difficult institutions do not become more permissive; they become more restrictive, since restriction is the only tool they have. Expect more examinations, more continuing-education requirements, and more insistence on a named individual signing. Falsified if any two of those professions in any two G7 countries materially reduce their entry requirements.
Fourth, and most against the mood. The interval between a capability being demonstrated and that capability being ordinary in regulated sectors will lengthen over the next decade rather than shorten. Everyone currently believes that everything is accelerating, and in the consumer domain everyone is right. In hospitals, courts, aircraft, schools and public administration the opposite is happening, because each new capability arrives with a larger burden of verification than the last, and the verification is performed by institutions that are not accelerating at all. The gap between what is possible and what is permitted will be visibly wider in 2035 than it is now, and this gap will be one of the defining political irritations of the decade. Falsified if the average approval-to-deployment interval in medicine or aviation shortens materially.
Fifth, offered at low confidence and included because a chapter of contrarian predictions containing only ones I am sure of would be a fraud. Fertility in at least one high-income country will stop falling and begin a modest sustained recovery before 2045, and it will not be caused by policy. The mechanism is unglamorous and rarely discussed: within any population, the subgroups that have children in the current environment are transmitting both their circumstances and their dispositions, and their share of the whole compounds. This is a slow, mechanical, uncomfortable argument, it is not a recommendation, and I hold it loosely. Falsified by uninterrupted decline through 2045 everywhere in the high-income world.
Now the item that this chapter owes the reader, which is the place where my own thesis should fail.
The pendulum argument predicts a return, and there is one domain where I expect none and would be embarrassed to be shown one. Automatic translation will not rebound. No restoring force will emerge, no counter-movement will make the language barrier meaningfully return, and by 2045 the idea of not understanding a foreign document will be as strange as the idea of not knowing the time. If a genuine reversal appears here — if fragmentation, regulation or deliberate withdrawal rebuilds the barrier — then the mechanism I have spent this book describing is either weaker or more selective than I have claimed, and the honest conclusion would be that I have found a pattern where there is a tendency.
And one further self-directed warning, which belongs in this chapter because it is a prediction about the author. If I am wrong across the board, I know the direction it will be in. I have spent two hundred pages watching a man overestimate how fast institutions dissolve, and the natural consequence of learning that lesson thoroughly is to overestimate how well they hold. Somewhere in the list above is an institution I have assumed will still be standing because standing is what institutions do in the story I have told myself, and it will not be standing, and the person reading this in twenty years will see which one immediately.
Chapter Thirty-Three — A Book on Somebody’s Desk in 2044
This chapter is addressed to whoever is holding this book eighteen years after it was printed, which is the same interval I have been working with throughout, and which I have chosen deliberately so that the arithmetic is easy for you.
You are probably not related to me, and if you are, I apologize for the recursion.
Start by stripping out the decade, using the instrument I described earlier and am now obliged to turn on myself. Almost everything in this book that concerns a specific technology will be embarrassing to you. I have written a great deal about artificial conversation because it is the salient thing in 2026 and it is very difficult, from inside a decade, to distinguish the salient thing from the important thing. If the important thing turns out to have been something I mention twice in passing, or do not mention at all, that will be the most predictable feature of this volume and you should not be too impressed with yourself for spotting it.
Second, look for what I have borrowed from the arguments going on around me. I have tried to keep the politics of the mid-twenties out of these pages, having watched what it did to the book on my desk, and I will certainly have failed in places I cannot see, because that is the nature of the failure. When you find the assumption I never bothered to state because everyone I knew shared it, that is the sentence to circle.
Third, the actual test, and it is not the scorecard from chapter thirty-one. Anyone can be lucky over eighteen years. The question is whether the machinery works: whether asking what pulls back, and how long it takes to arrive, and whether the obstacle is a wall or a price, produced better readings than the confident linear extrapolations of my period. If those questions helped and my specific answers did not, the book succeeded. If my answers held and the questions turn out to have been useless — if the world in your time is running in straight lines and the pendulum was a conceit of a stable era — then I got the right results by accident and you should throw this away.
Now the part I owe you, which is my best guess at my own error.
Each generation of forecasters is characteristically wrong in the direction of the previous generation’s correction, overdone. Malthus and his successors saw limits everywhere, because limits were all history had ever shown. The century that followed learned that limits could be broken and overcorrected into a cornucopian confidence that treated every warning as a joke waiting to happen, which is the mood in which the metals table of 1972 was laughed at rather than examined. My father’s generation learned that institutions were failing to keep up with technology and overcorrected into the conviction that institutions would simply dissolve — the state, the school, the city, the family, all shortly to be obsolete.
I have spent this book documenting how wrong that was, and the lesson I have drawn is that institutions are far more durable than technologists expect. Which means my characteristic error is already determined. I will have overestimated how well things hold. Somewhere in these pages is an arrangement I have treated as permanent furniture — because permanence has become the shape of my argument — and it will have gone, and the going will have been obvious in retrospect to everyone but me. This is not modesty. It is the single most reliable prediction in the book, because it is a prediction about a pattern I have observed three times and am therefore inside for the fourth.
What I would like to survive, if anything does, is smaller than a thesis. It is the three verdicts: that the honest categories are happened, reversed, and not yet, and that the third one requires a date or it is worthless. Almost every argument about the future in my time is conducted between people who insist on only two, and the shouting is largely produced by the missing column.
A last thing about the man whose book started this. He was not a professional, he had no standing in any of the fields he wrote about, and he put a date on his opinions when nobody required him to and nobody was going to check. Everything I have been able to do here rests on that, and it is a strange sort of inheritance — not the conclusions, which I have spent a book dismantling, but the habit of writing the thing down where it can be found and holding still while somebody grades it.
So the instruction to you is the one he gave me without meaning to. Take the list in chapter thirty-one, find the sentences that say what would prove me wrong, and check them. Be fair about the mechanisms and merciless about the dates. Then write your own down, with years attached, and leave them somewhere a person in 2062 can find them.
Get Peter Kriger’s stories in your inbox
Join Medium for free to get updates from this writer.
That is the whole of the method, and it is available to anybody, and almost nobody does it.
Conclusion
The book is back on the shelf now, which is where it will stay.
I have been asked, while writing this, what I think of it, and the question turns out to be harder than it sounds, because the honest answer involves two things that do not sit comfortably together. As a description of the twenty-first century it is wrong more often than it is right, and the errors are not small ones. As a document, it is worth more than a great many books that were right about more things, because it is checkable, and almost nothing is.
That combination is the argument of this whole volume compressed into a sentence, so let me unfold it one final time and then stop.
We have been asking the wrong thing of forecasts. We ask whether they will come true, which is a question about the future, and the future is precisely the thing nobody has access to. What a forecast actually contains, reliably and in every case, is an extremely accurate description of the present in which it was written — of what its author believed was solid, what he assumed would move, what he found intolerable, and what he could not see because everyone around him also could not see it. The prophecy is a guess. The diagnosis underneath it is data.
Read that way, the book on my desk becomes a much better book than it is. Its author could not imagine that people would keep paying to live near one another, which tells you what he thought a city was. He assumed a school was a delivery mechanism, which tells you what he thought education was. He believed that offering people a simulation of something would satisfy their appetite for it, which tells you what he thought a person was. Every error is a window, and the windows all look onto the same room, which is 2007, and which no historian could reconstruct from the newspapers of that year with anything like the same precision.
So the first conclusion is that old forecasts should be read the way one reads a letter rather than a timetable. Nobody checks a letter for accuracy.
The second conclusion is more practical and I have been building toward it since the second chapter. The forecasts of our own moment are not harmless entertainment. They are shaping what governments spend, what firms build, what parents advise their children to study, and what a great many people are quietly frightened of. And they are produced by an activity with no memory, no scoreboard, and no cost attached to being wrong. Every other field that matters has built an institution for remembering its failures in detail — the crash investigation, the clinical trial registry, the structural inquiry — and forecasting, which influences at least as much, has nothing. This is not a scandal because nobody has decided it is one. It is simply a hole in the arrangements that has been there so long that it has stopped being visible, which is the most durable kind of hole there is.
I do not expect this to be fixed. I have said as much, with a date on it. But the absence of an institution does not prevent an individual from behaving as though one existed, and that is the only recommendation I am going to make in this book.
Write your predictions down. Put years on them. Say what would prove you wrong. Then put the paper somewhere you will find it, and when you find it, read it honestly rather than kindly.
This costs nothing and almost nobody does it, and the reason nobody does it is worth being blunt about: it is unpleasant. A dated prediction converts an enjoyable feeling of insight into a liability that will eventually be presented for payment. The vagueness that runs through professional forecasting is not primarily a trick played on audiences. It is a comfort the forecaster is extending to himself, and any of us will extend it to ourselves the moment we are not obliged to be specific.
What you get in exchange is not the ability to see the future. Nothing in this book suggests that is available. What you get is calibration — a running record of the shape of your own errors, which is a genuinely different kind of knowledge and is the only kind that improves with age. After twenty years of this a person knows whether they habitually overestimate speed, or underestimate resistance, or systematically fail to notice the institutional actor in the corner of the picture. That knowledge is not glamorous and cannot be sold, and it is worth more than any individual correct guess, because it applies to every guess that comes afterward.
Before that, three things this book has not said, because a volume full of failed prophecies is very easy to misquote and I would rather do the misquoting myself and get it out of the way.
It has not said that warnings should be ignored. The single clearest success story in these pages is a warning that was believed and acted upon, and the atmosphere above Antarctica is measurably better for it. What it has said is that a warning is a claim like any other and should carry the same equipment as any other — a date, a mechanism, and a statement of what would show it to be wrong. A warning without those is not more urgent than a forecast. It is less accountable than one.
It has not said that limits are imaginary. The metals did not run out, and the aquifers are running out, and the difference between those two sentences is the presence of an owner and a price. Anyone who takes from this book a general confidence that human ingenuity will handle it has read the cheerful half and skipped the chapter where I drew the boundary.
And it has not said that the old forecasters were fools. I have now spent a considerable time in the company of people who were confidently wrong about the century I am living in, and the overwhelming impression is not stupidity. It is that they were reasoning carefully from what was in front of them, that what was in front of them was insufficient, and that this will be equally true of me and of you and of everybody who has ever tried to say what happens next. The condescension of the present toward the past is the least earned emotion available to a human being, since it is produced entirely by the accident of arriving later.
There is also an asymmetry in all of this that I have mentioned in passing and want to state plainly at the end, because it is the reason the subject deserves attention at all rather than merely amusement. The person who makes a bad forecast almost never pays for it. The people who pay are the ones who acted on it — who trained for a profession that was said to be disappearing and was not, or did not train for one that was said to be disappearing and was not, who bought or did not buy, moved or did not move, had children early or late or not at all, on the strength of a confident sentence written by somebody who had moved on to other sentences by the time the consequences landed. Forecasting is one of the few activities in which the risk is transferred wholly to the audience by the structure of the thing, without anybody arranging it, and this is precisely why the absence of a ledger is not a technicality.
Which is also, I think, the argument for reading amateurs. A professional forecast is shaped by the market it is sold into: it must be striking enough to publish, safe enough not to end a career, and vague enough to survive. An amateur has none of those pressures and consequently says what he actually believes, which makes his errors clean. The book on my desk is wrong in an unusually informative way for exactly this reason — nothing in it was written to be defensible. When you want to know what a period genuinely thought, do not go to its experts. Go to the people who had no reason to be careful.
I want to end on the thing this book has been circling without quite landing on, which is why any of it matters to somebody who does not write about the future for a living.
Every person makes forecasts continuously and stakes real things on them. You choose a training that takes six years on an implicit claim about what will be needed in six years. You take a mortgage on a claim about a neighborhood. You have children, or do not, partly on a claim about what kind of world is coming. You leave a country or stay in one. None of these is announced as a prediction and all of them are predictions, and the shape of a life is largely determined by which of them turned out to be roughly right.
Which means the four questions from the thirtieth chapter are not an intellectual game. What has to be true first, and who controls it. What pulls back, and how long it takes. Wall, price, or people. What would count as failure, and by when. Applied to your own decisions, they will not tell you what happens. They will tell you which of your assumptions is load-bearing, and that is the one to watch, and watching it is most of what can actually be done.
The man who wrote the book on my desk got a great deal wrong. He also, sitting in a small town at the far end of a decade nobody now remembers clearly, typed a sad sentence into a primitive program on a webpage and concluded from that afternoon that people would one day tell machines the things they could not tell anyone else. Nobody paid attention. It was not that kind of book, and he was not that kind of author, and the world was busy.
The future, when it arrived, turned out to be full of things people had already said. It usually is. That is not a reason to stop saying them. It is a reason to write down the date.
Case Studies
Case Study One — Concorde and the Supersonic Century
In the early nineteen-sixties every serious projection of civil aviation assumed the same thing: that speed would keep increasing, as it had every decade since the aeroplane was invented, and that by the year 2000 ordinary long-distance travel would be supersonic.
The reasoning was not speculative. Aviation had gone from wood and canvas to the jet airliner within fifty years, military aircraft were already flying at twice the speed of sound, and the commercial logic appeared unanswerable, since a machine that completes twice as many journeys earns twice as much. Boeing, backed by the American government, began work on a supersonic transport intended to carry nearly three hundred passengers. Britain and France, unwilling to be left behind, signed a treaty in 1962 committing both governments to a joint aircraft. The Soviet Union built its own. Airlines placed options on more than a hundred aircraft between them, from every major carrier in the world.
The challenge, as everyone understood it at the time, was engineering. Sustained supersonic flight in a passenger aircraft demanded a wing that worked at both low and high speed, an airframe that would tolerate the heat of friction, engines that could cruise beyond the sound barrier without consuming impossible quantities of fuel, and a fuselage that expanded measurably in flight. All of these problems were solved. Concorde entered service in 1976 and flew for twenty-seven years without a design-related loss of hull for the first twenty-four of them, crossing the Atlantic in under three and a half hours. Judged as engineering, the project succeeded completely.
What defeated it was everything that was not engineering, and the sequence is worth following because none of it appears in the forecasts of 1962.
The first obstacle was noise, and specifically the sonic boom, which is not a nuisance to passengers but to everybody living under the flight path. In 1973 the United States prohibited civil supersonic flight over its own territory, and other countries followed. At a stroke the most valuable routes in the world — transcontinental American ones — became unavailable, and the aircraft was confined to ocean crossings. The engineering was untouched by this. The market was halved.
The second obstacle was fuel. The oil price shocks of the nineteen-seventies arrived precisely as the aircraft entered service, and a machine whose entire economic case rested on burning a great deal of fuel very fast found itself competing against wide-bodied subsonic aircraft whose case rested on carrying a great many people cheaply. The industry, offered a choice between speed and cost, chose cost, decisively and permanently. Boeing cancelled its programme in 1971. Of the more than a hundred options held by airlines, essentially all were abandoned; the only carriers that operated Concorde were the two national airlines whose governments had built it.
The third obstacle was the one nobody could have priced, which is that the thing being sold turned out not to be speed. Business travellers, the only passengers who could justify the fare, discovered over the following decades that they valued arriving rested and being able to work en route at least as much as arriving early, and the wide-bodied cabin with a flat bed delivered that better than three and a half hours in a narrow tube. When the aircraft was finally withdrawn in 2003, following a fatal accident and the collapse in air travel after 2001, the withdrawal was not resisted by the market.
The measurable results are unambiguous. Fourteen aircraft entered commercial service out of a projected market of hundreds. Development costs, met by two treasuries, ran to several times the original estimate. Meanwhile the actual trajectory of civil aviation over the same period went in the opposite direction from the forecast: cruise speeds today are essentially unchanged from 1960, while the real cost of a seat-mile has fallen by roughly an order of magnitude and the number of passengers carried has multiplied many times over. The industry did not get faster. It got cheaper, and cheapness turned out to be the thing that transformed how humanity moves.
The relevance to the present is direct, because several companies are again promising supersonic passenger service, and the arguments being made are the arguments of 1962 with new materials. The four questions from this book apply cleanly. What has to be true first is not the airframe but the overland ban, which is controlled by legislatures answerable to people under flight paths, and which has not moved in fifty years. What pulls back is fuel cost per passenger, which is structurally worse for supersonic flight and now carries an environmental charge that did not exist before. Whether the obstacle is a wall, a price or people is answered by the historical record: it was people, then price, and never a wall. And what would count as failure is a question none of the current promoters have been asked to answer with a date.
Case Study Two — Too Cheap to Meter
In September 1954 the chairman of the United States Atomic Energy Commission told a gathering of science writers that it was not too much to expect that their children would enjoy electrical energy too cheap to meter.
The phrase entered the language and has been used against the nuclear industry ever since, generally without the surrounding context, which was a speech about the long-term promise of atomic energy in general rather than a costed projection for fission reactors. But the sentiment was not eccentric. Uranium contains an extraordinary quantity of energy by weight, the fuel cost of a nuclear plant is a small fraction of its output value, and if fuel is nearly free then electricity should be nearly free. The reasoning is sound and the conclusion was wrong, and the reason is instructive for every forecast in which a dominant cost is eliminated.
The challenge that emerged was that fuel had never been the binding cost. What determines the price of nuclear electricity is the capital cost of the plant, the cost of borrowing over the years it takes to build, and the cost of demonstrating to a regulator that it is safe. Each of those grew rather than shrank. Early reactors in the nineteen-sixties were built quickly and relatively cheaply. Later ones took longer, cost more per unit of capacity, and were subject to requirements that changed while construction was in progress, which is the most expensive thing that can happen to a large project.
Two accidents transformed the trajectory. The partial meltdown at Three Mile Island in 1979 caused no measurable public health consequence and enormous regulatory and political consequence. Chernobyl in 1986 caused both. What followed was not merely tighter rules but a collapse in the rate of new construction across most of the Western world, and the collapse itself raised costs further, because an industry that builds one plant a decade forgets how, loses its supply chain, and has to relearn on every project what a continuous programme would have retained.
The results can be stated precisely. In countries that built continuously — France constructed the bulk of its fleet within about fifteen years and derives most of its electricity from it — costs stayed manageable and the electricity is among the least carbon-intensive in Europe. In countries that built intermittently, recent projects have run years late and multiples over budget. Nuclear electricity is not too cheap to meter anywhere. It is, in most markets, among the more expensive forms of new generation to build, and among the cheaper to run once built, which is a combination that no simple forecast anticipated.
And now the pendulum, which is the reason this case belongs in a book about rebound rather than merely in a book about cost overruns.
Beginning in the early twenty-twenties, the direction reversed. Countries that had legislated to close their reactors began extending them. Plants that had been shut for economic reasons were examined for restarting, and at least one has been restarted. The reasons were not a change in the technology, which is essentially unchanged: they were the recognition that electricity demand was about to rise sharply, that intermittent generation needs something firm alongside it, and that energy imported from unfriendly countries carries a price that does not appear on the invoice. The same environmental movements that had opposed nuclear power for forty years divided over it.
Two forecasts have therefore failed within one lifetime, in opposite directions. The forecast of unlimited cheap atomic power failed because the eliminated cost was not the binding one. The subsequent forecast of nuclear power’s terminal decline, held with equal confidence by the late nineteen-nineties, failed because the value of firm low-carbon generation was not visible from where that forecast was standing. Anybody now making confident claims about small modular reactors should be asked the same question that neither camp answered: which cost is actually binding, and who controls it?
Case Study Three — The Aswan High Dam
For five thousand years Egyptian agriculture had depended on an annual flood that arrived in late summer, deposited silt across the valley, and receded.
The flood was generous and unreliable. A high flood destroyed villages; a low one produced famine. The idea of controlling it was very old, and by the middle of the twentieth century it was technically achievable. Construction of the High Dam at Aswan began in 1960 and was completed in 1970, creating one of the largest artificial lakes in the world. The forecasts attached to it were emphatic and, in the categories they addressed, correct: perennial irrigation instead of a single annual crop, protection against both flood and drought, and a large quantity of electricity for a country that had almost none.
Every one of those was delivered. The dam has generated a substantial share of Egypt’s electricity for decades. Cultivated area expanded and much of it moved to multiple harvests a year. When the Nile ran unusually low through the nineteen-eighties, the stored water in the reservoir prevented a famine that would certainly have occurred otherwise, and when it ran unusually high in the following decade, the flood was absorbed. Measured against its stated objectives, the project succeeded.
The challenge that the forecasts did not address was that the flood had been doing several jobs at once, and only one of them was supplying water.
The silt was the second job. Something in the region of a hundred million tonnes of sediment a year had been arriving on Egyptian fields, and it now settles in the reservoir instead. Egyptian agriculture, which had fertilized itself since the pharaohs, became dependent on manufactured fertilizer, at a cost that has to be paid every year forever. The third job was replenishing the coast: the Nile delta had been maintained against the Mediterranean by the sediment the river carried, and with the supply cut the shoreline began retreating, in places by metres a year, requiring engineered defenses of its own. The fourth job was flushing: the annual flood had washed salts out of the soil, and perennial irrigation without it has produced waterlogging and salinity across a significant fraction of the irrigated area, requiring an enormous drainage programme.
There was also a health consequence that was predicted by some specialists and disregarded. Permanent canals, unlike seasonal ones, support the snails that carry schistosomiasis, and prevalence rose in the newly irrigated areas. Egypt subsequently mounted a large control programme using mass drug administration, and prevalence has fallen very substantially since the nineteen-nineties, which is a genuine success and one that took thirty years and considerable expense to achieve.
The measurable balance sheet is therefore mixed in a specific and instructive way. The benefits were large, arrived quickly, and were exactly as forecast. The costs were also large, arrived slowly, were not forecast, and are permanent obligations rather than one-time charges: fertilizer every season, drainage every season, coastal defense indefinitely, and a reservoir that is slowly filling with the silt that used to go somewhere useful.
This is the unbundling problem described earlier in this book, appearing in a river. The forecasters saw the flood as a water event, because water was the thing they intended to control, and the flood was simultaneously a fertilizer delivery, a coastal maintenance system and a soil-flushing mechanism. Nobody had ever needed to name those functions, because nothing had ever threatened them.
The present-day application is uncomfortably broad. Every large intervention in an old system — a river, a forest, a labor market, an institution — is a proposal to control one function of something that is quietly performing several. The discipline that would have helped at Aswan is the one recommended in the thirtieth chapter: before removing anything, list what it is currently doing, including the jobs nobody has ever had to name, and ask what will perform each of them afterward.
Case Study Four — The Cane Toads of Queensland
In June 1935 a hundred and one toads arrived in northern Queensland from Hawaii, and by the end of the following year their descendants, numbering more than sixty thousand, had been released into the sugarcane fields.
The problem they were brought to solve was real. Two species of beetle were damaging the Queensland cane crop; the adults ate the leaves and the larvae ate the roots, and there was no effective chemical control at the time. The cane toad had been used elsewhere in the world against sugarcane pests, and reports of its success in the Caribbean circulated among sugar entomologists. An Australian delegate returned from an international conference persuaded, the toads were imported, bred, and released across the cane districts with the endorsement of the industry’s research organization.
The forecast was that the toads would eat the beetles. The toads did not eat the beetles. The beetles in question lived largely in the upper parts of the cane plant and the toad is a ground-dwelling animal that cannot climb; the adults were out of reach and the larvae were underground. The toad, being an opportunist with a very broad appetite, ate more or less everything else instead.
What followed is the most thoroughly documented biological invasion in the world, largely because Australia has spent ninety years measuring it. The toad spread west and north across the continent. The rate of advance, which was on the order of ten kilometres a year in the early decades, has accelerated to several times that, and the biological reason is itself remarkable: the toads at the leading edge have longer legs and move further, and they breed with each other, so the invasion front continuously selects for the animals best at invading. The toads now occupy well over a million square kilometres of territory they did not previously occupy.
The damage is not caused by what the toad eats but by what eats the toad. The animal carries a potent toxin in glands behind its head, and Australian predators had no evolutionary experience of it. Populations of northern quolls, several goanna species, and freshwater crocodiles collapsed in the wake of the invasion front, in some places by the great majority of individuals, because a predator that has always been able to eat a frog has no reason to stop.
Control has been attempted with essentially everything available. Bounties, hand collection, trapping, fencing of water points, and a long search for a biological or genetic control agent, none of which has produced continental-scale results. What has produced measurable results is stranger and more interesting: teaching the predators. Wildlife managers have distributed small, non-lethal toads and toad sausages in advance of the invasion front, so that quolls and goannas experience nausea rather than death and learn to avoid the prey. Trials of this conditioned taste aversion have shown substantially improved survival in treated populations. Some affected species have also begun recovering on their own, apparently through selection for individuals that are either resistant or disinclined.
The results, then, after ninety years: the original pest problem was never solved by the toad and was eventually addressed by pesticides and agronomic change; the introduced species is permanent and still expanding; several native predators suffered severe declines and some are recovering; and the most effective intervention has turned out to be preparing the ecosystem for the arrival rather than preventing it.
The forecasting lesson is not that the entomologists were reckless, though several contemporary naturalists did object and were overruled. It is that the prediction concerned a single interaction — toad eats beetle — inside a system with an enormous number of other interactions, none of which was modeled because none of them was the point. A forecast that specifies one relationship in a web is not a forecast about the web. And the modern applications are not hypothetical: every proposal to release a modified organism, introduce a control species, or remove a single element from an ecosystem is the same shape of claim, made with better instruments and the same structural limitation.
Case Study Five — The Sparrows
In 1958 the government of China identified four pests to be eliminated from the country: mosquitoes, flies, rats and sparrows.
The reasoning about sparrows was arithmetical and, as far as it went, accurate. A sparrow eats grain. Multiply the grain consumed per bird by the estimated number of birds and you obtain a figure for the food being taken from the people each year, and the figure was large. Officials calculated that eliminating the birds would feed a substantial number of additional citizens. The campaign was executed with extraordinary organizational energy: nests were destroyed, eggs broken, and across whole provinces the population turned out to bang pots and drums continuously so that the birds could not land, until they fell from exhaustion. Hundreds of millions of birds were killed.
The challenge that the arithmetic did not include was that the sparrow’s diet is not exclusively grain. Sparrows also eat insects, in very large quantities, particularly while feeding young. Removing the birds removed a predator, and the insects that the birds had been eating — including locusts — were released from that pressure.
Locust outbreaks followed. The consequences of what came next in China between 1959 and 1961 were catastrophic and their causes were multiple: an agricultural reorganization that disrupted production and reporting, requisitioning policies that removed grain from areas that had none to spare, unsound cultivation techniques imposed from above, and adverse weather in some regions. Honest accounting does not attribute the famine to the sparrows. What the sparrow campaign contributed was an additional pressure on harvests at the worst possible moment, and — more importantly for this book — an unusually clear demonstration of the mechanism by which the whole disaster occurred.
The campaign was reversed. Sparrows were quietly removed from the list of four pests in 1960 and replaced with bedbugs, and, according to accounts of the period, birds were sought from abroad to help restore the population.
What makes this case valuable rather than merely grim is that the correction was available in advance. Ornithologists in China had argued, before and during the campaign, that the sparrow’s insect consumption made its net effect on agriculture ambiguous at best. Stomach content studies existed. The objection was not a matter of philosophical opposition to the programme; it was a technical finding about what the birds ate, and it was set aside because the campaign had already been declared and because the arithmetic on the other side was simpler and more persuasive.
That is the structural point, and it generalizes far beyond agriculture or beyond any one political system. A calculation that counts a single flow — grain consumed by birds — will always be more persuasive than an argument about a network, because the calculation produces a number and the argument produces a caution. Numbers travel. Cautions do not. Anyone who has watched a policy debate in a democracy will recognize the pattern, in which a costed benefit defeats an uncosted risk not because anybody is corrupt but because one of them fits into a headline.
The second lesson concerns the speed of the reversal, which was about two years. Ecological systems have delays measured in seasons rather than decades, which meant the error announced itself quickly. Most of the failures in this book took twenty or thirty years to become undeniable, and the difference is not a difference in the quality of the reasoning. It is purely a difference in how long the system took to answer. A forecaster who happens to work in a fast-responding domain will appear far more accountable than one working in a slow one, and it is entirely an accident of the subject matter.
The present-day version of this case is any proposal to remove a single element from a complex system on the strength of one quantified flow: a species, a subsidy, a regulation, a category of employment, a course from a curriculum. The question worth asking in the room, before the arithmetic is finished, is what else the thing is doing that has never needed to be written down.
Case Study Six — Prohibition
On the sixteenth of January 1920 the manufacture, sale and transport of intoxicating liquor became illegal across the United States, and a great many intelligent people expected the country to become sober, prosperous and orderly.
This was not a fringe position. The campaign behind it had been running for the better part of a century, it was supported by a substantial body of medical opinion, and its arguments were not foolish: alcohol consumption in nineteenth-century America was genuinely enormous, and the associated damage to health, to workplace safety and to families was real and documented. The forecast was that removing the legal supply would remove the consumption, and that removing the consumption would remove the damage. Prison populations were expected to fall so far that some jurisdictions discussed closing facilities.
The first result was that the forecast partially came true, which is the part usually left out of the story. Consumption did fall, sharply, in the early years — by most reconstructions to something like a third of its previous level. Deaths from cirrhosis fell. Arrests for public drunkenness fell. For roughly the first two or three years, the policy did what it was designed to do.
Then the restoring force arrived, and it arrived from a direction the campaigners had not modeled at all: not from drinkers, but from suppliers.
A demand that persists after supply becomes illegal does not vanish; it becomes enormously more profitable to serve. The margins available to anyone willing to break the law were large enough to fund organization, transport, bribery and violence at a scale that American criminal enterprise had never previously been able to finance. Consumption began climbing again through the twenties. The product itself became more dangerous, because when alcohol must be smuggled, concentrated spirits travel better than beer, and because unregulated manufacture produced quantities of genuinely poisonous liquor — including, notoriously, industrial alcohol deliberately denatured by the government and then redistilled badly by bootleggers, which killed thousands.
Three further consequences were measurable and none had been anticipated. Federal and state treasuries lost a large revenue stream at the moment they could least afford it, which became decisive after 1929. Enforcement costs rose continuously while compliance fell. And the law manufactured a very large class of otherwise law-abiding citizens who broke it habitually, which is corrosive to the general authority of law in a way that is difficult to quantify and easy to observe.
Repeal came in December 1933, thirteen years and eleven months after the experiment began, by constitutional amendment — the only amendment in American history to repeal another.
The honest verdict is more interesting than either of the standard ones. Prohibition did reduce drinking; per-capita consumption did not return to its pre-1920 level for decades afterward, which suggests a genuine and lasting effect on habit. It also generated organized crime at national scale, killed people through adulterated supply, cost the treasury heavily, and collapsed politically. Both of those statements are true, and the reason the case is remembered as a simple failure is that the costs were vivid and concentrated while the benefit was diffuse and statistical.
The structural lesson concerns the difference between suppressing a supply and reducing a demand. A policy that attacks supply while demand persists does not eliminate the transaction; it transfers the transaction to whoever is willing to accept legal risk, and thereby funds them. This is not an argument for or against any particular prohibition — it is a statement about the mechanism, and the mechanism operates identically regardless of the substance or activity involved. Any forecast about banning something should be required to state where the demand is expected to go, and if the answer is that it will disappear, the forecaster should be asked how confident he is and by what date the claim can be checked.
Case Study Seven — The Bridge That Wobbled
The Millennium Bridge across the Thames opened on the tenth of June 2000, and closed two days later.
It had been designed by a distinguished engineering firm and a distinguished architect, analyzed thoroughly, and built to a specification that satisfied every relevant code. The structure was, by the standards of the profession, correct. What happened on the opening weekend was that it swayed sideways, sufficiently that people stopped, held the rails, and adjusted their walk, and the sway increased.
The design had accounted for vertical loading, wind, and the well-known phenomenon of soldiers marching in step, which is why troops break step on bridges. What it had not accounted for was a feedback loop that had never been properly characterized in a footbridge, and the description of it is one of the neatest illustrations of this book’s central mechanism available anywhere.
A person walking produces a small sideways force with each step. In a crowd, those forces are random and cancel out. But if the deck moves sideways even slightly, walkers adjust their gait to keep their balance, and the adjustment is naturally in time with the movement. Once a few people are in time, their forces reinforce the movement, which makes the movement larger, which makes more people adjust, which puts more of them in time. The crowd does not decide to synchronize. Synchronization is produced by the deck, and the deck’s motion is produced by the synchronization, and once the number of people on the bridge passes a threshold the loop takes over within seconds.
Note that this is a positive feedback loop — the response amplifies rather than opposes — which is exactly why it produced a spectacular failure rather than a quiet adjustment. Most of this book concerns negative feedback, the force that pulls a system back. The Millennium Bridge is the counter-case: when the loop reinforces, the outcome is not oscillation around a center but rapid divergence until something physical stops it.
The engineers’ response was exemplary and is the reason this case belongs in a book about accountability rather than only about error. The bridge was closed. A research programme was commissioned. Crowds of volunteers were walked across the structure in instrumented tests to characterize the effect quantitatively, since the existing literature was insufficient. The relationship between crowd size and the onset of instability was established. Thirty-seven viscous dampers and a number of tuned mass dampers were then fitted beneath the deck, at a cost of a substantial fraction of the original construction budget, and the bridge reopened in February 2002. It has not wobbled since, and it now carries millions of crossings a year.
Three measurable outcomes are worth recording. The remediation took twenty months and cost several million pounds against an original build cost of about eighteen million. The instrumented studies produced the first proper quantitative account of synchronous lateral excitation, which is now standard in bridge design codes worldwide and has been applied to structures on several continents. And no one was hurt.
What makes this case genuinely useful is the contrast with almost everything else in this book. Here was a failure of prediction — a well-resourced, technically expert forecast that omitted a real mechanism. Unlike the forecasts examined elsewhere in these pages, this one was refuted within forty-eight hours, in public, in front of a hundred thousand people, in a manner that could not be reinterpreted, and it was attached to a physical object whose owners were legally responsible for it. Under those conditions the profession investigated, published, corrected the codes, and improved.
That is the whole argument of the second chapter of this book compressed into one bridge. Forecasting does not improve because forecasters are unaccountable, and it improves whenever they are not. The variable is not intelligence and never was.
Case Study Eight — DDT and the Peregrine
The insecticide DDT was, by any measure, one of the most successful chemicals ever synthesized, and the man who identified its properties received a Nobel Prize for it in 1948.
It killed the lice that spread typhus, the mosquitoes that spread malaria, and the insects that ate crops, and it did so cheaply, persistently and with low acute toxicity to people. It was used at enormous scale from the nineteen-forties onward, in agriculture, in public health campaigns, and domestically. Malaria was eliminated from large regions where it had been endemic for centuries. The forecast attached to it was that insect-borne disease and crop loss were essentially solved problems.
The challenge emerged from precisely the property that made it valuable, which was persistence. A chemical that does not break down remains in the environment, and one that dissolves in fat rather than water accumulates in tissue and concentrates as it moves up a food chain. A bird at the top of that chain accumulates a dose vastly greater than anything present in the water or the insects.
The effect on birds of prey was specific and, once identified, unmistakable. A breakdown product of DDT interferes with the deposition of calcium in eggshells. Thinner shells break under the weight of the incubating parent, and reproduction fails while the adults remain apparently healthy — which is why the collapse was not noticed until populations had already fallen severely. Peregrine falcons were extirpated from the eastern United States entirely. Bald eagles, ospreys and brown pelicans declined sharply. Museum egg collections, measured decades later, allowed the thinning to be dated precisely to the period of heavy use.
Rachel Carson’s book in 1962 brought the argument to the public and was met with a well-funded campaign against her personally, which is a recurring feature of these cases and should be recorded as such. The United States restricted agricultural use in 1972; other countries followed; and the Stockholm Convention of 2001 restricted the chemical internationally while explicitly permitting its continued use for disease vector control, which is the compromise that survives today.
The measurable results are among the clearest in environmental history. Shell thickness recovered. Peregrine falcons, reintroduced from captive breeding, now nest in eastern American cities and were removed from the endangered list in 1999. Bald eagles recovered to the point of delisting in 2007. Osprey and pelican populations rebounded. Concentrations in human tissue and in wildlife fell steadily across the following decades.
And the case has an honest complication that this book is obliged to state, because leaving it out would be exactly the selective reporting the book complains about. The withdrawal of DDT from broad use coincided with the failure of the global malaria eradication campaign, and there has been a long, genuine and unresolved argument about how much of that failure is attributable to the restrictions and how much to mosquito resistance, which was already developing rapidly from agricultural overuse, and to the collapse of funding and administrative capacity. The confident version of the story on either side — that restricting the chemical cost millions of lives, or that it cost none — is not supported by the evidence, and the honest position is that vector control was made harder in ways that are difficult to quantify and that the agricultural use which was banned was also the use driving resistance.
What the case demonstrates for forecasting is the failure of a whole category: the persistent substance. Any material designed not to break down is a bet that we have correctly identified everything it will touch over the following decades, and that bet has now been lost several times — with this chemical, with the refrigerant gases described elsewhere in this book, with lead in fuel, and currently with a family of fluorinated compounds whose defining virtue is again that nothing degrades them. The pattern is reliable enough to be a rule: when durability is the selling point, the unmodeled consequences will also be durable, and they will appear at the top of some chain nobody was measuring.
Case Study Nine — The Man Who Measured the Air
In the nineteen-forties a young geochemist named Clair Patterson set out to determine the age of the Earth by measuring lead isotopes in meteorites, and found that he could not, because everything in his laboratory was contaminated with lead.
The contamination was not a laboratory failure. It was everywhere: in the dust, the water, the glassware, the air, on his own skin. To get a clean measurement he had to invent the ultra-clean laboratory, and in 1953 he produced the figure of roughly four and a half billion years that has stood ever since. But the contamination itself had become the more interesting question. Where was all the lead coming from?
The answer was that tetraethyl lead had been added to petrol since the nineteen-twenties as an anti-knock agent, and the resulting emissions had distributed lead across the entire surface of the planet. Patterson demonstrated this by measuring ice cores from Greenland, where snowfall preserves a dated record: concentrations in layers from before the industrial era were very low, and rose sharply from the nineteen-twenties onward. He also measured deep ocean water against surface water and found the same signature. There was no ambiguity about the source and no ambiguity about the scale.
The challenge was not scientific. It was that the industry had an existing account of the matter, supported by its own funded research, according to which the lead found in human blood was natural and normal. Patterson’s finding was that the baseline against which normality had been assessed was itself contaminated: everyone tested was carrying a body burden orders of magnitude above what a pre-industrial human would have had, and the comparison had been drawn between poisoned people and slightly more poisoned people.
What followed is the least edifying and most useful part of the story. Patterson lost contracts. He was excluded from a national research committee on atmospheric lead despite being the leading authority on the subject. Industry-funded scientists testified against his conclusions for years. He continued measuring.
The evidence accumulated from an independent direction as pediatric researchers established that lead exposure at levels well below those causing acute symptoms produced measurable deficits in children’s cognitive development, with no identifiable threshold below which the effect disappeared. The United States began phasing lead out of petrol in the nineteen-seventies; most of the world followed over the next three decades; the last country ceased using leaded fuel in 2021.
The measured results are among the most striking of any environmental intervention ever undertaken. Average blood lead levels in the American population fell by the great majority within about fifteen years of the phase-out beginning, tracking the removal so closely that the two curves are nearly superimposable. Comparable declines followed in every country that acted. There is a substantial and contested literature arguing that the removal also produced measurable gains in cognitive test scores and reductions in violent crime with a delay of about two decades, corresponding to the exposed cohorts reaching adulthood; the correlations are strong across many countries and the causal claim remains debated, which is the honest way to state it.
For this book the case makes two points. The first is about baselines: a forecast or an assessment is only as good as the reference state it is compared against, and if the reference state has itself shifted, every comparison drawn from it is wrong in the same direction. This is not a rare problem. It is the standing condition of anyone assessing a slow change from inside it.
The second is about who pays for the delay. The technical evidence was substantially in place by the nineteen-sixties. The full global phase-out completed in 2021. The interval between knowing and acting was, in this instance, roughly half a century, and it was not caused by scientific uncertainty. It was caused by an interested party with the resources to sustain a public argument, and the cost of the interval was borne by children who had no part in it.
Case Study Ten — The Delta Works
On the night of the thirty-first of January 1953 a storm surge in the North Sea coincided with a spring tide, overtopped and breached the dikes of the southwestern Netherlands, and drowned more than eighteen hundred people.
The warning had been issued. A Dutch engineer had argued through the nineteen-thirties and forties that the sea defenses of the delta region were inadequate, that the coastline was too long to defend dike by dike, and that the estuaries should be closed. The plan existed on paper. It had not been funded, partly because the country had spent the intervening years under occupation and then rebuilding, and partly because a plan to prevent something that has not happened is always competing against things that have.
Within twenty days of the flood the government had established a commission, and the plan that had been ignored for two decades became national policy. The Delta Works, executed over the following four decades, closed the estuary mouths, shortened the coastline requiring defense by hundreds of kilometres, and raised the design standard to an explicitly stated probability: the most densely populated province was to be protected against the flood expected once in ten thousand years, with lower but still extraordinary standards elsewhere. That choice of a stated numerical standard, arrived at by weighing the cost of construction against the value of what was behind the dikes, was itself an innovation, and it is the reason Dutch flood policy can be audited rather than merely asserted.
The engineering challenge shifted midway through, and the reason is instructive. The original design called for the Eastern Scheldt to be closed completely with a solid dam, which is the simplest and cheapest solution. By the nineteen-seventies, opposition from fishermen and from a newly influential environmental movement had established that a solid dam would destroy a large and productive tidal ecosystem and the shellfish industry depending on it. The design was changed to a storm surge barrier — a structure of sixty-two enormous gates, normally open to the tide and closed only when a surge threatens — at several times the cost of the dam it replaced. It was completed in 1986.
The results are measurable and largely favorable. No comparable flood disaster has occurred in the protected areas in seventy years. The Netherlands maintains a substantial fraction of its territory below sea level in continuous productive use. Dutch hydraulic engineering became a major export, consulted after storm surges and floods on several continents. And the tidal ecosystem of the Eastern Scheldt survives, though not unchanged: the barrier reduces tidal flow, which has produced a slow erosion of the intertidal sandbanks that is now itself a management problem requiring periodic replenishment.
The policy has also continued to evolve rather than hardening, which is the rarest property in this whole book. Beginning around 2000, Dutch planners concluded that raising defenses indefinitely was not the correct response to rising seas, and adopted a programme giving rivers more room to flood safely — deliberately reversing centuries of practice by moving dikes back and lowering floodplains, on the grounds that a system with somewhere to put the water is more robust than a system that only holds it out.
For a book about forecasting, this case is the one to keep. The warning was correct and ignored, at a cost of eighteen hundred lives. The response was proportionate, was executed, and worked. The design was revised in the face of a consequence that had not originally been valued. The standard was stated as a number, which made it checkable. And when the underlying conditions changed, the strategy was reversed rather than defended. That combination — a stated standard, an audited outcome, and a willingness to reverse — is what every field discussed in these pages lacks, and it exists here for the simple reason that the sea provides the scoring that forecasting otherwise never receives.
Case Study Eleven — The Sea That Was Spent
The Aral Sea was once the fourth largest lake in the world, and the decision to consume it was taken deliberately, in the open, with the consequences understood in advance.
This is what distinguishes it from most environmental catastrophes and from most of the cases in this book. Soviet planners in the nineteen-sixties diverted the two rivers feeding the lake — the Amu Darya and the Syr Darya — into irrigation canals across the Central Asian deserts, in order to grow cotton on a scale that would end Soviet dependence on imports. They knew the lake would shrink. Planning documents of the period discuss it. The judgment made was that the lake was, in effect, an inefficient use of water: it evaporated into the sky and produced nothing but fish, while the same water on a cotton field produced a commodity. It was a trade, and it was made knowingly.
The forecast attached to the trade was that cotton output would rise substantially and that the lake would decline to a smaller but stable size. The first half was correct. Central Asia became one of the world’s major cotton producers, and the republics involved built their economies on it.
The second half was catastrophically wrong, and the reason is a lesson in how a system behaves once a threshold is passed. A lake is not merely a container; it is a balance between inflow and evaporation, and evaporation depends on surface area. Reduce the inflow and the lake shrinks, which reduces surface area, which reduces evaporation and should produce a new equilibrium — that was the planners’ model, and it is correct up to a point. What the model omitted was that as the lake shrank its salinity rose, its thermal behavior changed, the irrigation canals themselves lost enormous quantities of water to seepage because they were unlined, and the demand on the rivers kept increasing rather than holding steady. The lake did not settle. It fell by more than ninety percent of its volume, split into separate bodies, and in its southeastern basin dried out entirely.
The consequences went far beyond the fishery, which collapsed completely and took an entire regional economy with it, stranding ships in the desert in photographs now famous. The exposed lakebed proved to be laden with salt and with residues of the pesticides and defoliants applied to the cotton for decades, and the wind lifts it. Dust storms distribute that material across the surrounding region, and rates of respiratory illness, kidney disease and certain cancers in the affected districts are elevated well above national averages. The lake had also moderated the local climate; without it, summers became hotter and winters colder, which shortened the growing season for the very cotton the whole exercise was undertaken to produce.
There is a partial recovery and it deserves recording, because it demonstrates the conditions under which a ratchet turns out not to be one. In 2005 Kazakhstan completed the Kok-Aral dam across the channel connecting the northern basin to the southern, effectively abandoning the south and concentrating the Syr Darya’s flow into the north. The northern lake rose several metres within a few years, well ahead of schedule. Salinity fell far enough for commercial fish species to return, the fishery reopened, and catches recovered to a significant fraction of historical levels. The southern basin remains desert.
The results, then: a resource consumed by deliberate decision on the basis of a trade that was correctly identified and badly priced; a secondary catastrophe in public health that nobody had costed; and a partial recovery achieved by a single well-placed engineering intervention costing a small fraction of what the original scheme did.
Two applications to the present. First, this is what the commons argument looks like in practice: the water had no price, the cotton did, and every actor in the system was individually rational. Second, and more useful, the recoverable part was recovered by giving up on the rest. The Kazakh decision was to save what could be saved and formally write off what could not, which is the option that is hardest to take politically and most often correct once a system has passed a threshold.
Case Study Twelve — New Coke
In April 1985 the Coca-Cola Company withdrew its ninety-nine-year-old formula and replaced it with a new one, having tested the decision more thoroughly than almost any consumer product decision in history.
The background was a genuine commercial problem rather than a whim. Coca-Cola’s share of the American soft drink market had been declining for years against Pepsi, and Pepsi had been running public blind taste comparisons in which most participants preferred its sweeter product. The company’s own research confirmed this repeatedly. Something in the region of two hundred thousand taste tests were conducted on the reformulated drink, and it beat both the original formula and Pepsi by clear margins.
The forecast, therefore, was not a guess. It was a projection from an enormous and correctly conducted body of data: people prefer the new formula in a taste test, therefore people will buy the new formula. The reasoning is unimpeachable and the conclusion was wrong within seventy-nine days.
What the tests measured was a sip. What the market bought was a relationship. The company received hundreds of thousands of protest contacts. Consumer groups organized. People bought and hoarded cases of the old formula. Bottlers complained. The reaction was regional and cultural as much as commercial, running strongest in the American South where the product functioned as a piece of local identity. Almost none of it was about flavor. The complaints were about having something taken away, and the company’s own subsequent analysis recognized this: the research had asked whether people liked the new drink and had never asked how they would feel about losing the old one.
On the eleventh of July 1985 the original formula returned as Coca-Cola Classic. Network television interrupted programming to report it. Within months the restored original was substantially outselling both the reformulation and Pepsi, and the company’s market position was better than it had been before the episode began — which produced a durable conspiracy theory that the whole thing had been engineered, denied by the executives involved, and implausible given the internal panic documented at the time.
The measurable results are unusually clean for a commercial case. Development and launch costs ran into the tens of millions of dollars. The reversal took seventy-nine days. The reformulated product survived in diminishing form for over a decade before disappearing. And the original brand’s sales and share both rose following the restoration.
The forecasting lesson is one of the sharpest available anywhere, and it is about the difference between an experiment and a decision. The taste tests were valid experiments answering the question they posed. The question they posed was not the question that mattered. A sip in a booth isolates flavor by design — that is what makes it a good experiment — and in isolating flavor it removes the memory, the habit, the ownership, the childhood, and the sense of being consulted, all of which are the actual product.
This failure mode is now everywhere and mostly unremarked. Any organization that tests a change by measuring the immediate response to it, in isolation, is running the same experiment: the metric moves, the decision is taken, and the thing the metric could not see is the thing that determines the outcome. The discipline that would have caught it is easy to state and rarely applied: before acting on a measurement, ask what the measurement had to remove in order to be clean, and whether the thing removed is the thing that matters.
Case Study Thirteen — Iridium
The engineering was magnificent and the market research was fiction, and the combination bankrupted one of the largest technology companies of the nineteen-nineties within nine months of launch.
The idea originated in the late nineteen-eighties, reportedly from an engineer’s wife being unable to make a telephone call from a remote location. The solution proposed was a constellation of low-orbiting satellites covering the entire surface of the planet, including the oceans and the poles, so that a handheld telephone would work anywhere on Earth without exception. Motorola committed to it and, over the following decade, built it: sixty-six operational satellites, launched on American, Russian and Chinese rockets, with the satellites relaying calls to one another in orbit so that coverage did not depend on ground stations. It cost roughly five billion dollars and it worked.
The forecast was that several million subscribers would pay a substantial monthly fee and several dollars a minute for the ability to call from anywhere. The forecast had been made in the late eighties and it was reasonable then, because in the late eighties terrestrial mobile coverage was patchy, expensive and confined to cities.
The failure was one of timing rather than of concept, and it is the purest example in this book of a prerequisite moving while a project was under construction. It took eleven years to design, finance, build and launch the system, and during those eleven years terrestrial cellular networks expanded across essentially every place where large numbers of people actually live. By 1998 the business traveller who had been the target customer could make a call from almost every city on Earth with an ordinary telephone, at an ordinary price. The remaining uncovered area was enormous in square kilometres and nearly empty of paying customers.
The product’s other problems compounded this. The handset was large and heavy, cost several thousand dollars, and required a clear view of the sky, which meant it did not work indoors, in a car, or in the streets of the dense cities where business travellers spend their time. Service began in November 1998. The company filed for bankruptcy protection in August 1999, having signed up a small fraction of the subscribers projected. At one point the deorbiting and destruction of the entire constellation was seriously contemplated.
What happened next is why this case belongs here rather than in a list of expensive failures. The assets were bought out of bankruptcy in 2000 for around twenty-five million dollars, which is roughly half a percent of what they cost to build. Freed of the obligation to service five billion dollars of capital, the same constellation providing the same service became a viable business almost immediately, serving the customers for whom global coverage is not a convenience but a requirement: shipping, aviation, the military, remote industry, scientific expeditions, disaster response and journalism. The company was subsequently profitable, publicly listed, and financed a complete replacement constellation.
The measurable outcome is therefore a five-billion-dollar loss to the original investors and a durable, useful, profitable service operated on the identical hardware. The technology was never wrong. The market was never as large as claimed. And the enterprise only worked once the price of the assets had been marked down to what that smaller market could support.
Two lessons for anybody reading a forecast today. The first is that a project with a decade-long build is a bet not on current conditions but on conditions a decade hence, and the competing technology gets to move during that decade while yours is under construction. The second is the specialization pattern described earlier in this book: the service did not die, it was demoted to the job it was uniquely good at, and it has been doing that job successfully ever since. Almost nothing disappears. Things get repriced.
Case Study Fourteen — The Segway
Before it was unveiled in December 2001, the device was known only by a code name, and the people who had seen it under confidentiality agreements said extraordinary things about it.
One prominent investor was quoted suggesting it might be a bigger deal than the personal computer. Another suggested that cities would be designed around it. The inventor, Dean Kamen, a serious engineer with a substantial record in medical devices, projected sales in the region of ten thousand units a week and a faster route to a billion dollars in revenue than any company in history. The secrecy generated a level of anticipation that no product could have satisfied.
The engineering was genuinely impressive. A two-wheeled platform that balances a standing rider using gyroscopes and accelerometers, responding to shifts in body position with no controls to learn, is a real achievement and it worked reliably from the beginning. Nobody has ever seriously disputed that the machine did what it claimed.
The forecast was that it would replace short urban journeys and reorganize the design of cities. Sales over the following decade ran to a small fraction of the projection — a few tens of thousands of units in total against a projection of that many per month.
The obstacles were not technical and every one of them belongs to a category this book has named. It was expensive, costing about as much as a used car for a device that travelled at the speed of a fast jog. It had no legal home: it was too fast for the pavement in the view of pedestrians and too slow and exposed for the road in the view of drivers, and cities spent years passing contradictory rules about where it could go, which meant a purchaser could not be sure of being allowed to use it. It could not be carried up stairs or folded into a train. And it made the rider conspicuous in a way that turned out to matter enormously, since a person standing upright and gliding along at head height above a crowd looks, to everybody else, faintly ridiculous. That last point was mocked at the time as trivial. It was decisive.
The product found its market in the places where those objections do not apply: security patrols, warehouse and campus movement, and above all guided tourism, where looking peculiar is part of the outing. Production ended in 2020.
And here is the part that makes this a case study rather than an anecdote. The function that the device was invented to serve — short individual urban journeys, faster than walking, without a car — was delivered about fifteen years later, at enormous scale, by electric scooters and bicycles distributed through rental fleets. The forecast that short urban trips would be transformed by small electric personal vehicles was correct. Cities have visibly reorganized around it, with dedicated lanes, parking, regulation and a substantial industry. The prediction failed on form and succeeded on function, exactly as the flying car did in the chapters above.
Why did the second attempt succeed? Not because the technology improved, though batteries did. Because the later form removed each of the specific obstacles: the scooter is cheap, is rented rather than bought so nobody carries the capital risk, folds or is simply left behind, fits the legal category of a bicycle in most jurisdictions, and does not make the rider look absurd. Nothing about the underlying human demand changed between 2001 and 2018. The packaging changed, and the packaging was the whole problem.
For a reader assessing forecasts today, the instruction is to separate the two claims that every product prediction bundles together. There is a claim about whether people want the capability, which is often right, and a claim about whether they want it in this object, which is usually wrong. The first can be assessed from what people already do. The second cannot be assessed at all, which is why so much confident money is lost on it.
Case Study Fifteen — The Tag That Was Going to Replace the Barcode
In the early two-thousands the retail industry was told, by consultants, technology vendors and its own largest members, that the printed barcode was finished.
The replacement was to be radio frequency identification: a tiny chip and antenna attached to every item, readable at a distance, in bulk, without line of sight. The promised consequences were dramatic and specific. A shopping trolley would be totalled instantly by being pushed through a gate. Inventory would be counted continuously and automatically. Theft would become difficult. Supply chains would know the location of every individual item at every moment. The world’s largest retailer issued a mandate in 2003 requiring its major suppliers to tag pallets and cases by January 2005, and the United States Department of Defense issued a similar requirement. Industry projections had the technology substantially displacing barcodes within about a decade.
The technology worked and still works. It was not a fraud and it was not vapour. The forecast nonetheless failed comprehensively at the item level, and the reasons are a catalogue of the failure modes this book has been describing.
The first was cost per unit. A printed barcode costs essentially nothing — it is ink already being applied to packaging that already exists. A tag containing a chip and an antenna cost, at the time of the mandate, somewhere in the range of tens of cents, and the projections assumed it would fall to about five cents within a few years. It did fall, substantially and continuously, but the economics of a supermarket are such that five cents on a can of soup is not a rounding error; it is a meaningful fraction of the margin. The technology was chasing a competitor whose price was zero and could not be undercut.
The second was physics, which no investment relaxes. Radio tags interact badly with metal and with water. A pallet of tinned goods or bottled liquid produces reflections and absorption that cause reads to fail, and a system that identifies ninety-seven percent of a pallet is not a system that can replace a process which identifies a hundred percent, because the exceptions have to be handled by a person, and the person is the cost you were trying to remove.
The third was that the benefits and the costs landed on different parties. The retailer issuing the mandate would gain the visibility; the suppliers were required to buy the tags and rebuild their packing lines. Nothing in the arrangement compensated them, and compliance was accordingly reluctant, partial and confined to the minimum required. The mandate was quietly relaxed.
The measurable outcome, two decades on, is that the barcode remains overwhelmingly dominant for grocery and general retail, and that radio tagging has become genuinely large and successful in precisely the places where its specific advantages outweigh its costs: apparel, where items are individually valuable, individually variable in size and color, and where accurate stock knowledge translates directly into sales; logistics and returnable containers; libraries; access control; passports; livestock; and contactless payment, which is the same family of technology and is now universal.
So this is not a failed technology. It is a technology that found its correct level, which is a smaller and more specialized level than the forecast described, and it took about fifteen years to get there. The pattern should now be familiar: a capability is real, its promoters project it across the whole of a domain, the economics and the physics confine it to the fraction of the domain where it is genuinely superior, and the resulting industry is substantial and unglamorous and nobody writes about it.
The transferable question is the one about who pays and who benefits. A forecast that requires one party to bear a cost so that a different party can capture a benefit is not a technology forecast at all. It is a prediction about bargaining power, and it should be assessed as one.
Case Study Sixteen — The Water Under Punjab
In the mid-nineteen-sixties India was importing grain under emergency arrangements and serious people were predicting famine on a scale that would kill millions.
What happened instead is one of the genuine triumphs of the twentieth century. High-yielding dwarf wheat and rice varieties, developed through international breeding programmes, were introduced across northern India from 1966 onward, together with fertilizer, assured irrigation and guaranteed purchase prices. Punjab, a state occupying a small percentage of India’s land, became the country’s granary. Wheat production in the state multiplied several times over within two decades. India stopped importing grain and began exporting it. The predicted famine did not occur, and the forecasters who had predicted it were wrong in the most useful possible way.
The forecast attached to the new agriculture was that yields would rise, and yields rose. What was not forecast, and what has become the defining problem of the region, concerns the input that nobody was charged for.
The new varieties are more productive and also thirstier, and rice in particular requires standing water in a region where the monsoon supplies it for only part of the year. The gap was closed with tube wells drawing on groundwater. The state subsidized the electricity to run the pumps — in some periods supplying it free to agriculture — which meant that the marginal cost of extracting another metre of water was, to the individual farmer, approximately zero. The price of the crop was guaranteed. The price of the water was not merely low; it did not exist as a signal at all.
The predictable consequence has been unfolding for forty years and is now measured by satellite. Water tables across large parts of Punjab and neighboring states have been falling at rates of well under a metre a year in some districts and considerably more in others, and the depletion is visible from orbit as a measurable loss of mass. Shallow wells that served for generations have failed and been replaced by deeper ones, which requires more powerful pumps and more electricity, which the state also subsidizes. In some districts the water now being drawn is not being replaced on any timescale relevant to human affairs.
Secondary consequences have accumulated. The intensive wheat-and-rice rotation has depleted soil nutrients that the previous mixed cropping had maintained. Fertilizer application per unit of output has risen while yield growth has slowed. Punjab also produces a great deal of rice straw that must be cleared quickly to plant the next crop, and burning it is the cheapest method, which contributes materially to the severe air pollution of the northern Indian winter — an entirely separate crisis whose origin lies in a cropping pattern adopted for food security thirty years earlier.
Attempts at correction have been made and are instructive in their difficulty. A law requiring rice transplanting to be delayed until closer to the monsoon has measurably slowed the decline in some areas. Programmes paying farmers to switch crops, install metering, or adopt water-saving methods have had limited uptake, because the individual farmer’s calculation has not changed: rice has a guaranteed price and a guaranteed buyer, alternatives do not, and the water is still free.
This case is the boundary condition from the eighth chapter of this book, appearing in the field. Resource optimism is correct where a resource has an owner and a price. Every element of the Green Revolution that had a price behaved well: seed, fertilizer and machinery were bought, sold, improved and economized on. The one input with no price is the one being consumed past the point of recovery, and it is being consumed by people acting entirely rationally within the incentives they were given.
It also demonstrates something uncomfortable about successful interventions. The famine forecast was wrong and the people who acted to prevent it were right, and the action they took created a slower problem with a longer fuse that is now harder to solve than the original one would have been. This is not an argument that they should have acted differently. Millions of people are alive who would not be. It is an argument that the ledger on any large intervention has to be kept open for fifty years, and that almost nobody keeps a ledger open for fifty years.
Case Study Seventeen — The Fifth Generation
In 1982 Japan’s Ministry of International Trade and Industry announced a ten-year national programme to build computers that could reason, converse and translate, and much of the rest of the world panicked.
The context matters. Japanese industry had spent the previous fifteen years taking successive markets — cameras, motorcycles, consumer electronics, automobiles, memory chips — from Western incumbents who had believed themselves secure. When the same coordinating ministry announced that it was next going to build intelligent machines, the announcement was taken very seriously indeed. The United States formed a research consortium in response. Britain launched a national programme. The European Community launched another. Books appeared warning that the West was about to lose the computing industry.
The technical plan was specific. The machines would be built around logic programming, executing inference in parallel across specialized hardware, with knowledge represented as explicit rules and facts. The intellectual case was strong and mainstream: if intelligence consists of drawing conclusions from what is known, then a machine that draws conclusions very fast from a great deal of explicitly represented knowledge should be intelligent. The programme was well funded, staffed with capable people, and executed largely as designed.
It delivered the hardware. Parallel inference machines were built and worked. It delivered software, a substantial body of research, and a generation of trained researchers. What it did not deliver was anything resembling the applications promised — machines that understood language, translated reliably, or reasoned usefully about the ordinary world — and when the programme concluded in 1992 the assessment was widely and correctly that it had failed in its central objective.
The reason is the mechanism described earlier in this book as the intuition of imitation. The plan assumed that to build a machine that reasons, you should implement reasoning: write down the rules, represent the knowledge, execute the logic. This is how reasoning appears from the inside, and it is how a philosopher would describe it. The difficulty is that the quantity of unstated knowledge required to understand an ordinary sentence about an ordinary situation is vast, and nobody has ever succeeded in writing it down, and the attempts to do so consumed decades and enormous budgets in several countries.
What eventually worked came from a direction the programme explicitly did not take, and which was regarded at the time as unprincipled: statistical methods that infer the regularities from very large quantities of data without anybody stating them. Machine translation, for instance, improved more from counting patterns across large collections of already-translated text than from any effort to encode grammar. The joke among researchers in that period, that the quality of a language system improved every time a linguist left the team, was unkind and directionally accurate.
The results, then. A ten-year national programme, several imitative programmes abroad, a great deal of hardware that was obsolete before it was finished because general-purpose processors improved faster than the specialized machines could be designed, and a research field that entered a prolonged funding winter following the disappointment. The competitive threat that had triggered the Western response never materialized. And the capabilities promised in 1982 arrived roughly forty years later, by an approach the programme had considered and set aside.
Two lessons that apply directly to any national technology programme being announced this year. The first is that committing a decade of resources to one architecture is a bet on that architecture, and the history of computing offers very little support for the proposition that anyone can identify the winning architecture ten years out. The second is that the reaction of rival countries to such an announcement is not evidence that the plan is sound; it is evidence about the rivals’ anxiety, and in this case three continents spent substantial public money responding to a threat assessment that turned out to be wrong in the same way the original plan was wrong.
Case Study Eighteen — The Newton
Apple’s MessagePad, introduced in 1993, was the product that made handheld computing a public joke, and it contained essentially every idea that made handheld computing universal fifteen years later.
The device was a tablet a little larger than a hand, with a touch screen operated by a stylus, no keyboard, and software organized around an assistant that would interpret what the user wrote and act on it. Writing an instruction about a meeting was intended to create the appointment, find the person, and prepare the message. The operating system was, by the standards of 1993, genuinely advanced. The chief executive who backed it, John Sculley, coined the term personal digital assistant to describe the category.
The forecast was that this category would become large and would substantially replace paper organizers within a few years. The category did eventually become large beyond any projection ever made for it, and this particular product did not participate.
What went wrong was concentrated in one feature. The handwriting recognition, on which the entire interaction model depended, attempted to interpret ordinary cursive writing by matching against a dictionary, and at launch it did so unreliably. It also required training, and the reviews were written by people who had not trained it. A newspaper cartoon strip mocked it for weeks, running a sequence in which a written word was transcribed as something absurd, and the phrase entered the language. Later versions improved the recognition substantially — by most accounts to a genuinely usable standard — but by then the joke was fixed and the joke was the product’s public identity. The line was discontinued in 1998.
The competitor that succeeded took the opposite approach to exactly the same problem. Palm’s device, released in 1996, did not attempt to read ordinary handwriting at all. It required the user to learn a simplified alphabet of single strokes, which is a worse idea on paper and a far better one in practice, because a system that is ninety-nine percent accurate on a restricted input is more useful than one that is eighty percent accurate on an unrestricted one. It was also smaller, far cheaper, and did less. It sold in the millions.
The measurable outcome is that Apple spent a reported several hundred million dollars over the programme’s life, sold a modest number of units, and withdrew. Palm captured the category. And a decade later the category was itself absorbed by a device that solved the input problem by abandoning the stylus and the handwriting entirely in favor of a finger and a capacitive screen — an approach that had been impossible in 1993 because the screens did not exist.
Three transferable observations. The first is that when a product depends on a single capability, the forecast for the product is exactly the forecast for that capability, and everything else in the specification is decoration. The second is that a reduced version of a hard problem, solved reliably, generally defeats the full version solved unreliably — which is a design principle and also a forecasting principle, since the ambitious approach attracts the attention and the restricted one takes the market.
The third is about reputation, and it is the least comfortable. The technical deficiency was corrected within a couple of years. The reputational deficiency was never corrected, because a joke has no update mechanism. A product that becomes a punchline is finished regardless of what happens to it afterward, and the same is true of a forecast, which is why several of the predictions examined in this book remain synonyms for foolishness long after they came true.
Case Study Nineteen — Wash Your Hands
In the Vienna General Hospital in the eighteen-forties there were two maternity clinics, and women knew which one they wanted.
In the first clinic, staffed by physicians and medical students, the death rate from puerperal fever ran at something in the region of one mother in ten, and in bad months considerably worse. In the second, staffed by midwives, it ran at around one in twenty-five. The difference was public knowledge; women arriving in labor begged to be admitted to the second clinic, and some preferred to give birth in the street, which was known to be safer.
Ignaz Semmelweis, an assistant in the first clinic, was tormented by this and worked through the possible explanations systematically, discarding those that did not fit the pattern. The answer arrived through a death. A colleague was cut with a scalpel during an autopsy and died with symptoms indistinguishable from those of the women. Semmelweis concluded that the physicians were carrying something from the dissection room, which they visited before the ward, and that the midwives, who performed no autopsies, were not.
He instituted washing in a chlorinated lime solution before examinations. Mortality in the first clinic fell within a year to approximately the level of the second, and in some months to near zero. He then obtained the same result at a second institution.
The intervention was cheap, simple, and demonstrably effective, and it was rejected.
The reasons deserve stating precisely, because the comfortable modern version of this story — that his colleagues were stupid — is wrong and unhelpful. Semmelweis had no mechanism to offer; germ theory did not yet exist, and his explanation invoked an unidentified cadaverous material, which is not a satisfying causal account. He published slowly and badly, delaying a full statement of his findings for fourteen years, and when he did write he attacked his opponents in language that made it professionally impossible to concede to him. His theory also appeared to conflict with the prevailing framework, in which disease arose from imbalances particular to each patient, so that a single cause of a single condition was a category error rather than merely a wrong answer. And the claim carried an accusation: it said that respectable men had been killing their patients with their own hands, which is a thing almost no professional body has ever accepted quickly.
He was not reappointed in Vienna, returned to Hungary, obtained the same results again, was again ignored, deteriorated, was committed to an asylum in 1865 and died there within two weeks, from an infection contracted in the institution. Within two decades Pasteur and Lister had supplied the mechanism he lacked, and the practice became universal.
The measurable results are stark: an intervention available in 1847, adopted generally in the eighteen-eighties, with the intervening deaths falling on women who had no part in the argument. And the modern sequel is the part that makes this case belong in a book about the gap between knowing and doing. Hand hygiene compliance in hospitals today, where the mechanism is undisputed and taught to every student, has been repeatedly measured well below what anyone would consider acceptable, and the measures that improved it were not arguments but architecture — dispensers placed where the hand already goes, audits, and someone whose job is to watch.
The forecasting content is this. A correct finding does not propagate on its own. It propagates according to whether it has a mechanism, whether its author can be conceded to without humiliation, whether it fits the framework in use, and whether anybody’s job is to enforce it. Semmelweis had the evidence and none of the other four, and the evidence alone bought nothing. Anyone who believes that being right is sufficient should read the register of the first clinic.
Case Study Twenty — One Child
The policy that limited most Chinese families to a single child was designed by people with a background in systems engineering rather than demography, and this shows in every feature of it.
It was introduced in 1979 on the basis of projections indicating that population growth would overwhelm the country’s capacity to feed and employ itself, and it was argued for using the language of control theory: the population was a system with a target state, and births were the input variable through which the state could be steered. Enforcement was carried out through birth permits, workplace and community monitoring, fines calibrated to income, and in some places and periods coercive measures that the central authorities condemned and local officials nonetheless applied under quota pressure.
The forecast was that fertility would fall sharply as a result of the policy. Fertility did fall sharply, and here the honest analysis becomes more interesting than the standard account in either direction. The steepest decline in Chinese fertility occurred during the nineteen-seventies, before the one-child policy, under a less coercive campaign encouraging later marriage, longer spacing and fewer children, and against a background of rapidly falling child mortality and rising education — the same combination that has reduced fertility in every country where it has occurred, without compulsion. By 1979 the rate had already fallen by roughly half. The one-child policy was applied to a decline that was substantially under way, and comparison with neighboring countries that had no such policy suggests that a large part of what followed would have happened regardless.
What the policy did produce, unambiguously and without precedent, were the consequences nobody had modeled.
The first is the sex ratio. In a society with a strong preference for sons and a limit of one child, the arithmetic is brutal, and the arrival of ultrasound made it actionable. Reported sex ratios at birth rose far above the biological norm through the nineteen-nineties and two-thousands, producing a cohort of tens of millions of men who will not find a partner within their own population. The consequences of this are still unfolding and no country has previous experience of them.
The second is the age structure. A birth rate suppressed for a generation produces, thirty years later, a workforce that shrinks while the retired population grows, and the arithmetic is fixed once the cohort exists — nothing done afterward can change how many people were born in 1995. China’s working-age population began declining in the mid-twenty-tens, and the ratio of workers to retirees is deteriorating faster than it did in any country that industrialized earlier, with the additional difficulty that the transition is occurring at a lower level of income per head.
The third is that the policy could not be switched off. Restrictions were relaxed to two children in 2016 and to three in 2021, accompanied by subsidies, extended leave, and public exhortation. Births rose briefly and then continued falling, reaching successive record lows. The generation now of childbearing age was raised in single-child families, in expensive cities, with housing and education costs calibrated to the assumption of one child, and it is not responding to the incentives being offered. The lever that was believed to control the system turned out to operate in only one direction.
The results, therefore: a decline in fertility that was largely happening anyway, achieved at very great cost in coercion; a sex imbalance without historical parallel; an age structure that will constrain the country’s economy for decades; and a demonstrated inability to reverse any of it. And the lesson is the oldest one in this book, arriving at the largest scale it can reach. A system was modeled as a machine with an input and an output. It was in fact a population of people making decisions about their own lives, and once those decisions changed, they did not change back when the policy did.
Glossary
Adoption curve. The typical shape of how a new thing spreads: slow at first, then rapid, then slowing as the remaining holdouts turn out to be the hardest to convince.
Agglomeration. The economic advantage that comes from many people and firms being close together, chiefly because it makes finding the right match easier.
Alarm fatigue. The predictable result of too many false warnings: people who once responded stop responding, including to the real one.
Amplitude. How far a swinging system travels from its resting point. Large amplitude is what makes a temporary movement look like a permanent change.
Aquifer. An underground body of water-bearing rock. Often drawn on faster than it refills, and, having no owner and no price, subject to no self-correcting mechanism.
Arrow model. The habit of imagining the future as a straight extension of a current trend. Natural, intuitive, and responsible for most forecasting failure.
Automation. Replacing human labor within an activity with machinery. Frequently confused with changing the activity itself, which is a much larger and rarer event.
Backfire. The rare case where a rebound is so large that an efficiency improvement raises total consumption rather than lowering it.
Banknote. A physical token of money. Used less and less for payment while being held more and more as a store of value and an emergency fallback.
Base rate. How often something happens in general. Ignoring it is the commonest way of being confidently wrong about a specific case.
Basic income. A regular unconditional payment to individuals regardless of employment. Trials have generally found modest effects on working and substantial effects on stability and health.
Bias blind spot. The finding that people readily accept that human judgment is distorted in general while believing their own present judgment to be sound.
Bullshit job. A term for paid work its own holder believes to be pointless. The idea that much employment exists to satisfy the requirement that people be employed.
Bundling. The condition of one object performing several distinct jobs at once, as a banknote performs payment, storage, privacy and fallback.
Calibration. Knowing the shape of your own errors — whether you habitually overestimate speed, underestimate resistance, and so on. The only forecasting skill that reliably improves with age.
Cashless society. An economy in which payment is entirely electronic. Widely predicted, partly arrived, and now resisted on grounds of resilience and inclusion.
Catharsis hypothesis. The belief that expressing an emotion drains it. Experimentally unsupported: rehearsing anger tends to increase subsequent aggression rather than reduce it.
Central bank digital currency. State-issued electronic money. Revives, in a new form, every question about surveillance and programmable spending that cash quietly avoided.
Chlorofluorocarbon. A family of industrial gases, once used in refrigeration and aerosols, identified in 1974 as a destroyer of stratospheric ozone and phased out by international agreement.
Cobweb cycle. The oscillation produced when producers respond to today’s price with goods that take time to grow, so that supply always arrives to meet a price that has since changed.
Cognitive dissonance. The discomfort of holding a belief that the evidence contradicts, and the tendency to resolve it by revising the evidence rather than the belief.
Cohort replacement. The mechanism by which social attitudes usually change: not by persuading the existing population but by the older group being replaced by a differently raised one.
Commons. A resource used by many and owned by none. The domain in which price signals do not operate and in which resource warnings have most often proved correct.
Compound growth. Growth in which each increase becomes part of the base for the next. Very difficult to picture accurately, which is why arguments built on it are so persuasive and so treacherous.
Confirmation bias. The tendency to notice and retain evidence that supports what one already believes and to overlook the rest.
Contact hypothesis. The finding that contact between groups reduces hostility only under specific conditions — rough equality, a shared goal, cooperation, and institutional support — and can increase it otherwise.
Cornucopianism. The confidence that human ingenuity will always outrun scarcity. Correct where a resource has an owner and a price; unfounded where it does not.
Counterfactual. What would have happened otherwise. Impossible to observe directly, which is why successful prevention is so hard to distinguish from an unnecessary alarm.
Credential inflation. The tendency of qualification requirements to rise over time without a corresponding rise in the difficulty of the work.
Cryonics. The practice of freezing a body after death in the expectation that future medicine will repair and revive it. Begun in the nineteen-sixties; no revival has occurred.
Damping. The loss of energy that stops a swinging system from returning exactly to where it began. Why a rebound never restores the original state.
Declared rule. A stated principle with nothing attached to it. Distinguished from a binding rule, which has a consequence for breaking it.
Delay. The interval between an action and the response it provokes. The single variable that most determines whether a system adjusts smoothly or swings wildly.
Demographic transition. The shift, observed in nearly every society that has become wealthier and more urban, from high birth and death rates to low ones.
Demonetization. The formal withdrawal of legal tender status from a currency or denomination.
Diffusion. The spread of a practice or technology through a population, generally far slower than its inventors expect.
Direct rebound. The portion of an efficiency saving consumed by doing more of the same activity, as when a cheaper car to run is driven further.
Disclosure effect. The finding that people reveal more, and conceal less, when they believe they are being interviewed by a machine rather than by a person.
Disconfirming event. An observation specified in advance that would show a prediction to be wrong. Rarely stated, and the absence of one is what makes most forecasts unscoreable.
Doomsday cult. A group organized around a dated prophecy of catastrophe. Studied chiefly for what its members do after the date passes, which is to revise rather than abandon.
Efficiency. Getting more output from the same input. Reliably beneficial and reliably less beneficial than the engineering calculation suggests, because behavior adjusts.
ELIZA effect. The tendency of people to attribute understanding and intention to a program whose workings they know to be simple.
Equilibrium. A state a system returns to after being disturbed. Its existence is what makes a restoring force possible.
Eradication. The permanent removal of a disease from the world. Achieved for smallpox, and a genuine example of a change that does not reverse.
Escape hatch. Any device that allows a failed prediction to avoid being scored: a moving date, a retreat to scenario, an unstated condition, or a claim that it is already happening.
Escape velocity. The claim that medical progress will soon extend life by more than a year for each year that passes, after which nobody alive need die.
Expert judgment. Forecasting by qualified specialists. Studied systematically over two decades and found, on average, to perform at close to the level of an informed guess.
Exponential growth. Growth proportional to current size. Reliable over short spans, and almost never sustained over long ones, because something eventually pushes back.
Extrapolation. Extending an observed trend beyond the data. The core operation of forecasting and the source of most of its errors.
Fallback function. The job an old technology keeps because it works when everything else fails, as cash does in a power cut.
Falsifiability. The property of a claim that some observation could show it to be wrong. Without it, a prediction is a mood rather than a statement.
Feedback loop. An arrangement in which the output of a process affects the process itself, either amplifying it or opposing it.
Fertility rate. The average number of children born per woman. Falling across the developed world and resistant to every policy so far attempted.
Forecast. A statement about what will happen. Distinguished from a scenario, which explores what would happen under stated assumptions.
Friction. Everything that slows a change down: habit, cost, regulation, liability, and people who would rather not. Absent from most forecasts and decisive in most outcomes.
Furniture. A prediction that came true so completely that it stopped being noticed as a prediction. The main reason forecasting appears less accurate than it is.
Governor. A device that limits an engine’s speed using the speed itself. The earliest visible example of a restoring force built into a machine.
Green Revolution. The mid-twentieth-century transformation of agriculture through improved varieties, fertilizer and irrigation, which defeated a widely predicted famine.
Guano. Accumulated seabird droppings, the dominant fertilizer of the nineteenth century, whose exhaustion prompted a resource panic resolved by industrial chemistry.
Haber-Bosch process. The industrial method for producing ammonia from atmospheric nitrogen. The reason a resource crisis in fertilizer did not occur.
Hedgehog and fox. A distinction between thinkers who drive everything through one large idea and those who hold many small ones loosely. The second group forecasts better.
Hierarchy of needs. A model arranging human needs from physical survival up to self-realization. Useful for understanding individuals and misleading when applied to institutions.
Hindsight bias. The feeling, after an event, that it was obviously going to happen. The main obstacle to being fair to earlier forecasters.
Hoarding. Holding a store of something against future difficulty. Explains why the quantity of banknotes rose while their use in payment fell.
Homeostasis. The maintenance of a stable internal state despite external change. The biological version of a restoring force.
Household formation. The rate at which new separate households are created. Shifted dramatically by the rise of people living alone.
Hubbert curve. The roughly symmetrical rise and fall of extraction from a finite oil field. Accurate for a defined set of fields; misleading when extended indefinitely.
Hybrid work. An arrangement combining attendance and remote work. The settled outcome of a period in which both total attendance and total dispersal were predicted.
Hydraulic model. The nineteenth-century picture of the mind as a vessel under pressure whose drives must be released. Persuasive, influential, and unsupported by experiment.
Hype cycle. The pattern in which a new technology attracts excessive expectation, disappoints, and then becomes useful more quietly than promised.
Illustrative model. A deliberately simplified calculation offered to demonstrate a point rather than to predict an outcome. Almost never survives transmission as such.
Imagined community. A large group of strangers who understand themselves as belonging together, made possible by shared printed language.
Immortalism. The belief that indefinite human lifespan is technically achievable and near. A recurring position for at least four centuries.
Incentive asymmetry. The arrangement whereby predicting disaster carries little professional risk and predicting calm carries a great deal.
Infant mortality. Deaths in the first year of life. Its reduction accounts for most of the twentieth-century increase in average life expectancy.
Institution. A durable arrangement performing several social functions at once. Rarely abolished; usually demoted and specialized.
Intensive parenting. The modern practice of high supervision and high direct engagement with children, which increased even as families became smaller and less stable.
Intuition of imitation. The assumption that to reproduce something one should copy it. Wrong at nearly every important juncture in the history of technology.
Jevons paradox. The observation that making the use of a resource more efficient can increase rather than decrease the total amount consumed.
Kibbutz. An Israeli collective settlement, some of which raised children communally on principle and abandoned the practice within a few decades.
Lag. See delay. The gap that turns a corrective mechanism into an oscillation.
Le Chatelier’s principle. The rule that a system in balance responds to a disturbance by shifting in the direction that relieves it.
Ledger. A kept record of predictions and their outcomes. Almost entirely absent from public forecasting, which is why the field does not improve.
Legal tender. Money that must be accepted in settlement of a debt. A legal status distinct from whether anyone actually uses it.
Legibility. The degree to which an institution can see and record what individuals are doing. Greatly increased by digital records.
Legibility trap. The pattern by which a technology apparently empowering individuals against an institution empowers the institution more, because only the institution can analyze at scale.
Leisure. Time not spent working. Predicted to expand indefinitely with productivity; expanded until roughly the nineteen-seventies and then largely stopped.
Liability. Answerability for an outcome. Frequently the true constraint on automation in professional work, and rarely mentioned in forecasts about it.
Life expectancy. The average age at death for a given population. Roughly doubled in the twentieth century, chiefly by preventing early deaths rather than extending old age.
Limits to Growth. A 1972 report modeling the interaction of population, industry, food, pollution and resources, remembered mainly for an illustrative table it was not offering as prediction.
Malthusian trap. The condition in which population growth cancels any gain in food supply, keeping most people near subsistence. Accurate for all recorded history before about 1800.
Marchetti’s constant. The observation that people across many centuries and technologies have spent roughly an hour a day travelling, so faster transport buys distance rather than time.
Marienthal study. A 1930s investigation of an Austrian village after its factory closed, documenting that prolonged unemployment destroys the structure of time as much as it destroys income.
Matching. The process by which unusual skills find unusual requirements. The main reason dense cities persist despite the cost of living in them.
Maximum lifespan. The greatest documented human age, currently a hundred and twenty-two, unexceeded since 1997 despite enormous effort.
Mechanized substitution. The characteristic error of visual futurism: removing the human from an activity while leaving the activity, and the arrangement surrounding it, untouched.
Montreal Protocol. The 1987 agreement phasing out ozone-destroying chemicals. The clearest case of a warning that was believed, acted upon, and vindicated.
Moore’s law. The observation that the number of components on a chip has doubled at regular intervals. A description of an industry’s achievement, frequently mistaken for a law of nature.
Negative feedback. A response that opposes the change that produced it. The mechanism behind every restoring force in this book.
Negativity bias. The tendency of bad information to be processed faster, remembered longer and weighted more heavily than good information of equal size.
Nitrogen fixation. Converting atmospheric nitrogen into a form plants can use. Performed naturally by bacteria and industrially since the early twentieth century.
Not proven. A third verdict, borrowed from Scots law, for a claim on which the evidence is genuinely insufficient in either direction.
One-way door. A change that cannot be undone at any affordable cost. See ratchet.
Ornithopter. A flying machine that flaps. The natural first design for anyone imitating birds, and a comprehensive failure.
Overshoot. Movement past the point of balance, caused by a corrective response arriving too late.
Ozone layer. The band of stratospheric ozone that absorbs ultraviolet radiation, damaged by industrial gases and now recovering following international action.
Paperless office. The predicted disappearance of paper from workplaces. Announced in the nineteen-seventies, contradicted for thirty years, and quietly achieved after the joke had set.
Partial credit. The grade given to a prediction that identified a real phenomenon and got its scale badly wrong.
Path dependence. The condition in which what is possible now depends on choices made earlier, often for reasons no longer relevant.
Peak oil. The point of maximum extraction. Correctly predicted for the United States in 1956 for the year it occurred, and wrongly generalized into a permanent global decline.
Pending verdict. A claim genuinely undecided. Admissible only with a stated date and a stated condition for closing it, or it becomes a hiding place.
Pendulum model. The picture of social and economic systems as oscillating around a moving center rather than travelling in a straight line.
Physical limit. A constraint that no amount of investment relaxes, such as the energy available in a quantity of fuel. Distinguished from a cost, which falls.
Picturephone. A commercial visual telephone demonstrated in 1964 and comprehensively unsuccessful, whose function arrived decades later by an entirely different route.
Plateau. The flattening of a curve that had appeared to be accelerating, generally arriving shortly after somebody has published a book about the acceleration.
Population bomb. A widely read 1968 warning of imminent mass famine from population growth, overtaken by agricultural change.
Positional good. Something valuable because others lack it. Immune to general prosperity, and therefore able to absorb surplus indefinitely.
Precautionary principle. The rule that action should be taken against a serious risk before the evidence is conclusive. Sound in principle and difficult to bound in practice.
Prediction market. A market in contracts paying out on future events. Effective for questions that resolve soon and cleanly, and useless for the questions this book concerns.
Prerequisite. The thing that must change before a predicted change can occur. Usually the slowest element and usually controlled by somebody outside the forecast.
Prevention defense. The claim that a warning did not fail but succeeded, by causing the action that averted the outcome. Sometimes true, always available, and therefore in need of a test.
Price signal. The information carried by a rising or falling price, which simultaneously funds conservation, substitution, exploration and recovery.
Print capitalism. The commercial printing of books in vernacular languages, which standardized those languages and helped manufacture national identities.
Productivity. Output per hour of work. Its twentieth-century increase was correctly forecast; the assumption that it would be taken as leisure was not.
Prophecy. A prediction delivered with the authority of conviction rather than evidence. Distinguished from a forecast mainly by the absence of a date.
Ratchet. A change that can go one way but not back, because the return journey is impossible or unaffordable.
Rebound effect. The portion of a saving from efficiency that is consumed by increased use. Usually partial, occasionally total.
Rectangularization. The reshaping of a survival curve so that more people reach the maximum age without the maximum itself increasing.
Rehearsal. A simulation of a satisfaction rather than the satisfaction itself. Trains an appetite instead of relieving it.
Remediation. Preventive repair carried out in anticipation of a failure, as before the year 2000 date change. Successful remediation is indistinguishable from an unnecessary panic.
Remote work. Employment conducted away from a shared workplace. Correctly forecast in mechanism and greatly overestimated in the proportion of work to which it applies.
Replacement level. The fertility rate at which a population sustains itself, roughly two point one children per woman.
Reserves. The quantity of a resource currently identified and worth extracting at present prices with present technology. An economic category, not a geological one.
Resilience. The capacity of a system to keep functioning when a component fails. The argument that has revived cash after its predicted demise.
Resources. The total quantity of a substance believed to exist, as distinct from reserves.
Restoring force. Whatever pulls a system back when it moves too far. Invisible while the movement is small, which is why forecasts omit it.
Retrofuturism. The study or enjoyment of how earlier periods imagined the future, chiefly valuable as evidence about those periods.
Reversal. A prediction that was tested and answered in the opposite direction, as distinct from one that merely failed to occur.
Risk transfer. The property of forecasting whereby the cost of error falls entirely on those who acted on the prediction rather than on those who made it.
Scenario. An exploration of what would follow from stated assumptions. Regularly published as a scenario and received as a prediction.
Scoring rule. A method of grading forecasts against outcomes, standard in weather prediction and essentially absent everywhere else.
Self-actualization. The realization of one’s full capacities, placed at the top of the hierarchy of needs. A concept of the mid-twentieth century that dates any text using it.
Self-defeating prophecy. A prediction that prevents its own fulfillment by changing behavior. Indistinguishable, afterward, from an error.
Self-fulfilling prophecy. A prediction that causes the thing it predicts, as a forecast of shortage causes hoarding.
Serendipity. Encountering something valuable that one was not looking for. The function of cities that networks have consistently failed to reproduce.
Settling time. How long a system takes to reach its new resting state after a disturbance. Determines what any given interval of hindsight can and cannot judge.
Shale. Dense rock containing oil and gas, extractable by techniques that became economic in the late two-thousands and overturned the peak oil forecast.
Simon-Ehrlich wager. A ten-year bet made in 1980 on the price of five metals, won by the optimist, and considerably less decisive than folklore records, since the outcome depended on the decade chosen.
Singularity. A hypothesized point at which technological change becomes too rapid for human comprehension. Notable for never carrying a falsification condition.
Skill. In forecasting, performance measured against a simple baseline such as chance or persistence. Weather forecasting has it; almost nothing else is measured for it.
Slow variable. The element in a system that changes least quickly and therefore governs the timing of everything else.
Sovereign individual. The idea that mobile, information-based wealth would free individuals from state taxation. Comprehensively reversed by the same technology’s usefulness to tax authorities.
Specialization. What happens to a technology that has been displaced: it retreats to the job it does best and survives there at reduced volume and higher meaning.
Static reserve index. Known reserves divided by current consumption. A crude illustrative figure regularly mistaken for an expiry date.
Status inversion. The process by which a thing becomes prestigious precisely because a cheaper automated version exists, as handmade goods did after the factory.
Stock and flow. The distinction between an accumulated quantity and a rate of change. Stocks such as trust and expertise are slow to build and fast to destroy.
Store of value. The function of holding purchasing power over time, which cash retained after losing its role in payment.
Sublimation. The redirection of a drive into a socially acceptable activity. A central idea of the hydraulic model and the basis of the failed prediction that virtual outlets would reduce real aggression.
Substitution. Replacing one thing with another that serves the same purpose. Succeeds when the substitute delivers the actual good and fails when it delivers a rehearsal of it.
Surplus. What is left over after needs are met. Historically spent on positional consumption rather than on leisure.
Survivorship bias. Drawing conclusions from the cases that remain visible while the failures have disappeared from the record.
Taboo. A prohibition maintained by shared feeling rather than enforcement. A stock, and therefore capable of being lost far faster than it was built.
Technocracy. Government by technical experts. The characteristic utopian form of the early twentieth century, in the way that psychological self-realization was of the late twentieth.
Technological determinism. The belief that a technology’s capabilities determine its social consequences. The single assumption most responsible for the failures catalogued in this book.
Telecommuting. An earlier term for remote work, current from the nineteen-seventies, accompanied by predictions of the emptying of cities.
Thick market. A market with many participants on both sides, allowing specialized matches that a thin market cannot support.
Threshold. The level of disturbance at which a restoring force begins to act. Below it, a trend appears unopposed.
Time-use survey. A study recording how populations actually spend their hours. The source of the finding that parents now spend more time with children than in the nineteen-sixties.
Transmission loss. The stripping of qualifications, ranges and conditions as a claim passes from research to press release to headline to memory.
Trend. A direction of movement observed over some interval. Real, and not by itself evidence about any interval that has not yet occurred.
Trust. Confidence between people and in institutions. A stock, accumulated over generations and capable of being destroyed in a season.
Turing test. A proposed test in which a machine is judged by whether its conversation is distinguishable from a person’s. The subject of one of the few public dated wagers on the future.
Unbundling. The separation of an object’s several functions when a new technology takes the one it performs best and leaves the rest behind.
Uncertainty. The honest condition of any statement about the future. Ordinarily expressed as a range, and ordinarily discarded before publication.
Utopia. An imagined ideal society. Most reliable as evidence about what its author’s own period found unbearable.
Vagueness contract. The tacit arrangement by which forecasts are phrased so that no outcome could refute them, allowing the accurate and the useless to be indistinguishable.
Verdict. The grade assigned to a past prediction. This book uses three: happened, reversed, and not proven.
Verification burden. The work required to confirm that a new capability is safe or correct. Growing faster than the capabilities themselves in regulated sectors.
Videotelephony. Speaking while seeing the other person. Predicted for over a century, commercially failed for decades, and now so ordinary that its arrival went unremarked.
Virtual personality. An invented online identity, whether operated by a person or generated by a program. A live concern well before the technology could support it.
Weather forecasting skill. The measurable and steadily improving accuracy of meteorological prediction. Achieved by daily public scoring rather than by superior intelligence.
Whale oil. The dominant lighting fuel of the early nineteenth century, displaced by petroleum products as its price rose.
Withering away of the state. The prediction, made in several traditions and repeatedly, that government would become unnecessary. Consistently contradicted by the state’s share of economic activity.
Working week. The number of hours worked in seven days. Fell steeply for a century, then largely stopped falling, defeating the most famous economic forecast of the twentieth century.
World3. The computer model underlying the 1972 limits report, concerned with long-run interactions rather than with the exhaustion dates for which it is remembered.
Y2K. The year 2000 date problem in computer systems, preceded by extensive documented remediation and followed by few failures, leaving its true severity permanently unresolvable.
Timeline
c. 380 BCE. Plato’s Republic proposes that the guardian class hold wives and children in common, the first sustained argument in Western thought for abolishing the family.
c. 350 BCE. Aristotle replies that property and affection held in common are cared for by nobody, establishing the oldest objection to utopian redesign.
1516. Thomas More publishes Utopia, which reorganizes property, labor and religion and leaves slavery in place, demonstrating that even radical imagination carries its era’s furniture.
c. 1450 to 1600. Printing spreads in vernacular languages, standardizing them and helping manufacture the national identities that later forecasters would repeatedly predict were about to dissolve.
1620. Francis Bacon argues that knowledge accumulates and that the future can therefore be better than the past, introducing progress as a forecastable quantity.
1627. Bacon’s New Atlantis imagines an institution devoted to systematic invention, including flying machines and submarines, and anticipates the research laboratory by three centuries.
1755. The Lisbon earthquake shatters confidence that the world is arranged for human benefit, and prediction begins to separate from providence.
1776. Adam Smith describes how prices coordinate the behavior of people who never meet, supplying the mechanism that would later defeat most resource forecasts.
1788. James Watt applies the centrifugal governor to the steam engine, producing the first widely used machine that regulates itself through negative feedback.
1798. Thomas Malthus publishes his essay on population anonymously, arguing that food supply cannot keep pace with reproduction. It is the best-supported prediction of its century and the most thoroughly overturned.
1830s to 1870s. The guano trade dominates world fertilizer, prompts the American Guano Islands Act of 1856, and creates the first modern panic about a resource running out.
1856. The Scots verdict of not proven, long established in criminal law, provides the model this book borrows for judging predictions still in progress.
1859. Charles Darwin acknowledges Malthus as the source of the struggle for existence, making a failed social forecast the foundation of a successful biological theory.
1863. Jules Verne writes Paris in the Twentieth Century, forecasting a city of glass towers, gas-driven vehicles and a communications network. It is rejected as implausible and published in 1994.
1865. William Stanley Jevons observes that more efficient steam engines increased rather than reduced Britain’s coal consumption, describing the rebound effect.
1870s. Cartoonists begin drawing the visual telephone, beginning the longest continuously predicted technology in the record.
1880s. The telegraph prompts widespread predictions that instantaneous communication will make war impossible.
1888. Edward Bellamy’s Looking Backward organizes the future as one vast efficient corporation, because the corporation was the most admired institution of its decade.
1899 to 1910. French artists produce the En L’An 2000 postcards, which mechanize the barber and the schoolteacher while leaving the barber’s shop and the classroom entirely intact.
1909 to 1913. Fritz Haber and Carl Bosch industrialize the fixation of atmospheric nitrogen, ending the fertilizer crisis by manufacturing the resource from the air.
1901 to 1914. H. G. Wells publishes his anticipations and his world-state utopias, establishing the technocratic future administered by competent engineers.
1928. Alexander Bogdanov, convinced that repeated blood transfusion rejuvenates the body, dies from one he administers to himself.
1930. John Maynard Keynes predicts a fifteen-hour working week for his grandchildren. His productivity arithmetic proves roughly right and his conclusion entirely wrong.
1933. The Marienthal study documents that prolonged unemployment destroys the structure of time as much as it destroys income, evidence that any proposal to abolish work must answer.
1936. The Soviet Union reverses its earlier campaign against the family, restricting divorce and rewarding large families, one of the fastest policy reversals of the century.
1946 onward. Early kibbutzim raise children communally on principle; the practice is dismantled over the following decades, largely at the insistence of mothers.
1950. Alan Turing proposes judging machine intelligence by conversation, creating the criterion around which artificial companionship would later be argued.
1954. Abraham Maslow’s hierarchy of needs is published in its influential form, later becoming the structural spine of the 2007 book examined here.
1956. M. King Hubbert predicts that United States oil production will peak between 1965 and 1970. It peaks in 1970, the cleanest dated forecasting success in this book.
1956. Psychologists publish their study of a group whose doomsday prophecy failed, establishing the template by which refuted predictions are converted into pending ones.
1956. The Aerocar receives aviation certification. Six are built and the flying car enters permanent residence in the museum of ridicule.
1961. A rocket belt is publicly demonstrated. It flies for about twenty seconds, which remains the limit, because the constraint is the energy stored in the fuel.
1964. The Picturephone is demonstrated at the New York World’s Fair and fails commercially, chiefly because many people making telephone calls did not wish to be seen.
1964. Robert Ettinger’s book launches the modern cryonics movement; the first person is frozen in 1967.
1965. Gordon Moore observes the regular doubling of components on a chip, a description of industrial achievement that will be treated for sixty years as a law of nature.
1966. Joseph Weizenbaum writes ELIZA and is disturbed to find that people who know exactly what it is confide in it anyway.
1968. Paul Ehrlich’s The Population Bomb forecasts imminent mass famine, and is overtaken within a decade by agricultural change already under way.
1970s. Agricultural transformation through improved varieties, fertilizer and irrigation averts the predicted famines, a success rarely credited to anybody by name.
1972. The Limits to Growth is published, including an illustrative table of resource exhaustion dates that will outlive and obscure everything else in the report.
1974. Mario Molina and F. Sherwood Rowland publish the hypothesis that chlorofluorocarbons destroy stratospheric ozone. Industry responds that it is only a hypothesis.
1975. An American business magazine announces the office of the future in which paper declines. Office paper consumption then rises for roughly three decades.
1977. Fred Hirsch’s work on positional goods explains why general prosperity cannot satisfy the demand for relative standing, and why the surplus was never taken as leisure.
1977. The last naturally occurring case of smallpox is recorded in Somalia; eradication is declared in 1980, a change that will not reverse.
1980. Julian Simon and Paul Ehrlich stake a ten-year wager on the price of five metals. Simon wins, and the result is treated as far more decisive than a single decade can support.
1980. Alvin Toffler’s electronic cottage predicts that networked work will return the household to the center of economic life and empty the commuting city.
1983. Ruth Schwartz Cowan documents that labor-saving household technology did not reduce domestic hours, because standards rose to absorb the saving.
1985. A British Antarctic Survey team reports a seasonal collapse in ozone so severe they initially doubt their instruments.
1987. The Montreal Protocol is signed and rapidly ratified, becoming the clearest case in history of a warning believed, acted upon, and vindicated.
1988 onward. Global scientific assessment of climate change begins, and with it the modern argument about how confidence intervals survive transmission to the public.
1990s. The Human Development and demographic literature confirms the fertility decline across widely different cultures and political systems, reversing the Malthusian assumption that prosperity increases births.
1994. Cesare Marchetti describes the roughly constant daily travel-time budget, explaining why faster transport expands cities rather than dissolving them.
1997. Jeanne Calment dies at a hundred and twenty-two, setting a documented maximum that has not been approached since.
1997. James Dale Davidson and William Rees-Mogg publish The Sovereign Individual, forecasting that mobile information wealth will end the taxing state.
1997. Frances Cairncross announces the death of distance, the most confident statement of the case that location would cease to matter.
1999 to 2000. Extensive documented remediation precedes the year 2000 date change, followed by few failures, leaving its true severity permanently unresolvable.
2001. Abigail Sellen and Richard Harper publish a study of the myth of the paperless office, finding that electronic mail increased paper consumption in the organizations they observed.
2002. The Long Now Foundation establishes a public register of dated wagers about the future, including the Kurzweil-Kapor bet on the Turing test by 2029. Uptake is negligible.
2002. Psychologists document the bias blind spot: people accept that judgment in general is distorted while exempting their own.
2005. Philip Tetlock publishes two decades of collected expert forecasts, finding average performance close to chance and confident single-idea thinkers performing worst.
2007. The book examined in these pages is printed by a small press in Florida, written in Ontario, and enters the record unrevised.
2008 onward. Techniques for extracting oil from dense rock become economic in the United States, which passes its 1970 peak and overturns the global peak oil doctrine.
2010 to 2017. International agreements on automatic exchange of financial account information effectively end traditional banking secrecy, reversing the sovereign individual forecast.
2013 onward. Global demand for printing and writing paper falls steadily, quietly delivering the paperless office roughly thirty-five years after it was announced.
2014. Controlled studies find that people disclose more, and show less fear of judgment, when interviewed by an evidently artificial interlocutor than by a human being.
2017 to 2019. Unconditional income trials in several countries report modest effects on employment and substantial improvements in health, stress and the capacity to plan.
2020 to 2021. A pandemic runs the remote work and home schooling experiment on the entire population without consent. Schools reopen; cities briefly empty.
2020 onward. Banknotes in circulation rise in the major currencies even as cash payment collapses, splitting money’s functions and refuting the simple cashless forecast.
2021. More than a hundred and thirty jurisdictions agree in principle to a minimum rate of corporate taxation, ending the tax competition predicted to be irresistible.
2022 onward. Several northern European states issue household preparedness advice recommending that physical cash be kept at home, and cities and states legislate to require its acceptance.
2023. Rents in the major metropolitan centers return to record levels, closing the shortest and most complete reversal in this book.
2024 to 2026. Conversational machines become an ordinary daily technology, fulfilling the 2007 prediction about disclosure and doing so by an architecture its author had described elsewhere and never connected.
2026. This book is written, its predictions are dated, and the ledger passes to whoever is holding it in 2044.
Literature
Anderson, Benedict. Imagined Communities. London: Verso, 1983. The argument that printing in vernacular languages manufactured national identity, which is the historical answer to every forecast that communication technology dissolves nations.
Allport, Gordon W. The Nature of Prejudice. Cambridge: Addison-Wesley, 1954. The source of the conditions under which contact between groups reduces hostility, and by implication the conditions under which it does not.
Bellamy, Edward. Looking Backward: 2000–1887. Boston: Ticknor, 1888. The most widely read American utopia, and the clearest case of a future imagined as the most admired institution of its own decade enlarged to cover everything.
Bushman, Brad J. Does Venting Anger Feed or Extinguish the Flame? Personality and Social Psychology Bulletin, 2002. The experimental refutation of the catharsis hypothesis.
Cairncross, Frances. The Death of Distance. Boston: Harvard Business School Press, 1997. The most confident statement of the case that location would cease to matter, useful now chiefly as a dated specimen.
Carson, Rachel. Silent Spring. Boston: Houghton Mifflin, 1962. The book that brought bioaccumulation to public attention, and a case study in what happens to a correct warning with a well-funded opponent.
Cowan, Ruth Schwartz. More Work for Mother. New York: Basic Books, 1983. Documents that labor-saving domestic technology did not reduce domestic labor, because standards rose to absorb the saving.
Davidson, James Dale, and William Rees-Mogg. The Sovereign Individual. New York: Simon and Schuster, 1997. The forecast that mobile information wealth would end the taxing state, quoted approvingly in the book examined here and comprehensively reversed since.
Ehrlich, Paul R. The Population Bomb. New York: Ballantine, 1968. The most famous Malthusian forecast of the twentieth century, and the origin of one of the few public dated wagers in the field.
Ettinger, Robert C. W. The Prospect of Immortality. New York: Doubleday, 1964. The founding text of modern cryonics.
Festinger, Leon, Henry W. Riecken, and Stanley Schachter. When Prophecy Fails. Minneapolis: University of Minnesota Press, 1956. The study of what a group does after its dated prophecy does not occur, and the template for every subsequent escape hatch.
Gawande, Atul. The Checklist Manifesto. New York: Metropolitan Books, 2009. Argues that failure in expert work is usually a failure of execution rather than of knowledge, which is the distinction between a declared rule and a binding one.
Graeber, David. Bullshit Jobs: A Theory. New York: Simon and Schuster, 2018. Arrives from the political left at a conclusion closely resembling the one reached in the book examined here from an entirely different direction.
Hirsch, Fred. Social Limits to Growth. Cambridge: Harvard University Press, 1977. Explains why general prosperity cannot satisfy demand for relative standing, and therefore why the productivity surplus was never taken as leisure.
Hubbert, M. King. Nuclear Energy and the Fossil Fuels. Shell Development Company, 1956. The paper containing the correct dated prediction of the United States oil peak, and the method that was subsequently overextended.
Jahoda, Marie, Paul F. Lazarsfeld, and Hans Zeisel. Marienthal: The Sociography of an Unemployed Community. Original German edition 1933. The evidence that unemployment destroys the structure of time as much as it destroys income.
Jevons, William Stanley. The Coal Question. London: Macmillan, 1865. Contains both a failed national resource forecast and the correct identification of the rebound effect, in the same volume.
Keynes, John Maynard. Economic Possibilities for our Grandchildren. 1930. Correct about compounding productivity, wrong about what the surplus would be spent on.
Kriger, Bruce. A Future Beyond Imagination. Coral Springs: Llumina Press, 2007. The primary source examined throughout this book: an unrevised, dated, wide-ranging forecast by an unaffiliated author, and the reason the present volume exists.
Kriger, Peter. When the Last Straw Is Small: Cascading Failure under Clustered Multi-Vector Shocks in Networked Societies. IIIR Computational Humanities and Cultural Systems, 2026. DOI 10.5281/zenodo.21405843. On why collapse deadlines are set by the number of simultaneous shocks rather than their size.
Kriger, Peter. Latency and the Unwanted War. IIIR Computational Humanities and Cultural Systems, 2026. DOI 10.5281/zenodo.21568121. On decision delay in institutional systems, which is the same variable that governs the oscillations described here.
Malthus, Thomas Robert. An Essay on the Principle of Population. London, 1798, with substantial revisions in later editions. The best-supported prediction of its century and the most thoroughly overturned; the later editions are more careful than the reputation.
Marchetti, Cesare. Anthropological Invariants in Travel Behavior. Technological Forecasting and Social Change, 1994. The constant daily travel-time budget, and the reason faster transport enlarges cities rather than dissolving them.
Maslow, Abraham H. Motivation and Personality. New York: Harper, 1954. The hierarchy of needs, which is the structural spine of the book examined here and the source of its characteristic blindness to institutions.
Meadows, Donella H., and colleagues. The Limits to Growth. New York: Universe Books, 1972. Remembered for an illustrative table it did not offer as prediction, which is itself the most useful lesson in the volume.
More, Thomas. Utopia. 1516. Redesigns property, labor and religion while leaving slavery in place, demonstrating that radical imagination carries its era’s furniture unexamined.
Nordhaus, William D. Do Real-Output and Real-Wage Measures Capture Reality? The History of Lighting Suggests Not. In The Economics of New Goods, University of Chicago Press, 1996. The long price history of light, and the clearest single demonstration of demand expanding to absorb efficiency.
Ostrom, Elinor. Governing the Commons. Cambridge University Press, 1990. On the conditions under which shared resources are successfully managed, which is the boundary condition on every optimistic resource forecast.
Patterson, Clair C. Contaminated and Natural Lead Environments of Man. Archives of Environmental Health, 1965. The paper establishing that the human baseline had itself been contaminated, and the origin of the leaded fuel phase-out.
Popper, Karl. The Poverty of Historicism. London: Routledge, 1957. The argument that large-scale historical prediction is impossible in principle, which every forecaster should read and almost none do.
Pronin, Emily, Daniel Y. Lin, and Lee Ross. The Bias Blind Spot. Personality and Social Psychology Bulletin, 2002. People accept that judgment in general is distorted and exempt their own.
Rogers, Everett M. Diffusion of Innovations. New York: Free Press, 1962. The standard account of how new practices spread, and of why the spread is slower than inventors expect.
Rosling, Hans. Factfulness. London: Sceptre, 2018. A corrective to systematic pessimism about measurable global trends, to be read alongside rather than instead of the warnings it corrects.
Scott, James C. Seeing Like a State. New Haven: Yale University Press, 1998. On legibility as the precondition of state power, which is why digital records strengthened the state that was predicted to wither.
Sellen, Abigail J., and Richard H. R. Harper. The Myth of the Paperless Office. Cambridge: MIT Press, 2001. Field evidence that electronic mail increased paper consumption, and an account of what paper is actually good at.
Simon, Julian L. The Ultimate Resource. Princeton University Press, 1981. The cornucopian case, correct where resources have owners and prices and silent where they do not.
Smil, Vaclav. Enriching the Earth. Cambridge: MIT Press, 2001. The history of nitrogen fixation, and the best single account of a resource crisis solved by manufacturing the resource.
Smil, Vaclav. Energy and Civilization: A History. Cambridge: MIT Press, 2017. The long record of energy transitions, all of which took longer than their advocates predicted.
Tainter, Joseph A. The Collapse of Complex Societies. Cambridge University Press, 1988. On declining returns to complexity, which is the mechanism behind several of the institutional failures described here.
Tetlock, Philip E. Expert Political Judgment. Princeton University Press, 2005. Two decades of collected and scored forecasts, and the finding that confident single-idea thinkers perform worst.
Tetlock, Philip E., and Dan Gardner. Superforecasting. New York: Crown, 2015. What the small minority who forecast well actually do, which is chiefly to update in small increments and keep score.
Toffler, Alvin. The Third Wave. New York: Morrow, 1980. The electronic cottage, and the most influential statement of the prediction that networked work would empty the commuting city.
Turner, Graham M. A Comparison of The Limits to Growth with Thirty Years of Reality. Global Environmental Change, 2008. An honest attempt to score a famous model against subsequent data, and a rare example of anybody bothering.
Vogel, Ezra F. Japan as Number One. Cambridge: Harvard University Press, 1979. Included as a specimen: an intelligent, well-informed extrapolation of a trend that reversed within a decade of publication.
Weizenbaum, Joseph. Computer Power and Human Reason. San Francisco: Freeman, 1976. The objection that certain tasks should not be delegated to machinery regardless of how well it performs them, by the man who built the first program people confided in.
Wells, H. G. Anticipations of the Reaction of Mechanical and Scientific Progress upon Human Life and Thought. London: Chapman and Hall, 1901. The founding document of technocratic futurism, accurate about roads and traffic and disastrous about people.
Winner, Langdon. Autonomous Technology. Cambridge: MIT Press, 1977. A sustained argument against technological determinism, which is the assumption most responsible for the failures catalogued in this book.
