Latest / Collective Intelligence: Why groups get smart, strange and stuck / Tulips, bubbles and witches
Transcript
- 0:00Hi, I’m Niccolò Pescetelli, and welcome to Collective Intelligence: why groups get smart, strange and stuck.
- 0:07Quick news: short video episodes are on the way, made by Georgina Denis, to sit alongside these long ones.
- 0:14Same show, half an hour shorter.
- 0:16Right. Today, a story you may already know.
- 0:20It is Holland, in the 1630s, and a merchant is doing very well.
- 0:24A ship has come in from the eastern Mediterranean loaded with silks and velvets, and a sailor turns up at the counting house to tell him so.
- 0:34The merchant is pleased, and in a generous mood he gives the sailor a red herring for his breakfast.
- 0:41The sailor is fond of onions.
- 0:42And lying on the counter, among the bales of cloth, there is something that looks very like an onion.
- 0:48He pockets it.
- 0:49He goes down to the quay, sits on a coil of rope, and eats it with his herring.
- 0:56It is not an onion.
- 0:57It is a tulip bulb, a rare one, and the merchant goes through the building in a panic looking for it.
- 1:04By the time they find the sailor he is finishing the last of it.
- 1:08That breakfast, the story goes, cost as much as it would have cost to feed a ship's entire crew for a year.
- 1:16I love this story.
- 1:17It has everything.
- 1:18A precious object mistaken for a cheap one.
- 1:21A common man accidentally eating a fortune.
- 1:25And behind it, a whole society that had lost its mind so thoroughly that a flower bulb could be worth more than a house.
- 1:33That is the tulip mania, and it is the founding example in the literature of crowd madness.
- 1:39It is where every history of financial bubbles begins.
- 1:43It is the reason the phrase "tulip mania" still gets reached for whenever a price does something frightening.
- 1:50The story of the sailor comes from a book published in London in 1841 by a journalist called Charles Mackay.
- 1:57The book is called Extraordinary Popular Delusions and the Madness of Crowds, and its title has outlived almost everything else about it.
- 2:06I still remember when I bought this book, years ago.
- 2:09I was a few weeks into an internship at UCL with Bahador Bahrami, whose work on how two people pool their judgements turns up later in this series.
- 2:19After he mentioned it to me, I went straight up to the Waterstones across from the UCL campus and bought it.
- 2:26A book promising that crowds go mad felt like the right thing to be carrying around while learning how they get things right.
- 2:34That is the pleasure of this book.
- 2:36It is also, as I'll explain later, the problem with it.
- 2:41Now, here is the first thing I want you to notice.
- 2:43Mackay did not see this happen.
- 2:45He was writing two hundred years later.
- 2:48And he does not claim to have found it in any record of the time.
- 2:52He says he took it from a travel book, written by someone else, who got it from somewhere else again.
- 3:00And the second thing is the one that will take the rest of this episode to unpack.
- 3:06When historians went back to the Dutch archives and looked for the tulip mania, they found something much smaller and much stranger than the story.
- 3:15Very few people involved.
- 3:16Nobody ruined. No economic collapse.
- 3:21What they found instead was a trail.
- 3:23A set of songs and pamphlets written after the event, mocking the people involved, copied from writer to writer across two centuries,
- 3:32getting more lurid every time, until they arrived on Mackay's desk and went out into the world as history.
- 3:41So the most famous example of a crowd losing its mind may itself be an example of something else.
- 3:48A story spreading through a crowd of writers, each one repeating the last, none of them checking.
- 3:55That is the thread I want to pull today.
- 3:58Last episode I ended on Condorcet, and on a century of writers who were certain that a crowd that talks to itself goes mad.
- 4:05Mackay is the first of them, and I want to take him seriously, which means doing two things at once.
- 4:11Asking what he got right about crowds.
- 4:14And asking how a book about the madness of crowds came to be so thoroughly shaped by the thing it describes.
- 4:23So what kind of book is this?
- 4:25It is enormous, for a start.
- 4:27Three volumes, close to three hundred thousand words.
- 4:30And it is less an argument than a museum.
- 4:33Mackay goes room by room through the follies of European history and puts each one in a glass case.
- 4:40The first volume is what he calls national delusions: the great financial manias, including the tulips and the South Sea Bubble, then a run of smaller exhibits.
- 4:49The cult of relics.
- 4:51Prophecies of the end of the world.
- 4:53The way politics and religion kept dictating how men wore their beards.
- 4:58Duels. The popular worship of highwaymen.
- 5:03The second volume goes darker.
- 5:05The Crusades. The witch mania.
- 5:07A wave of poisonings.
- 5:09Haunted houses. The third is about what he calls philosophical delusions: alchemists who spent their lives trying to make gold,
- 5:19fortune-tellers, and the magnetisers, the followers of Mesmer.
- 5:24I will come back to the magnetisers in a future episode, because something happened there that I think is among the most important moments in this whole story.
- 5:34Today I am leaving them alone.
- 5:36What holds the museum together is one idea, and Mackay states it in his preface in a sentence that has been quoted ever since.
- 5:45Men, he says, think in herds, and it will be seen that they go mad in herds, while they only recover their senses slowly, and one by one.
- 5:56The usual reading is simply that crowds are stupid.
- 6:00That when people gather they become worse than they are alone.
- 6:06And some of Mackay's book does say that, loudly and repeatedly.
- 6:10But look at the shape.
- 6:12The going mad happens together, in a herd.
- 6:16The recovery happens separately, one person at a time.
- 6:21That is an asymmetry, and it is a very specific one.
- 6:25It says that whatever makes a crowd lose its judgement is something that travels between people.
- 6:31And whatever brings judgement back is something each person has to do alone.
- 6:37Now put that next to the last two episodes.
- 6:40Galton's ox worked because every guess was made alone.
- 6:44Condorcet's theorem works because every vote is cast alone.
- 6:49In both cases, the magic of the crowd comes from people's errors being independent, so that they cancel out.
- 6:55And in both cases I spent a long time on what happens when independence fails.
- 7:02When one loud man in the queue says eleven hundred, and then forty people drift towards his number.
- 7:08Mackay is describing that failure, over and over, across four centuries of European history.
- 7:14He did not have the mathematics.
- 7:16He did not have the word correlation.
- 7:19But the sentence about herds is, I think, the correct observation.
- 7:24The madness is something that happens between people, not inside them.
- 7:29Which raises a question that Mackay never quite asks, and which is going to organise the rest of this episode.
- 7:37If the problem is between people, is it actually madness?
- 7:42Because madness implies that the individuals have stopped reasoning.
- 7:46That each person in the crowd has become a fool.
- 7:49But there is another possibility, which is that each person in the crowd is reasoning perfectly sensibly,
- 7:55given what they can see, and that the crowd as a whole still ends up somewhere absurd.
- 8:03Those are very different diagnoses.
- 8:05They predict different things.
- 8:07And they suggest very different cures.
- 8:11If the problem is foolish individuals, you fix it with education, and that takes you straight back to Condorcet.
- 8:18If the problem is sensible individuals arranged the wrong way, then education will not help at all, and you need to change that arrangement.
- 8:28I think the evidence points more towards the second.
- 8:31But to get there I need to show you Mackay's two best cases, which are the South Sea Bubble and the tulips.
- 8:39And then I need to show you what happened when someone finally worked out the mathematics of what he was describing, a hundred and fifty years after he wrote it down.
- 8:50Okay. Two bubbles, one of them is real.
- 8:54Starting in London, in 1720.
- 8:57The South Sea Company had been granted a monopoly on British trade to the South Seas, and its name comes from that grant.
- 9:07But what the company really had by 1720 was a scheme.
- 9:12The British government was carrying a large national debt,
- 9:15and the company offered to take a great slice of it off the government's hands, converting people's government debt into company shares.
- 9:26For that to work out for everybody, the shares needed to rise.
- 9:30So the company did everything it could to make them rise, and for a while it worked spectacularly.
- 9:37Now watch what happens around it, because this is the part Mackay is best at.
- 9:42When the shares take off, everyone who owns them looks rich.
- 9:46Everyone who does not own them feels stupid.
- 9:50And suddently, a crowd of promoters appears, selling the second group a way to catch up.
- 9:56Mackay lists their schemes with obvious delight.
- 9:59One was to build a wheel for perpetual motion, and asked for a million pounds of capital.
- 10:06Another promised to improve the breeding of horses in England.
- 10:10Dozens more lived for a few weeks and then burst.
- 10:14They were called bubbles, which is where the word comes from.
- 10:17And underneath all of it, feeding it, there were rumours.
- 10:21While the company's directors, and above all its chairman, Sir John Blunt, did everything they could to lift the share price,
- 10:29stories went round London that Spain was about to open its colonies to English trade,
- 10:35and that silver from the great mines of South America would pour into England until it was as common as iron.
- 10:44Mexico would empty its gold mines in exchange for English wool.
- 10:48None of this was happening.
- 10:50But notice what a rumour is, in the terms of this series.
- 10:54It is a belief that has become visible without the evidence behind it.
- 10:59You hear that people are saying Spain is about to open its trade.
- 11:03You do not hear why they are saying it, or whether anybody who says it knows anything.
- 11:09And the more people you hear saying it, the more it sounds like knowledge.
- 11:15And then there is the one Mackay calls the most absurd of all.
- 11:19A company, as Mackay gives its title, for carrying on an undertaking of great advantage, but nobody to know what it is.
- 11:29This one is my favourite thing in the whole book.
- 11:31An unknown promoter opens an office in the City one morning.
- 11:36He asks for a small deposit on each share, promising to reveal the undertaking later.
- 11:42By the time he closes the door in the afternoon, a thousand people have paid.
- 11:47He has made two thousand pounds in five hours.
- 11:50And that same evening he leaves for the Continent and is never heard from again.
- 11:56Mackay vouches for this by saying it was reported by scores of credible witnesses.
- 12:05I want you to hold on to that phrase, because it is his standard of evidence all the way through the book, which is pretty ironic, right?
- 12:14And London was not even the first.
- 12:17Across the Channel, in 1719 and 1720, a Scottish financier called John Law, born in Edinburgh,
- 12:24had talked the French state into letting him run its finances, and built a company around the development of Louisiana and the Mississippi.
- 12:34When Law offered fifty thousand new shares, Mackay says, around three hundred thousand people applied for them.
- 12:42His house was in a narrow Paris street, the rue Quincampoix, and the street became a trading floor.
- 12:51Dukes and countesses stood in it for hours waiting to learn whether they had been allotted shares.
- 12:58Rents in the street multiplied many times over.
- 13:01A cobbler made a small fortune renting out his stall as a desk.
- 13:07And Mackay passes on a story, and he is careful to call it a story, of a hunchback who earned good money lending out his hump for people to write their contracts on.
- 13:18That one, notice, Mackay does flag as hearsay.
- 13:22It is one of the few places in the book where he does.
- 13:26Law's scheme collapsed in 1720 as well.
- 13:29So did the South Sea shares, by the autumn, and a great many people in both countries were ruined.
- 13:36That part is not in dispute.
- 13:38The two great bubbles of 1720 were real.
- 13:42And there is one ruined investor in particular whom everybody remembers, although he is not in Mackay's chapter at all.
- 13:49Isaac Newton. The story goes like this.
- 13:53Newton buys early, sells sensibly, and makes a solid profit.
- 13:57Then he watches his friends get richer as the shares keep climbing, cannot stand it, buys back in near the top, and loses a fortune.
- 14:08And when someone asks him about it afterwards, he says he could calculate the motions of the heavenly bodies, but not the madness of people.
- 14:19In 2019 the mathematician Andrew Odlyzko went through everything that survives about Newton's money, and his verdict is interesting.
- 14:30The folk tale is almost certainly true in outline.
- 14:34Newton did make money early and then lose heavily.
- 14:38But it rests on very few pieces of hard evidence, and some of those have been misquoted for centuries.
- 14:46So what about the famous line?
- 14:48The earliest version anyone has found appears in 1820, nearly a century after Newton died, reported second-hand.
- 14:56It says only that he could not calculate the madness of the people.
- 15:01The heavenly bodies are not there.
- 15:03Somebody added them later, because they make the sentence better.
- 15:08So even the most famous remark about the madness of crowds was improved in the retelling, by a crowd.
- 15:15Now the tulips.
- 15:17Mackay's account goes roughly like this.
- 15:20In the 1630s the tulip becomes the fashionable flower of the Dutch rich.
- 15:26Rare varieties, especially the striped and feathered ones, become enormously valuable.
- 15:32Bulbs are sold by weight in tiny units.
- 15:35A single bulb of the most famous variety is quoted at five and a half thousand florins.
- 15:41By 1636 there are tulip markets on the stock exchanges.
- 15:47And the whole of Dutch society piles in, in Mackay's telling,
- 15:51from nobles down to chimney-sweeps and old women selling clothes, selling their houses and land to buy bulbs.
- 15:59Then, in early 1637, the market breaks.
- 16:03Prices collapse. People who had agreed to buy bulbs at high prices refuse to pay.
- 16:09The courts decline to enforce the contracts, treating them as gambling debts.
- 16:14The matter goes up to the provincial council at The Hague, which cannot sort it out either.
- 16:20And the country, Mackay says, took years to recover.
- 16:25That is the story.
- 16:27Then the historian Anne Goldgar sets out to check it, by spending years in the Dutch archives,
- 16:34reading the actual contracts, notarial records and court papers of the 1630s.
- 16:40Her book came out in 2007.
- 16:43Here is what she found.
- 16:46She could identify thirty-seven people who spent more than three hundred guilders on bulbs, which was roughly a master craftsman's income for a year.
- 16:56Thirty-seven. Not a nation.
- 16:59They were mostly prosperous merchants and skilled artisans, and they were connected: by family, by religion, by living on the same streets.
- 17:09Most chains of trading involved five buyers or fewer.
- 17:14And she could not find the wreckage.
- 17:16No documented suicides caused by tulips.
- 17:19No bankruptcies caused by tulips.
- 17:21People who turn up in the bankruptcy records around that time were, if anything, buying assets from other bankrupts, which is not what ruined people do.
- 17:32No measurable damage to the Dutch economy at all.
- 17:36Peter Garber, an economist, had reached a similar view independently,
- 17:41arguing that most tulip prices can be explained by the ordinary economics of rare and fashionable goods.
- 17:48So where did the story come from?
- 17:51According to Goldgar, from songs.
- 17:54After the crash in 1637, satirical songs and pamphlets appeared mocking the people who had traded tulips.
- 18:02They were written to moralise, to make fun, to warn.
- 18:07Later writers picked them up and repeated them.
- 18:10They went into a German history of inventions, which was translated into English.
- 18:16And Mackay, two centuries after the event, took that material and treated it as a factual account.
- 18:23Which means the sailor with his herring is not a record of anything.
- 18:27It is a joke that was being told about some rich merchants in the 1630s, that got repeated until it became a fact.
- 18:38So here is the question I left hanging.
- 18:40If the tulip mania was small, and the South Sea Bubble was real, and in both cases people seem to have piled in together, what was going on inside those people?
- 18:52Were they mad?
- 18:53In 1992, two papers came out that answer this, and I think between them they are the missing half of Mackay's book.
- 19:02The first, by the economist Abhijit Banerjee, starts with an image I want you to keep.
- 19:09You are in an unfamiliar town, looking for dinner.
- 19:13There are two restaurants side by side.
- 19:15You have a faint hunch, from something you read once, that the one on the left is better.
- 19:24But when you get there, the one on the left is empty and the one on the right has a couple of people eating in it.
- 19:31So you go in to the one on the right.
- 19:34And that is a perfectly sensible thing to do.
- 19:36Those diners might know something.
- 19:38Your hunch was just a hunch.
- 19:41Now the next person arrives.
- 19:43They also have a faint private hunch.
- 19:45Maybe it favours the left.
- 19:47But they see three people in the right-hand restaurant and nobody in the left.
- 19:51So they go right.
- 19:53And the next.
- 19:55And the next.
- 19:56By the time the evening is over, the right-hand restaurant is full and the left is empty,
- 20:02and it is entirely possible that most of the people who ate there had a private hunch that the other one was better.
- 20:11Nobody ever found out, because nobody ever acted on it.
- 20:15Every hunch after the first couple got swallowed by what the room already looked like.
- 20:22The second paper, by Sushil Bikhchandani, David Hirshleifer and Ivo Welch, gave this a name and made it precise.
- 20:31They called it an information cascade.
- 20:34The set-up is very simple, and it is worth walking through, because the whole thing turns on one feature.
- 20:41People make a choice one after another.
- 20:44Each has a private piece of evidence, a signal, which is right more often than not.
- 20:50And each can see what everybody before them chose.
- 20:54What they cannot see is why.
- 20:56They see the choice.
- 20:57They do not see the signal behind it.
- 21:00Walk it forward.
- 21:02The first person has only their own signal, so they follow it, and their choice reveals it.
- 21:08The second person sees the first person's choice, which tells them the first person's signal, and adds their own.
- 21:15If the two agree, fine.
- 21:16If they disagree, it is a coin toss and they follow their own.
- 21:21Now the third person.
- 21:23Suppose the first two chose the same way.
- 21:25The third person's own signal is a single piece of evidence.
- 21:28What they can infer from the room is two pieces of evidence pointing the other way.
- 21:34Even if their own signal disagrees, it cannot outweigh two.
- 21:38So the rational thing to do is to ignore it and follow the crowd.
- 21:43And here is the trap.
- 21:44Because the third person followed the crowd regardless of their signal, their choice tells the fourth person nothing about what they privately knew.
- 21:53So the fourth person is in exactly the same position as the third.
- 21:57And so is the fifth.
- 21:59And so is everybody after.
- 22:02From that moment on, the crowd grows, and the amount of information in it stops growing.
- 22:07Every new person adds a body to the queue and nothing to the evidence.
- 22:14I want to know how much this costs, so I run it in a simulation.
- 22:19Take a hundred and one people, each of whom is right sixty per cent of the time on their own.
- 22:26Exactly the jury from the Condorcet episode.
- 22:29If they vote blind, each on their own signal, and you take the majority, they are right about ninety-eight per cent of the time.
- 22:38That was Condorcet's result.
- 22:40If instead they choose one after another, each able to see what the others chose, they are right about sixty-nine per cent of the time.
- 22:52Same people. Same private knowledge.
- 22:53Same question. The only difference is that they can see each other, and it costs them almost thirty points.
- 23:01And the reason is visible if you count how many people ever actually use their own evidence before the cascade locks in.
- 23:08The typical answer is two.
- 23:10Two people's private knowledge, and then ninety-nine people repeating it.
- 23:15A crowd of a hundred and one that knows what two people know.
- 23:21If you would rather watch this than take my word for it, I have built a version you can play with:
- 23:26a street with two restaurants, diners arriving one at a time, and sliders for how good everybody's private tip is.
- 23:36You can run a thousand evenings and see how often the street ends up in the worse restaurant,
- 23:42or take a turn yourself, which I recommend, because the wrong choice feels completely sensible from the inside.
- 23:51The link is in the notes.
- 23:53It has this model and a few more that you can play with.
- 23:58So here is the answer to the question about madness, and it is the one I think Mackay needed.
- 24:03In a cascade, nobody is mad.
- 24:06Every single person is doing the reasonable thing with what they can see.
- 24:11The absurdity lives in the arrangement, in the fact that choices travel between people and reasons do not.
- 24:19And there is a second prediction in the model that fits Mackay's stories uncomfortably well.
- 24:25A cascade rests on very little information, just those first couple of choices, however many people have piled on top.
- 24:34So it is fragile.
- 24:36A small piece of new public evidence can tip the whole thing over, and the entire crowd reverses at once.
- 24:43That is what a crash looks like from the inside.
- 24:46Nothing much changes in the world, and suddenly everybody is heading for the door.
- 24:53Now I want to take the same machine and show you it running on a crueller fuel.
- 24:59Mackay's chapter on the witch mania is the longest in the book apart from the Crusades and the alchemists, and it is hard reading.
- 25:08Across early modern Europe, historians now estimate that roughly a hundred thousand people were put on trial for witchcraft between about 1400 and 1750,
- 25:19and that around half of them were executed.
- 25:23The historian Brian Levack's range is forty to sixty thousand dead.
- 25:28What makes it the same machine is the evidence.
- 25:31In a bubble, the visible signal that everyone reads is the price.
- 25:35A rising price looks like proof that other people know something, so you buy, and your buying pushes the price up, and that looks like more proof.
- 25:46In a witch hunt, the visible signal is the confession.
- 25:50And confessions were very often extracted under torture, and under torture people were asked to name their accomplices.
- 25:58So one accusation produced a confession, and the confession produced names, and the names produced more accusations.
- 26:07Mackay records one man in France who named more than twelve hundred accomplices, many of whom were then arrested and executed.
- 26:15And the machinery that produced those confessions was built on purpose.
- 26:20Mackay traces the start of what he calls the witch mania proper to the 1480s,
- 26:25when Pope Innocent VIII issued a bull describing witches as a menace to Christendom and appointed inquisitors in every country with the power to convict and punish them.
- 26:38Almost immediately, Mackay says, a new profession appeared: men whose whole working life was finding and burning witches.
- 26:47They had a manual.
- 26:48The Malleus Maleficarum, the Hammer of Witches, which Mackay credits to an inquisitor called Sprenger,
- 26:56laid down a standard form of trial and a standard list of questions for the accused.
- 27:03Did you meet the devil at night?
- 27:05Did you attend the sabbath?
- 27:07And those questions, Mackay says, were put under torture.
- 27:11Now think about that with the mathematics of the last three episodes in your head.
- 27:15When confessions from villages hundreds of miles apart describe the same midnight meetings, the same pacts, the same sabbaths, it looks like overwhelming corroboration.
- 27:26Independent witnesses, all telling the same story.
- 27:29But they were not independent.
- 27:31They were answers to the same questions, from the same manual, extracted by the same method.
- 27:36The agreement between them was put there by the procedure.
- 27:40It is correlation, manufactured, and it is exactly the kind that the jury theorem says will drive a crowd, with total confidence, towards the wrong answer.
- 27:52Think about what that does to the evidence.
- 27:54Every new confession looks like independent confirmation that witchcraft is real and widespread.
- 28:00It was nothing of the kind.
- 28:01It was manufactured by the same process that it was then used to justify.
- 28:07And some of Mackay's own sources for what witches supposedly believed and did were demonologists who had sat on the trials themselves,
- 28:17writing up the confessions they had helped to extract.
- 28:22So the witch hunt is a cascade with one extra and terrible feature.
- 28:27In the restaurant, the diners at least chose freely.
- 28:29In the witch hunt, the institution produced the signals it then treated as proof.
- 28:36I should be careful here, because a cascade explains how a belief spreads, and not why it starts, or why it starts in one place rather than another.
- 28:45Other scholars have worked on that.
- 28:48The economist Emily Oster found that witch trials across eleven European regions rose in periods of cold weather and hardship,
- 28:56which suggests that people went looking for someone to blame when the harvest failed.
- 29:02Peter Leeson and Jacob Russ argued that trials were fiercest where Catholic and Protestant churches were competing for followers,
- 29:10as a way of demonstrating which one could protect you, but other researchers dispute that.
- 29:17These explanations are about where the spark lands.
- 29:20The cascade is about why the fire spreads.
- 29:25I owe you a caveat, and it is a big one.
- 29:29That number I gave you, sixty-nine per cent against ninety-eight, came out of a computer.
- 29:35I built a crowd of perfectly rational people who reason flawlessly about what they can see and never deviate from it.
- 29:43That is the theory's worst case.
- 29:46It tells you what cascades can do in principle.
- 29:48It does not tell you what real people do, and the whole point of this series is that you check.
- 29:54People have checked, and the results are more interesting than the theory.
- 29:59The classic test is from 1997, by the economists Lisa Anderson and Charles Holt.
- 30:05They put students in a room with two urns, each holding a different mix of coloured balls, and secretly chose one of the urns.
- 30:14Each student drew a single ball in private, which was their signal, and then announced out loud which urn they thought it was.
- 30:23One after another, each able to hear the guesses before them.
- 30:27That is the cascade, stripped to its bones.
- 30:30And cascades did form.
- 30:32In the situations where the theory said a cascade should start, one started in forty-one cases out of fifty-six.
- 30:40Roughly three times in four, people set aside their own ball and went with the room.
- 30:46So the effect is real.
- 30:47It is not something economists imagined.
- 30:51But then a group of researchers, Jacob Goeree, Thomas Palfrey, Brian Rogers and Richard McKelvey, asked a sharper question in a paper published in 2007.
- 31:03What happens if you let the sequence run for a long time?
- 31:09The theory says a cascade, once started, lasts forever, because nobody after the lock-in ever adds information.
- 31:16What they found was the opposite.
- 31:19The cascades in their lab were fleeting.
- 31:21They formed, and then they broke.
- 31:24And when a cascade reversed, it usually reversed towards the right answer.
- 31:29The reason is almost funny.
- 31:31Real people are not perfectly rational.
- 31:33Every so often someone ignores the crowd and goes with their own ball, even when the maths says they should not.
- 31:42And that stubborn person does something the perfect reasoners never do.
- 31:47They put fresh evidence back into the room.
- 31:50A few of those, and the crowd can see that the trail it was following was thinner than it looked.
- 31:57Notice who the culprits are.
- 31:59In a cascade, the flawless reasoners are the problem.
- 32:03They are precisely the people who stop contributing what they know.
- 32:07A little stubbornness, a little irrational faith in your own ball, turns out to be a public service.
- 32:14Then there is the question of who cascades.
- 32:16In a study published the same year, Jonathan Alevy, Michael Haigh and John List ran the urn experiment with professional traders from the Chicago Board of Trade,
- 32:29alongside students, across more than fifteen hundred decisions.
- 32:34The traders behaved differently.
- 32:36They paid more attention to how good the trail of earlier choices actually was, and they were more willing to throw it out when it was poor.
- 32:46Their expertise showed up less in knowing the answer than in knowing when not to follow.
- 32:54Those are lab studies, with urns.
- 32:56The obvious objection is that nobody lives in an urn experiment.
- 33:00So two further studies went looking for the same thing in the wild.
- 33:04In 2006 Matthew Salganik, Peter Dodds and Duncan Watts built an online music market.
- 33:11More than fourteen thousand people listened to songs by unknown bands and downloaded the ones they liked.
- 33:20Some participants could see how many times each song had already been downloaded.
- 33:25Others could not.
- 33:27The difference was stark.
- 33:29Where people could see each other's choices, success became far more unequal, a few songs ran away with everything, and it became far less predictable which ones.
- 33:42The best songs rarely flopped and the worst rarely won.
- 33:46But in between, the winner was close to a lottery, decided by who happened to download what early on.
- 33:54And in 2013 Lev Muchnik, Sinan Aral and Sean Taylor ran an experiment on a real news website, on more than a hundred thousand comments.
- 34:05On some comments, chosen at random, they planted a single fake up-vote as the first vote.
- 34:12On others, a single fake down-vote.
- 34:16The fake up-vote kept paying off: comments that got one ended up rated higher than they would have been otherwise.
- 34:24The fake down-vote mostly did not stick, because people pushed back and corrected it.
- 34:29The crowd piled onto praise and resisted criticism.
- 34:34So here is where I think the evidence leaves the theory.
- 34:37The theory is right that cascades form, right that they throw information away, and right that a tiny early signal can steer a very large crowd.
- 34:49It is wrong that they lock in forever.
- 34:52Real crowds leak.
- 34:54Stubborn individuals keep putting their own evidence back in, and over a long enough run that leak often brings the crowd back towards the truth.
- 35:04Which is, when you think about it, exactly Mackay's sentence.
- 35:08They go mad in herds, and recover one by one.
- 35:12And it gives me a way to say what made the witch hunts so much worse than the tulips, although I want to flag that this is my reading and not a finding.
- 35:22In a market, the stubborn person who refuses to follow the crowd loses some money, at most.
- 35:28In a witch hunt, the stubborn person who says the accused is innocent, or that the confessions are worthless, risks becoming the next accused.
- 35:38The machinery did not only manufacture agreeing evidence.
- 35:41It made disagreement dangerous.
- 35:44It plugged the leak.
- 35:46And a cascade that cannot leak does not self-correct.
- 35:49It runs for centuries.
- 35:52Now go back to the tulips, because I think the machine explains something about them that is even stranger than the bubble.
- 36:02Put yourself in the position of a writer in, say, the 1700s, who wants to say something about human folly.
- 36:09You have heard that there was a tulip madness in Holland.
- 36:12You find a pamphlet that describes it.
- 36:14It is vivid.
- 36:15It has good stories in it.
- 36:17And you have no easy way to check it, because the archives are in another country, in another language, and nobody thinks a flower market is worth the trip.
- 36:25So you repeat it.
- 36:27That is a perfectly reasonable thing to do.
- 36:29Somebody wrote it down, and it sounds right.
- 36:33The next writer finds your book, and the pamphlet behind it, and now there are two sources that agree.
- 36:39Two. The third writer sees two accounts pointing the same way and a vague feeling that it might have been exaggerated,
- 36:48and a vague feeling cannot outweigh two published authorities.
- 36:52So the third writer repeats it too.
- 36:56This is my interpretation, so hold it at that weight.
- 36:59But look at the shape of it.
- 37:01It is the restaurant.
- 37:04Each writer sees what earlier writers concluded.
- 37:07None of them sees the evidence the conclusion rests on, because there was almost no evidence, only satirical songs written to mock people.
- 37:16From very early on, every new retelling adds another voice and nothing to the evidence.
- 37:22And by the time the story reaches Mackay, it has been repeated so many times that its sheer repetition looks like proof.
- 37:30And notice how the story was finally checked.
- 37:34Not by a crowd.
- 37:35By one historian, sitting in an archive, reading notarial contracts one at a time, for years.
- 37:42Mackay's own sentence turns out to describe what happened to his book.
- 37:47The going wrong happened in a herd.
- 37:50The recovery happened slowly, and one by one.
- 37:53Mackay's phrase for his standard of evidence was scores of credible witnesses.
- 38:00That is exactly the wrong test for a cascade.
- 38:02A cascade is precisely the situation in which scores of credible witnesses all say the same thing because they are repeating each other.
- 38:10And the same thing happened to Newton's remark.
- 38:14Somebody reported a plain sentence, second-hand, a century after his death.
- 38:19Then it acquired the heavenly bodies.
- 38:21Then it acquired him saying it at a particular moment, to a particular person.
- 38:26Each retelling a little better than the last.
- 38:29Each one resting on the one before.
- 38:32The madness of crowds, it turns out, is also a story that crowds tell.
- 38:37And if you want to be depressed about it, notice that it is a story about how gullible other people are,
- 38:43which makes it exactly the kind of story everybody enjoys passing on.
- 38:49So what does any of this change, now?
- 38:53Here is the thing I keep coming back to.
- 38:56Every system you use that ranks things by popularity works like that urn experiment, run in public.
- 39:03The number of stars on a product.
- 39:05The number of people who have watched a video.
- 39:08The order of results on a page.
- 39:11The list of what is trending.
- 39:13In each case you see what other people chose, you do not see why, and you are invited to treat their choices as evidence.
- 39:22Sometimes that works.
- 39:25If the early choices were made independently, by people who actually knew something, then following them is sensible, and the popularity carries real information.
- 39:34But the cascade model tells you exactly when it stops working, and the answer is: very early.
- 39:41Once the first handful of choices point one way, nobody after them has any reason to add their own evidence.
- 39:48The count keeps going up.
- 39:50The information stops.
- 39:53So a thing with ten thousand ratings may know what the first dozen raters knew, and not a great deal more.
- 40:02And as the news-site experiment showed, a single planted early vote can move where it ends up.
- 40:09The same shape runs through public argument.
- 40:12Somebody accuses somebody else of something.
- 40:15A few people repeat it.
- 40:17Then a great many people repeat it, each of them seeing that many others already believe it, and none of them in a position to check.
- 40:24That is not madness either.
- 40:27Everybody is doing the sensible thing with what they can see.
- 40:31And the witch hunt is a warning about the case where the process itself starts generating the evidence it then relies on,
- 40:39where being accused is taken as a sign of guilt because so many other people have been accused.
- 40:48And markets, which is where Mackay started, run on exactly this.
- 40:52A price is a choice made visible.
- 40:54It tells you what people did and not why they did it.
- 40:58When prices rise, the rise itself looks like information,
- 41:02and people who buy because it is rising push it higher, and the thing can climb a long way resting on very little.
- 41:10And then a small piece of news tips it, and everybody leaves at once.
- 41:15So what do you do about it?
- 41:17The mathematics is actually quite clear on this, and it points back to the first two episodes of the series.
- 41:23What goes wrong in a cascade is the order of operations.
- 41:27Everybody reveals a choice before anybody's private evidence has been collected.
- 41:32The fix is to reverse that.
- 41:35Collect the private judgements first, independently, before anyone can see anyone else's.
- 41:41Then reveal them.
- 41:43We already know how to do this, and we do it in places where it matters.
- 41:47A secret ballot is exactly this, right?
- 41:49So is, for example, a sealed-bid auction, where everybody commits a price without seeing the others.
- 41:55So is asking the members of a panel to write down their view before anyone speaks, so that the most senior person in the room does not set the anchor for everyone else.
- 42:08The mammogram study from the Galton episode worked because the radiologists read the images separately,
- 42:14and only then they were able to talk and have their readings combined.
- 42:23In each case, you are keeping people in separate rooms for exactly as long as it takes to get their evidence out of them.
- 42:29And then you let them talk.
- 42:32So this is a rule about when conversation should happen, and nothing in it tells you to stop talking.
- 42:37Mackay did not have this.
- 42:41He had four centuries of disasters, a talent for telling them, and a belief that the cure for crowd folly was for each person to recover their senses alone.
- 42:52He was half right.
- 42:53The recovery does happen one by one.
- 42:55But the folly lived in the arrangement rather than in anyone's senses, and you cannot fix an arrangement by improving the people inside it.
- 43:05Which brings me to the man who took Mackay's museum and tried to turn it into a science.
- 43:10A French writer, at the end of the nineteenth century, who read all these stories and concluded something much darker than Mackay did.
- 43:18He did not think the individuals in a crowd were reasoning sensibly and being led astray by the arrangement.
- 43:24He thought that when people gather, the individuals disappear altogether, and something else takes their place.
- 43:31A single creature, with a mind of its own, and not a pleasant one.
- 43:36He is the reason we still talk about the psychology of crowds.
- 43:39And he chose, as one of his examples, the very assembly that hunted Condorcet.
- 43:46That is next time.
- 43:47Thanks for listening.
- 43:48Mackay's book is Extraordinary Popular Delusions and the Madness of Crowds, 1841, and it is out of copyright and easy to find.
- 43:55Anne Goldgar's Tulipmania is from 2007, and Peter Garber's Famous First Bubbles from 2000.
- 44:03Andrew Odlyzko's work on Newton is in Notes and Records of the Royal Society, 2019.
- 44:10The cascade papers are Banerjee, and Bikhchandani, Hirshleifer and Welch, both 1992.
- 44:15On the witch trials, Emily Oster in 2004, and Leeson and Russ in 2018.
- 44:20All in the notes.
- 44:22I'll see you next time.