by David Orrell
DON’T BLAME THE BUTTERFLY
My own introduction to predictability occurred when I returned to university, after several years as a jobbing mathematician, to do a Ph.D. on model error in weather forecasting. Model error represents the difference between the model—typically a set of mathematical equations based on physical “laws”—and the actual system it is supposed to emulate. For example, the trajectory of an arrow is something that can be determined reasonably accurately from the arrow’s starting position and its velocity—the initial conditions—using the laws of physics. But if there’s a gust of wind that is not included in the model, then the arrow will depart slightly from its predicted path. That’s model error. It might not be important, unless you happen to be the person waiting at the other end with the apple on your head.
In weather forecasting, there had been little investigation of model error, despite the fact that even forecasters agreed that predictions usually missed their target after about two or three days. From my work experience (by which I mean glaring discrepancies between my calculations and reality), I knew that even engineered systems, where all the forces and material properties are exactly known, could still be hard to model accurately. The atmosphere was a horrendously complex system in comparison, so model error should have been huge. However, the dominant theory was that forecasts went wrong not because of any deficiency in the model, but because small errors in the initial condition were magnified by chaos—the so-called butterfly effect. Storms like the one that hit the north coast of Venezuela could, in principle, be caused by an insect flapping its wings somewhere on the other side of the world. My work over the next couple of years was aimed at developing a technique for measuring model error that filtered out the effects of chaos. When our group’s results (which showed that most forecast error was a result of the model, with chaos a relatively minor effect) were accepted for publication and presented at conferences, there was initially no reaction from the meteorological community. But then the story was picked up by the media. Soon there were reports in newspapers and magazines and on radio shows in Europe, North America, and elsewhere. The reaction seemed out of proportion to the actual scientific interest; but everyone is interested in the weather, and everyone knows that the forecasts are wrong. The idea that the cause could be the models was apparently big news. Perhaps storms could be better predicted.
While this seemed a positive development, it didn’t go down well. Criticizing models was apparently a good way to annoy the weather gods, or at least the weathermen.12 Eventually, over a few years, the storm clouds dissipated. The work was published, and I moved on to different things. However, I remained struck by the deep blanket of denial that settled over those in the meteorological community, and by their emotional reaction to criticism, where any questioning of the model was interpreted as a personal attack. No matter the evidence to the contrary, they always believed that their models were right. It raised questions in my mind about the science and sociology of forecasting, questions this book attempts to answer.
I learned later that this lack of zeal in investigating model error was not unique. The work of Will Keepin and Brian Wynne at the International Institute for Applied Systems Analysis (IIASA) in Austria was proof of that. In 1981, the institute had just finished a multi-million-dollar project, involving over a hundred scientists, that was supposed to forecast the world’s future energy consumption. After accounting for factors such as demographics and projected oil reserves, the computer model predicted that demand for energy would rise enormously over the next fifty years, and could be met only by building over a hundred nuclear power plants a year. Needless to say, this would have required a huge expansion in the nuclear industry, creating lucrative jobs for anyone with expertise in the energy field—such as, for example, the scientists who had built the model.
To Keepin, however, the model was so flexible, its connection to reality so tenuous, that “it was a bit like the Wizard of Oz. . . . Some guy was pulling on levers and making a big show, but it was a show determined by the little guy behind the curtain.”13 Almost alone in his criticism of the model, he decided to resign from his job. Just then, the British scientist Brian Wynne came to IIASA on a twoyear contract to study the politics of science. He was looking for an insider to tell all about the energy model. After hearing Keepin’s story, Wynne decided, to the institute’s horror, to devote his two years to studying the other scientists’ reaction to Keepin’s critique. In 1984, the results were published in the top journal Nature.14 The paper showed that the model had been biased in favour of nuclear and fossil-fuel energy producers, and that the model developers had tried to conceal its shortcomings.
Similarly, when the mathematician Benoit Mandelbrot questioned the assumptions behind modern finance, he found himself “about as welcome in the established church of economics as a heretical Arian at the Council of Nicene.”15 As the philosopher Thomas Kuhn pointed out, in science it is common, and even healthy, for new ideas to be met with skepticism by the establishment.16 But if scientific models are used to set policy, and to make important public and private decisions, then we need to know how accurate they are. This is made difficult by the nature of the models, which are written in a highly specialized language that can be understood only by other scientists with experience in the field. Since such people often share the biases of the model makers, glaring problems go undetected, or unremarked.17 Indeed, scientific institutions have become expert at deflecting serious public scrutiny of their work.18 So to what extent can we trust their predictions for the future?
WHERE IS THIS GOING?
Forecasting has always attracted fraudsters and con men. When Kepler was trying to promote his predictive model of the sun, moon, and planets to Emperor Rudolph II, his competition was not so much other astronomers, but savants like the Englishman Edward Kelley, who preferred a talking mirror to calculations and was eventually jailed by the emperor for his poor performance. More recently, studies have shown that social forecasting, scientific and otherwise, is about as accurate as random guessing, despite the vast numbers of highly paid experts employed to do it.19 If the futurologists of the 1960s had been right, for example, I would probably be writing this in an orbital space station as my personal robot tended to my toenails.
The accuracy of forecasts would not be so important if all that were at stake was the weekend weather or the likely return from government bonds in the next quarter. But like the residents of Caracas, we are becoming aware that the future need not resemble the recent past; the coming storms in weather, health, and wealth may be more intense than the kind we have grown used to.
It is only in the past few decades that human activities, like our use of large-scale industry and the automobile, have become comparable in scale to the workings of the planet itself. And it is even more recently that we have started to learn of holes in the ozone layer, the spread of chemical pollutants through the food chain, and the collapse of ocean fisheries because of overfishing. We have passed a kind of tipping point in our relationship with the world; our actions now influence its workings at every level. It used to be that the world happened to us; now we happen to the world as well. One day, as our children survey the damaged planet they have inherited, we may hear the question (asked less fairly of the Venezuelan president), Are you responsible?
The next fifty or one hundred years are going to be crucial, and we need to have a guide. In many ways, science got us into this fix, but will it help us to get out? And even if it can’t tell us exactly where the world is headed, can it help us predict our own future health or play the stock market? To answer these questions, it is necessary to understand how scientists go about making forecasts. Equally important, though, is the history and sociology of science. Like any complex process, prediction is path-dependent: it matters how we got here.
My own short experience with the weathermen was, in the scale of things, a minor affair, a tempest in a teacup. I wasn’t burned at the stake like Giord
ano Bruno, who had tried to convince the Inquisition that space was infinite, or threatened with torture and imprisonment for mocking the pope and arguing that the earth went around the sun, like Galileo Galilei. However, it did make me realize that in many ways, science has become rather like the Catholic Church of Galileo’s time, and about as receptive to criticism. And just as we once looked to the Church to predict the future—just keep your head down until the Second Coming—we now look to the scientists for guidance.
We are all predictors, living by our forecasts. The most primitive bacteria have the ability to sense the presence of food and move towards it. Living beings are constantly interacting with their environment, reading and displaying information. Successful strategies—knowing when to hunt, when to run, when to sit it out—are coded in the genes. The practice of speed dating is based on the idea that first impressions count: within a few minutes of meeting a prospective mate, we somehow fast-forward through the whole relationship and predict whether it will work (and because this affects how much effort we put into the relationship, it can become a self-fulfilling prophecy).20 We may not judge a book by its cover, but it certainly helps to scan the first few pages for a summary.
In science, though, forecasting plays a special role. Predicting the future is not a side activity of science, but has come to be seen as its primary pursuit.21 A scientific theory is generally considered valid only if it can be used to predict the behaviour of a system. A theory that doesn’t predict may be a beautiful or elegant idea, but it’s no more functional, in the view of many scientists, than a piece of modern art. We may all be predictors, but for scientists, it’s their profession. In this book, however, I will argue the following:
Mathematical models interpret the world in simple mechanical terms. Scientific prediction, from ancient astronomy up to and including chaos theory, has been based on a highly abstracted, mechanistic view of the world, which is of limited applicability in the context of complex systems.
Living things have properties that elude prediction. Systems where predictions are of interest—in biology, economics, or climate change—are either alive, influenced by life, or have a similar level of complexity to living beings. They are difficult to predict not because of simple technical reasons, which can be overcome with faster computers or better data, but because they have evolved to be that way. We pinpoint the causes of prediction error.
Forecasting has a large psychological component. The desire to explain the world in terms of simple cause-and-effect relationships is a fundamental characteristic of human beings.Predictions often tell us more about group psychology than they do about reality. Many prognosticators anticipated chaos in the financial markets in the new millennium, but the cause was supposed to be the Millennium Bug, not the collapse of Internet stocks. And accurate predictions, such as those that pointed out the vulnerability of New Orleans to a hurricane, are often ignored.22
Some predictions are still possible. One type of prediction relates to overall function and can be used to make general warnings. The other type involves specific forecasts about the future. Mathematical models are better at the first than they are the second (Niels Bohr was right: predicting the future is hard).
We need to change our approach to prediction. The current debate between climate modellers who argue that global warming is an imminent threat and skeptics who demand further proof can be resolved only with a fundamental shift in the kinds of predictions we make.
This book is divided into three main parts. The first is a brief history of the science of prediction. It will argue that modern forecasters are drawing on a long tradition of modelling the physical universe that stretches back to the ancient Greeks; and that throughout history forecasters have not just peered into the future but have helped shape the world we live in. Everything from our economic system to our relationship with nature and our own bodies has been profoundly affected by the early predictors, the model makers, the champions of cause and effect.
Of course, reading any such abbreviated history is a little like listening to a classic rock station on the radio. Just as each band is allowed to have only a handful of representative songs, so the great scientists have their life’s work boiled down to a couple of greatest hits: Pythagoras and the Music of the Spheres, Kepler and his Harmony of the World, Galileo and his Stones. Minor scientists—the supporting acts—seem never to have existed, except as part of the occasional quirky sideshow. Unlike most classic rockers, though, the great scientists considered here are all European males. This is not because prediction is not practised by other races, or indeed by females, but because culture has played a role in the development of prediction as it is currently practised by scientists. And as the science historian Evelyn Fox Keller has pointed out, science has not been a gender-neutral pursuit.23 These are subjects we will return to.
The second part of the book examines forecasting practice in the specific areas of weather, health, and wealth, and describes in detail the techniques currently employed by the scientists who make prediction their living. Like siblings, these three main areas of scientific prediction grew up together, share DNA, and show similar traits. To understand one, it helps to know the others. Finally, in the third part of the book, we see how these separate strands come together in a long-term forecast for the planet—culminating in a look at predictions for the year 2100.
The ultimate aim of the book is to make a forecast about forecasting, and to try to answer the question, Can scientists really look into the future? To find the answer, we must begin with the spiritual and intellectual forebears of modern numerical prediction—a secretive cult in ancient Greece led by a man they claimed was the son of Apollo.
PAST
1 SLINGS AND ARROWS
THE BEGINNINGS OF PREDICTION
All things are full of gods.
—Thales, Greek philosopher and mathematician
The truth of the model is not the truth of the phenomenon. It is a common confusion between these two kinds of truth—the norm in magic—that sometimes sanctifies the model (which is regarded as part of the real world) and gives the scientist the role of priest.
—Antoine Danchin, Pasteur Institute biologist
GAIA
According to Greek mythology, the first oracle, the maker of forecasts, was the earth goddess Gaia. She held forth at Delphi, which was named after the Greek word delphus, for “womb,” and was literally the womb of the earth. Geographically, Delphi is located on a gentle slope on Mount Parnassus, about 150 kilometres northwest of Athens. On one side, the area is towered over by 300-metre cliffs that are known as the Phaedriades, or Shining Ones, because of the almost metallic way they catch the morning and evening light. The ground is nourished by the Castalian spring, which flows through a cleft in the cliffs. Below, a gorge filled with olive trees leads down to the Gulf of Corinth. The whole area is prone to storms, landslides, and other outbursts of the gods. It is watched over by birds of prey who ride the thermals of the cliffs.
The ancient Greeks believed this beautiful and dramatic place to be the centre of the universe. A legend states that the god Zeus released two eagles, one from the east, one from the west. When they met at Delphi, Zeus placed a stone, the omphalos, to mark the spot. Gaia’s prophecies were sung out by a mythical figure referred to as Sybil, who inhaled trance-inducing vapours from a fissure in the mountain. The site was guarded by Gaia’s daughter, the fearsome serpent Python, who lurked in the nearby Castalian spring.
Like his father Zeus, the god Apollo had an interesting and complicated life. He was god, among other things, of reason, music, plague, and archery. He had many love affairs, with both goddesses and mortal humans. But the young, inexperienced god’s first big achievement—the one that put him on the map—was to slay the giant serpent Python:
E’re now the God his arrows had not try’d
But on the trembling deer, or mountain goat;
At this new quarry he prepares to shoot.
Tho
ugh ev’ry shaft took place, he spent the store
Of his full quiver, and ’twas long before
Th’ expiring serpent wallow’d in his gore.1
Since Python was Gaia’s daughter, amends had to be made for this violent deed. Apollo worked for eight years as a cowherd to purify himself. But once that was done, he returned to Delphi and, in a hostile takeover, claimed the oracle from Gaia. From that moment on, he was known as Pythian Apollo, the god of prophecy, and Delphi was his main shrine.2
That’s the mythology. Archaeological excavations have shown that from 1500 to 1100 B.C., the site was occupied by small Bronze Age Mycenean settlements that were dedicated to the Mother Earth deity. The new god Apollo arrived, perhaps via invading Dorians, and began to dominate. So in both versions, a power shift takes place between Gaia and Apollo. The chaos theorist Ralph Abraham refers to this time in human history as a major bifurcation point, where “the goddess submerged into the collective unconscious, while her statues underwent gender-change operations.”3 The result was the most successful prediction business in history. For almost a thousand years, the Delphic Oracle called the shots in business, politics, religion, and war.
The biographer Plutarch, best known for his lives of famous Greeks and Romans, served as a priest at Delphi, and from his histories we have some knowledge of the inner workings of the Delphic sanctuary.4 The oracle, known as the Pythia, was always a woman, since women were thought to be more receptive to Apollo’s oracular powers. Like a telegenic TV presenter, the Pythia didn’t make the forecasts herself, but only channelled the predictive power of Apollo. The main job requirements were enthousiasmos (which in its original sense meant not enthusiasm but “possessed by a god”) and faithfulness to Apollo. She was not allowed to have intimate relations with anyone, even a husband, for Apollo was a jealous god. A case in point was Cassandra. Apollo attempted to seduce her by granting her prophetic powers, but she refused him. In revenge, he cursed her so that no one would pay attention to her predictions.
