The Future of Everything: The Science of Prediction

by David Orrell

  Despite the efficiency of the human immune system, it occasionally loses the battle for prediction. AIDS, caused by the HIV virus, has infected tens of millions worldwide, and over 25 million in Africa alone; tuberculosis (from a bacterium) and malaria (a microscopic parasite) kill millions each year; and new diseases are constantly emerging. Creutzfeldt-Jakob disease (CJD), the human analog to mad cow disease, is caused not by a microbe but by mis-folded proteins known as prions. In the 1990s, it was feared that CJD could kill millions of people, though more recent estimates are far lower.76 The 2003 SARS outbreak was caused by a novel virus that was much less transmissible than influenza, but it still managed to spread from rural China to five countries within a single day. It killed under a thousand, far fewer than a normal flu, but the resulting panic caused an economic crisis in much of Asia and in cities like Toronto, with a total price tag estimated at $30 billion. It showed that we are only a viral mutation or two away from a modern-day plague—but it also showed that the response can be more damaging than the problem itself.

  In 2005 and 2006, the world was working itself into a similar panic over a disease that is highly contagious, extremely lethal, and absolutely terrifying—at least for birds. Like the 1918 influenza, which took more lives than the First World War, the H5N1 strain of avian flu is made up mostly of genes from our feathered friends. But unlike the 1918 flu, it has yet to become easily transmissible between humans.77 Spread around the world by migratory birds such as ducks, which can host the virus without dying from it, avian flu has killed hundreds of millions of chickens and other birds. Of the humans who have contracted it, usually through direct contact with birds, it has killed about half. (This compares with a mortality rate of 2 to 3 percent for the 1918 pandemic.)78 If it mutates to a different, human-transmissible form, with even half the original virulence, then things would look extremely bad. But as Laurie Garrett, the author of The Coming Plague, pointed out in May 2005, “We have no idea what exact genetic changes this would require, how difficult it is for the virus to make those changes and whether or not the virus would significantly sacrifice its virulence level in the process.”79 There is no “normal” size for an outbreak. Avian flu may therefore turn into something far worse than the 1918 epidemic, or it may never happen.80

  Mathematicians and epidemiologists can model the spread of disease with computer simulations, which in their cruder form divide a population into three classes: those who have the disease, those who are susceptible, and those who are immune (possibly because they have died). The models assume that people encounter each other randomly, like molecules in a gas, and pass on the disease at a rate that depends on its transmissibility.81 Epidemics, if they become established, are then seen to follow a typical S-shaped curve, rather like that in figure 7.1 (see page 282).

  More sophisticated models simulate a detailed population that statistically matches the properties of a given city. Each “individual” in the model will interact with a certain number of people each day. The number of interactions varies just as it does in a real town, so that students attending class come into contact with many people, while those who work at home do not. These simulations have shown that the speed with which health officials act—by isolating patients and telling people to stay at home—is critical.82 Indeed, the impact of SARS was much lower in Vancouver than in Toronto because healthcare workers there immediately managed to contain the disease.

  However, the most important factor is the nature of the disease itself, and we cannot predict what exactly will emerge from the global ecosystem.83 The process by which viruses incorporate new genes is inherently random. The evolution of a microbe as it adapts to a new environment is also highly variable and unpredictable. In one experiment, two samples of a virus that usually infects the bacterium E. coli were introduced to Salmonella instead. The two virus samples adapted in completely different ways, and after just ten days, they were genetically distinct.84

  Nor can we predict the lethality or transmissibility of a disease by sequencing a microbe’s DNA, since symptoms (like traits) are an emergent feature of the complex interaction between the microbe and its host. This was scarily demonstrated in 2001 by Australian researchers who were trying to develop a mouse contraceptive as a population-control device. Their approach involved the insertion of the gene interleukin-4 into the genome of mousepox (a version of smallpox that afflicts mice but not humans). The gene was expected to boost antibody production, but instead it transformed the virus into a raging killer that took out all the laboratory mice.85 A major concern is that biotechnologists will accidentally create novel diseases that control the populations of both mice and men.

  According to legend, Apollo’s arrow enabled Pythagoras to cure plagues, but there seems little that mathematical models can do to offer similar protection. Building a model of an epidemic in progress is rather like working out how to double the cube at the altar of Apollo during the Athenian plague. So how can we prepare ourselves for the emergence of new diseases?

  The best option, it seems, is to take a cue from our own immune systems, both innate and acquired. The innate response is to couple systems such as the Global Public Health Intelligence Network,86 which automatically searches news reports and websites worldwide to pick up signs of an outbreak, with available technologies such as antiviral drugs87 and antibiotics. The acquired response is vaccination tailored to the particular disease. The World Health Organization is attempting to improve systems for the design, production, and distribution of vaccines.88 New biotechnologies can play an important role by speeding the process, and high-throughput techniques can monitor the development of dangerous pathogens. Equally important will be low-tech plans to maintain basic health care and food supplies during an outbreak.

  Of course, as Timothy Geithner, president of the Federal Reserve Bank of New York, remarked on the slightly different subject of hedge funds, “It is hard to motivate people to buy more insurance against adverse outcomes when the risks seem remote and hard to measure and when present conditions seem favourable.” 89 The price tag of the next epidemic is uncertain, but a protective network can be accurately costed.90 It is natural for those footing the bill to demand proof that the investment is worthwhile. But how can we prove whether avian flu is the revenge of chickens on the human race? As Garrett says, “The bottom line for policymakers: Science does not know the answer.”91 It is hard to get the balance right, and even our own immune system overreacts sometimes. All we can do is watch for coming storms.

  WE DON’T KNOW

  It might seem in this chapter that we have fallen into the trap of assuming that the future will resemble the past: just because we cannot predict atmospheric, biological, or economic systems now does not mean that we will not be able to do so one day, once we have better computers, observation systems, and models. Statements about the limitations of human ingenuity have a way of being proven wrong. However, the constraints here are the result of the nature of the systems themselves. In cellular automata, computational irreducibility does not (by definition) go away with a better computer, and emergent properties cannot (by definition) be expressed in terms of simple physical laws. Similarly, the dual nature of complex real-world systems means that they can be based on simple, local rules—avian flu is just a few bird genes—but at the same time be uncomputable. Errors in model parameteriza-tions are magnified by the complex feedback loops that characterize such systems. In a hundred years, we won’t have an equation for the Game of Life, and (to venture a prediction) we won’t have an equation for life.

  Even if models do not have predictive accuracy, they are still useful tools for understanding the present, envisaging future scenarios and educating policy makers and the public. Scientific research into global warming—along with pictures of calving icebergs—has highlighted the changes that are currently happening in the climate system, and provoked debate about the possible consequences. Similarly, models of the spread of disease have helped focus people’s minds o
n the possible consequences of a pandemic, and economic models have shown how dependent we are on our relationship with the rest of the biosphere. For, in the long run, what unites our future weather, health and wealth is that they all rely on the state of the planet.

  To summarize:

  Prediction is a holistic business. Our future weather, health,and wealth depend on interrelated effects and must be treated in an integrated fashion.

  Long-term prediction is no easier than short-term prediction. The comparison with reality is just farther away.

  We cannot accurately predict systems such as the climate for two reasons: (1) We don’t have the equations. In an uncom-putable system, they don’t exist; and (2) The ones we have are sensitive to errors in parameterization. Small changes to existing models often result in a wide spread of different predictions.

  We cannot accurately state the uncertainty in predictions. For the same two reasons.

  The effects of climate change on health and the economy (and their effects on the climate) are even harder to forecast. When different models are combined, the uncertainties multiply.

  The emergence of new diseases is inherently random and unpredictable. Avian flu may be the next big killer—but a bigger worry is the one that no one has heard about yet.

  Simple predictions are still possible. These usually take the form of general warnings rather than precise statements.

  Models can help us understand system fragilities. A warmer climate may cause tundra to melt and rainforests to burn, thus releasing their massive stores of carbon. However, the models cannot predict the exact probability of such events, or their exact consequences.

  Uncertainty means that discussions become polarized between opposing camps of optimists and pessimists. This, combined with the difficulty in costing unknown future perils, creates a bias towards inaction.

  Given that optimists and pessimists are unlikely to convince one another on purely theoretical grounds of our future impact on the planet, and its impact on us, it is hard to imagine how the debate will be resolved until events unfold. To better prepare for an uncertain future, perhaps we need to discontinue the search for more accurate numerical predictions and do the opposite instead. We take up this theme in the next chapter.

  8 BACK TO THE DRAWING BOARD

  FIGURING OUT WHERE WE WENT WRONG

  Hereafter, when they come to model Heaven

  And calculate the stars, how they will wield

  The mighty frame, how build, unbuild, contrive

  To save appearances, how gird the sphere

  With centric and eccentric scribbled o’er,

  Cycle and epicycle, orb in orb.

  —John Milton, Paradise Lost

  You don’t need a weather model to know which way the wind blows.

  —After Bob Dylan

  UNDER THE VOLCANO

  In Autumn 2004, all eyes, along with a large number of highly sensitive scientific instruments, were focused on Mount St. Helens, in the Cascade Mountains. Back in May 1980, Mount St. Helens had been hit by a magnitude 5.1 earthquake. The earthquake had set off a massive volcanic eruption, along with the largest landslide in the earth’s recorded history. A plume of gas and ash reached twenty-five kilometres high in less than fifteen minutes. Some 500 million tons of ash were sent into the atmosphere, blotting out the sun for hundreds of miles around; the ash had spread around the world in two weeks. Fifty-seven people died in the blast; some managed to outrace it by driving at speeds of up to 150 kilometres an hour.1 Now it looked like the mountain might do it again. Trembling to a series of smaller earthquakes, burping the occasional stream of hot ash—was this the prelude to another major eruption or just a passing phase?

  Mount St. Helens is part of the Ring of Fire, a network of volcanoes and faults that forms a circle around the Pacific Ocean and marks the boundaries of geological plates whose pressure and shear against each other is balanced by the forces of friction. Any small slip or shudder alters this balance and sends molten rock to the surface. The lava in Mount St. Helens is squeezed up into a six-kilometre-wide chamber. Heat and pressure fracture the rock, turn groundwater and glacial ice in the crater to steam, and push up against the lava dome, a cap of cooled magma that acts like a cork in a bottle of champagne.

  Scientists from around the world camped out near the mountain to monitor its movements. Small aircraft circled like flies, with onboard devices sniffing for signs of gases that might indicate the magma was rising. Global-positioning instruments and seismometers picked up minute movements and tremors, and microphones near the crater listened for any sign of activity.

  If anything was being more closely monitored and recorded than the mountain, it was the musings of the scientists on the evening news. The question every reporter from around the world wanted answered was, Will the mountain blow? The answers were always guarded and usually uninformative. There could be a major eruption, but it’s unlikely. The most probable outcome is a minor eruption. Or it could stop now and lapse back into a coma, not saying a word for a hundred years.

  In the end, nothing happened. The mountain calmed down. The foreign reporters returned to Japan or France or wherever-they were from. Washington State moved its attention back to the upcoming presidential election, and the Ring of Fire seemed snuffed out—until December 26, 2004, when it roared back to life not as a volcanic eruption in Washington, but as a massive earthquake on the other side of the world.

  The magnitude 9.0 underwater quake off the west coast of Sumatra was caused when the floor of the Indian Ocean slipped under the Burma plate, creating a seabed cliff over 10 metres high and 1,000 kilometres long. The sudden movement initiated a tsunami that raced across the ocean at 900 kilometres per hour towards the beaches of Indonesia, Thailand, India, and Sri Lanka. As the waves neared the coast, they slowed and grew to a height of several metres before slamming into the land. Their destructive force swept away people on the beach, and in villages and cities. The death toll of nearly a quarter million made it one of the biggest natural disasters in history.

  No reporters anticipated this event. Nor were there tidal gauges or buoys to give an early warning. Yet the earthquake had been felt around the world, and within fifteen minutes of its occurence, scientists had informed several countries (including Thailand and Indonesia) that dangerous waves might be generated; however, because such warnings usually turn out to be false alarms, the information was not acted on. As one professor stated, “We have believed as a community that the Indian Ocean is fairly immune to tsunamis of the kind that took place.”2

  PREDICTING EARTHQUAKES

  Earthquakes are caused by the slipping and sliding of huge plates of rock that float on the earth’s molten core. When the mechanics of plate tectonics were first understood in the 1960s, it heralded an era of optimism about earthquake prediction. Billions of dollars were funnelled into research programs. Japan, which found itself on the intersection of three such plates, spent ¥160 billion over thirty years. Theories abounded that earthquakes followed some predictable pattern or had reliable precursors.

  Unfortunately, the research has shown little success. Earthquakes, like the one that devastated the Kashmir area of India and Pakistan in October 2005, are as unpredictable today as they were thirty years ago. No precursors have been identified, and most of the research programs have been wound up.

  The problem is that earthquakes represent a shift in balance between two opposing forces: the dynamical force that grinds one plate against another, and the frictional force that resists movement. Tension is released not in a smooth, continuous manner, but in sudden fits and starts, felt as tremors or quakes. The situation is in some respects similar to collapses in financial markets, which are the result of a sudden shift in balance between buyers and sellers. Like financial crashes, earthquakes occur at unpredictable intervals, and their magnitude tends to follow a power-law distribution: there are many smaller ones, and fewer large ones.

  Even if the timing of earthquak
es eludes prediction, it is possible to forecast where they are most likely to occur from the location of the tectonic plates. Engineers can also calculate how manmade structures will react and design building codes for areas at risk. Whether their advice is heeded is another question—it hasn’t been in much of northern California, which rides along the San Andreas fault, part of the Ring of Fire.

  We would all like to know what is on the horizon for the earth and sky over the next one hundred years. Are we living under a volcano? Will global warming or other human causes trigger a major disaster, or will the threat fizzle out like Mount St. Helens? Yet our best models of weather, health, and wealth regularly fail to predict even short-term phenomena. So perhaps an equally important question is, What are we doing wrong? Is there something fundamentally askew in our mental approach, rather than just the technical execution? Are we looking for answers in the wrong place? In this chapter, we put our three sibling oracles on the analyst’s couch to find out why they keep going astray—and whether their behaviour can be blamed on the parents.

  CAUSE AND EFFECT

  In the nineteenth century, the psychologist Franz Brentano observed that people divide the world into two categories and handle each in different ways.3 The first category consists of things that act spontaneously and with intentionality—in short, are alive. The second includes things that obey physical laws. The distinction between the two classes is not necessarily for complicated philosophical or religious reasons, or because we assign a mysterious “vital force” to one and not the other, but is due to the simple empirical fact that they behave differently. Living beings have evolved in a way that gives them special properties. Kick a stone, and a sense of physics will explain what happens; kiss a person, and it’s more complicated.