The Black Swan

by Nassim Nicholas Taleb

  The same idea applies to debt—it makes you fragile, very fragile under perturbations, particularly when we switch from the assumption of Mediocristan to that of Extremistan. We currently learn in business schools to engage in borrowing (by the same professors who teach the Gaussian bell curve, that Great Intellectual Fraud, among other pseudosciences), against all historical traditions, when all Mediterranean cultures developed through time a dogma against debt. Felix qui nihil debet goes the Roman proverb: “Happy is he who owes nothing.” Grandmothers who survived the Great Depression would have advised the exact opposite of debt: redundancy; they would urge us to have several years of income in cash before taking any personal risk—exactly my barbell idea of Chapter 11, in which one keeps high cash reserves while taking more aggressive risks but with a small portion of the portfolio. Had banks done that, there would have been no bank crises in history.

  We have documents since the Babylonians showing the ills of debt; Near Eastern religions banned debt. This tells me that one of the purposes of religion and tradition has been to enforce interdicts—simply to protect people against their own epistemic arrogance. Why? Debt implies a strong statement about the future, and a high degree of reliance on forecasts. If you borrow a hundred dollars and invest in a project, you still owe a hundred dollars even if you fail in the project (but you do a lot better in the event you succeed). So debt is dangerous if you have some overconfidence about the future and are Black Swan blind, which we all tend to be. And forecasting is harmful since people (especially governments) borrow in response to a forecast (or use the forecast as a cognitive excuse to borrow). My Scandal of Prediction (i.e., bogus predictions that seem to be there to satisfy psychological needs) is compounded by the Scandal of Debt: borrowing makes you more vulnerable to forecast errors.

  Big is Ugly—and Fragile

  Second, Mother Nature does not like anything too big. The largest land animal is the elephant, and there is a reason for that. If I went on a rampage and shot an elephant, I might be put in jail, and get yelled at by my mother, but I would hardly disturb the ecology of Mother Nature. On the other hand, my point about banks in Chapter 14—that if you shot a large bank, I would “shiver at the consequences” and that “if one falls, they all fall”—was subsequently illustrated by events: one bank failure, that of Lehman Brothers, in September 2008, brought down the entire edifice. Mother Nature does not limit the interactions between entities; it just limits the size of its units. (Hence my idea is not to stop globalization and ban the Internet; as we will see, much more stability would be achieved by stopping governments from helping companies when they become large and by giving back advantages to the small guy.)

  But there is another reason for man-made structures not to get too large. The notion of “economies of scale”—that companies save money when they become large, hence more efficient—is often, apparently behind company expansions and mergers. It is prevalent in the collective consciousness without evidence for it; in fact, the evidence would suggest the opposite. Yet, for obvious reasons, people keep doing these mergers—they are not good for companies, they are good for Wall Street bonuses; a company getting larger is good for the CEO. Well, I realized that as they become larger, companies appear to be more “efficient,” but they are also much more vulnerable to outside contingencies, those contingencies commonly known as “Black Swans” after a book of that name. All that under the illusion of more stability. Add the fact that when companies are large, they need to optimize so as to satisfy Wall Street analysts. Wall Street analysts (MBA types) will pressure companies to sell the extra kidney and ditch insurance to raise their “earnings per share” and “improve their bottom line”—hence eventually contributing to their bankruptcy.

  Charles Tapiero and I have shown mathematically that a certain class of unforeseen errors and random shocks hurts large organisms vastly more than smaller ones. In another paper, we computed the costs to society of such size; don’t forget that companies, when they fall, cost us.

  The problem with governments is that they will tend to support these fragile organisms “because they are large employers” and because they have lobbyists, the kind of phony but visible advertised contributions so decried by Bastiat. Large companies get government support and become progressively larger and more fragile, and, in a way, run government, another prophetic view of Karl Marx and Friedrich Engels. Hairdressers and small businesses on the other hand, fail without anyone caring about them; they need to be efficient and to obey the laws of nature.

  Climate Change and “Too Big” Polluters

  I have been asked frequently on how to deal with climate change in connection with the Black Swan idea and my work on decision making under opacity. The position I suggest should be based both on ignorance and on deference to the wisdom of Mother Nature, since it is older than us, hence wiser than us, and has been proven much smarter than scientists. We do not understand enough about Mother Nature to mess with her—and I do not trust the models used to forecast climate change. Simply, we are facing nonlinearities and magnifications of errors coming from the so-called butterfly effects we saw in Chapter 11, actually discovered by Lorenz using weather-forecasting models. Small changes in input, coming from measurement error, can lead to massively divergent projections—and that generously assumes that we have the right equations.

  We have polluted for years, causing much damage to the environment, while the scientists currently making these complicated forecasting models were not sticking their necks out and trying to stop us from building these risks (they resemble those “risk experts” in the economic domain who fight the previous war)—these are the scientists now trying to impose the solutions on us. But the skepticism about models that I propose does not lead to the conclusions endorsed by anti-environmentalists and pro-market fundamentalists. Quite the contrary: we need to be hyper-conservationists ecologically, since we do not know what we are harming with now. That’s the sound policy under conditions of ignorance and epistemic opacity. To those who say “We have no proof that we are harming nature,” a sound response is “We have no proof that we are not harming nature, either;” the burden of the proof is not on the ecological conservationist, but on someone disrupting an old system. Furthermore we should not “try to correct” the harm done, as we may be creating another problem we do not know much about currently.

  One practical solution I have come up with, based on the nonlinearities in the damage (under the assumption that harm increases disproportionately with the quantities released), and using the same mathematical reasoning that led to my opposing the “too big” concept, is to spread the damage across pollutants—should we need to pollute, of course. Let us carry on a thought experiment.

  Case 1: You give the patient a dose of cyanide, hemlock, or some poisonous substance, assuming they are equally harmful—and assuming, for the case of this experiment, the absence of super-additivity (that is, no synergetic effects).

  Case 2: You give the patient a tenth of a dose of each of ten such substances, for the same total amount of poison.

  Clearly we can see that Case 2, by spreading the poison ingested across substances, is at the worst equally harmful (if all the poisonous substances act in the same way), and at the best close to harmless to the patient.

  Species Density

  Mother Nature does not like too much connectivity and globalization—(biological, cultural, or economic). One of the privileges I got as a result of the book was meeting Nathan Myhrvold, the type of person I wish were cloned so I could have one copy here in New York, one in Europe, and one in Lebanon. I started meeting with him regularly; every single meeting has led to a big idea, or the rediscovery of my own ideas through the brain of a more intelligent person—he could easily claim co-authorship of my next book. The problem is that, unlike Spyros and those very few others, he does not have his conversations while walking (though I met him in excellent restaurants).

  Myhrvold enlightened me about an additional w
ay to interpret and prove how globalization takes us into Extremistan: the notion of species density. Simply, larger environments are more scalable than smaller ones—allowing the biggest to get even bigger, at the expense of the smallest, through the mechanism of preferential attachment we saw in Chapter 14. We have evidence that small islands have many more species per square meter than larger ones, and, of course, than continents. As we travel more on this planet, epidemics will be more acute—we will have a germ population dominated by a few numbers, and the successful killer will spread vastly more effectively. Cultural life will be dominated by fewer persons: we have fewer books per reader in English than in Italian (this includes bad books). Companies will be more uneven in size. And fads will be more acute. So will runs on the banks, of course.

  Once again, I am not saying that we need to stop globalization and prevent travel. We just need to be aware of the side effects, the trade-offs—and few people are. I see the risks of a very strange acute virus spreading throughout the planet.

  The Other Types of Redundancy

  The other categories of redundancy, more complicated and subtle, explain how elements of nature exploit positive Black Swans (and have an additional toolkit for surviving negative ones). I will discuss this very briefly here, as it is mostly behind my next work on the exploitation of Black Swans, through tinkering or the domestication of uncertainty.

  Functional redundancy, studied by biologists, is as follows: unlike organ redundancy—the availability of spare parts, where the same function can be performed by identical elements—very often the same function can be performed by two different structures. Sometimes the term degeneracy is used (by Gerald Edelman and Joseph Gally).

  There is another redundancy: when an organ can be employed to perform a certain function that is not its current central one. My friend Peter Bevelin links this idea to the “spandrels of San Marco,” after an essay by Steven Jay Gould. There, the necessary space between arches in the Venetian cathedral of San Marco has led to art that is now central to our aesthetic experience while visiting the place. In what is now called the spandrel effect, an auxiliary offshoot of a certain adaptation leads to a new function. I can also see the adaptation as having a dormant potential function that could wake up in the right environment.

  The best way to illustrate such redundancy is with an aspect of the life story of the colorful philosopher of science Paul Feyerabend. Feyerabend was permanently impotent from a war injury, yet he married four times, and was a womanizer to the point of leaving a trail of devastated boyfriends and husbands whose partners he snatched, and an equally long one of broken hearts, including those of many of his students (in his day, certain privileges were allowed to professors, particularly flamboyant professors of philosophy). This was a particular achievement given his impotence. So there were other parts of the body that came to satisfy whatever it was that made women attached to him.

  Mother Nature initially created the mouth to eat, perhaps to breathe, perhaps for some other function linked to the existence of the tongue. Then new functions emerged that were most probably not part of the initial plan. Some people use the mouth and tongue to kiss, or to do something more involved to which Feyerabend allegedly had recourse.

  Over the past three years I became obsessed with the notion that, under epistemic limitations—some opacity concerning the future—progress (and survival) cannot take place without one of these types of redundancy. You don’t know today what may be needed tomorrow. This conflicts very sharply with the notion of teleological design we all got from reading Aristotle, which has shaped medieval Arabic-western thought. For Aristotle, an object had a clear purpose set by its designer. An eye was there to see, a nose to smell. This is a rationalistic argument, another manifestation of what I call Platonicity. Yet anything that has a secondary use, and one you did not pay for, will present an extra opportunity should a heretofore unknown application emerge or a new environment appear. The organism with the largest number of secondary uses is the one that will gain the most from environmental randomness and epistemic opacity!

  Take aspirin. Forty years ago, aspirin’s raison d’être was its antipyretic (fever-reducing) effect. Later it was used for its analgesic (pain-reducing) effect. It has also been used for its anti-inflammatory properties. It is now used mostly as a blood thinner to avoid second (or first) heart attacks. The same thing applies to almost all drugs—many are used for secondary and tertiary properties.

  I have just glanced at the desk in my business, nonliterary office (I separate the functional from the aesthetic). A laptop computer is propped up on a book, as I like to have some incline. The book is a French biography of the fiery Lou Andreas Salomé (Nietzsche’s and Freud’s friend) that I can very safely say I will never read; it was selected for its optimal thickness for the task. This makes me reflect on the foolishness of thinking that books are there to be read and could be replaced by electronic files. Think of the spate of functional redundancies provided by books. You cannot impress your neighbors with electronic files. You cannot prop up your ego with electronic files. Objects seem to have invisible but significant auxiliary functions that we are not aware of consciously, but that allow them to thrive—and on occasion, as with decorator books, the auxiliary function becomes the principal one.

  So when you have a lot of functional redundancies, randomness helps on balance, but under one condition—that you can benefit from the randomness more than you can be hurt by it (an argument I call more technically convexity to uncertainty). This is certainly the case with many engineering applications, in which tools emerge from other tools.

  Also, I am currently absorbed in the study of the history of medicine, which struggled under this Aristotelian illusion of purpose, with Galen’s rationalistic methods that killed so many people while physicians thought they were curing them. Our psychology conspires: people like to go to a precise destination, rather than face some degree of uncertainty, even if beneficial. And research itself, the way it is designed and funded, seems to be teleological, aiming for precise results rather than looking for maximal exposures to forking avenues.

  I have given more complicated names to this idea, in addition to convexity, like optionality—since you have the option of taking the freebie from randomness—but this is still work in progress for me. The progress coming from the second type of randomness is what I call tinkering, or bricolage, the subject of my next book.

  Distinctions Without a Difference, Differences Without a Distinction

  Another benefit of duplication. I have, throughout this book, focused on the absence of practical distinctions between the various notions of luck, uncertainty, randomness, incompleteness of information, and fortuitous occurrences using the simple criterion of predictability, which makes them all functionally equal. Probability can be degrees of belief, what one uses to make a bet, or something more physical associated with true randomness (called “ontic,” on which later). To paraphrase Gerd Gigerenzer, a “50 percent chance of rain tomorrow” in London might mean that it will rain half the day, while in Germany it will mean that half the experts think it will rain, and (I am adding), in Brooklyn, that the betting market at the bar is such that one would pay 50 cents to get a dollar if it rains.

  For scientists, the treatment is the same. We use the same equation to describe a probability distribution, regardless of whether the probability is a degree of belief or something designed by Zeus, who, we believe, calls the shots. For us probabilists (persons who work with probability in a scientific context), the probability of an event, however it may be defined, is, simply, a weight between 0 and 1, called the measure of the set concerned. Giving different names and symbols would be distracting and would prevent the transfer of analytical results from one domain to another.

  For a philosopher, it is altogether another matter. I had two lunches with the (analytical) philosopher Paul Boghossian, three years apart, one upon the completion of the first edition of The Black Swan, the second upo
n the completion of this essay. During the first conversation he said that, from a philosophical point of view, it is a mistake to conflate probability as a measure of someone’s rational degree of belief with probability as a property of events in the world. To me, this implied that we should not use the same mathematical language, say, the same symbol, p, and write down the same equation for the different types of probabilities. I spent three years wondering if he was right or wrong, whether this was a good redundancy. Then I had lunch with him again, though in a better (and even more friendly) restaurant.

  He alerted me to a phrase philosophers use: “distinction without a difference.” Then I realized the following: that there are distinctions philosophers use that make sense philosophically, but do not seem to make sense in practice, but that may be necessary if you go deeper into the idea, and may make sense in practice under a change of environment.

  For consider the opposite: differences without a distinction. They can be brutally misleading. People use the same term, measuring, for measuring a table using a ruler, and for measuring risk—when the second is a forecast, or something of the sort. And the word measuring conveys an illusion of knowledge that can be severely distorting: we will see that we are psychologically very vulnerable to terms used and how things are framed. So if we used measuring for the table, and forecasting for risk, we would have fewer turkeys blowing up from Black Swans.

  Mixing vocabulary has been very common in history. Let me take the idea of chance again. At some point in history the same Latin word, felix (from felicitas) was used to designate both someone lucky and someone happy. (The conflation of happiness and luck was explainable in an antique context: the goddess Felicitas represented both.) The English word luck comes from the Germanic Glück, happiness. An ancient would have seen the distinction between the two concepts as a waste, since all lucky people seem happy (not thinking that one could be happy without being lucky). But in a modern context we need to extricate luck from happiness—utility from probability—in order to perform any psychological analysis of decision making. (True, it is hard to disentangle the two from observing people making decisions in a probabilistic environment. People may be so fearful of bad things that may happen to them that they tend to overpay for insurance, which in turn may make us mistakenly think that they believe the adverse event has a high probability.) So we can see now that the absence of such precision made the language of the ancients quite confusing to us; but to the ancients, the distinction would have been a redundancy.