Antifragile: Things That Gain from Disorder

by Taleb, Nassim Nicholas

  Not quite, Professor Dawkins. Gerd Gigerenzer et al. counter by saying that none of that is done. They write the following:

  Instead, experiments have shown that players rely on several heuristics. The gaze heuristic is the simplest one and works if the ball is already high up in the air: Fix your gaze on the ball, start running, and adjust your running speed so that the angle of gaze remains constant. A player who relies on the gaze heuristic can ignore all causal variables necessary to compute the trajectory of the ball—the initial distance, velocity, angle, air resistance, speed and direction of wind, and spin, among others. By paying attention to only one variable, the player will end up where the ball comes down without computing the exact spot.

  The same heuristic is also used by animal species for catching prey and for intercepting potential mates. In pursuit and predation, bats, birds, and dragonflies maintain a constant optical angle between themselves and their prey, as do dogs when catching a Frisbee.

  Additional examples:

  To choose a mate, a peahen uses a heuristic: Rather than investigating all peacocks posing and displaying in a lek eager to get her attention or weighting and adding all male features to calculate the one with the highest expected utility, she investigates only three or four, and chooses the one with the largest number of eyespots.

  Just like humans. Another example:

  To measure the area of a nest cavity, a narrow crack in a rock, an ant has no yardstick but a rule of thumb: Run around on an irregular path for a fixed period while laying down a pheromone trail, and then leave. Return, move around on a different irregular path, and estimate the size of the cavity by the frequency of encountering the old trail. This heuristic is remarkably precise.

  Other: Czerlinski and Gigerenzer et al. (1999), Goldstein and Gigerenzer (1999), Gigerenzer (2008).

  Makridakis, forecasting, and less is more: Makridakis et al. (1982, 1993), Makridakis and Hibon (2000), Makridakis and Taleb (2009).

  Heuristic to measure risks: Taleb, Canetti et al. (2012)—with IMF staff.

  Lindy Effects and Associated Topics

  The Lindy effect was demonstrated in Mandelbrot (1997). Initially he used it for the artistic production, bounded by the life of the producer. In our conversations toward the end of his life, I suggested the boundary perishable/nonperishable and he agreed that the nonperishable would be powerlaw distributed while the perishable (the initial Lindy story) worked as a mere metaphor. Depending on whether we condition for knowledge of the initial time, the remaining lifetime for the exponential remains constant regardless of future condition, for powerlaw increases with time since inception, by a factor of (α/1-α), where α is the tail exponent; for Gaussian or semi-Gaussian it decreases.

  Gott: Gott (1993, 1994) presented the Copernican idea but did not properly condition the probability; corrected in Caves (2000). See discussion in Rees (2003), a treatment of the paradox in Bostrom (2002).

  Survival papers and distributional properties: Often powerlaws are mistaken for exponential distributions, owing to lack of data in the tails. So I assume a priori that an exponential is likely to be powerlaw, but not the reverse, as the error in the opposite direction is vastly less likely. Pigolotti et al. (2005). For empires, Arbesman (2011), Khmaladze et al. (2007, 2010), Taagepera (1978, 1979). For firms: Fujiwara. Also Turchin (2003, 2009).

  Conditional expected time of survival across distributions: Sornette and Knopoff (1997). They show how, paradoxically, the longer one waits for an earthquake, the longer he would be expected to wait.

  Other Neomania

  Le Corbusier: Christopher Caldwell, “Revolting High Rises,” New York Times, November 27, 2005.

  Cairns and ancient measures: Cairns (2007). His work was brought to my attention by Yoav Brand, who graciously offered me his book after a lecture.

  Nonteleological design: How buildings mutate and change, Brand (1995).

  The Dog: Moral, ii. 11; 1208 b 11. “And he says that when a dog was accustomed always to sleep on the same tile, Empedokles was asked why the dog always sleeps on the same tile, and he answered that the dog had some likeness to the tile, so that the likeness is the reason for its frequenting it.”

  General and Philosophical Discussions of Medicine

  Medicina soror philosophiae: For reflective histories of medicine, Mudry (2006), Pigeaud (2006); Camguillem (1995) discussion of iatrogenics. For the spirit, Pager (1996), Bates (1995).

  Islamic medicine: Porman and Savage-Smith (2007), Djebbar (2001).

  De motu animali and attempts to mathematize medicine: In Wear (1995). Let me reiterate: math is good, the wrong math is not good.

  Ancient medicine: Edelstein (1987), Lonrig (1998). Vivian Nutton’s Ancient Medicine (Nutton [2004]) is informative, but near-silent about the empiricists, and not too detailed about ancient practices outside of a few standard treatises. More on medicine (skeptics and methodists) in the monumental Zeller (1905) or even better the superb Les Sceptiques Grecs by Brochard.

  Oranges: As they are named in Modern Greek, portokali, a corruption of “Portuguese”—further corrupted in Levantine Arabic into burduqan, and present under that name in the Sicilian dialect.

  Medical heuristics: Palmieri (2003).

  Medieval and Renaissance: French (2003).

  General history: Conrad et al. (1995), Porter (2002, 2003), Meslin et al. (2006), Kennedy (2004).

  Iatrogenics: Sharpe and Faden (1998), most complete; Illich (1995) the first movement; Hadler (2009) for the back, Duffin (1999), Welsh et al. (2011) on overdiagnosis (though no argument about noise/signal and filtering), Lebrun (1995).

  Agency and iatrogenics: Just a random example: “Surgeons do more operations if they’re on the board of surgery centers,” June 22, 2012, “The Daily Stat,” Harvard Business Review.

  More amusing historical perspective of iatrogenics: Gustave Jules A. Witkowski, 1889, Le mal qu’on a dit des médecins.

  Rationalism/Galenism: Garicia-Ballester (1995).

  Montaigne: “Mais ils ont cet heur, selon Nicocles, que le soleil esclaire leur succez, et la terre cache leur faute; et, outre-cela, ils ont une façon bien avantageuse de se servir de toutes sortes d’evenemens, car ce que la fortune, ce que la nature, ou quelque autre cause estrangere (desquelles le nombre est infini) produit en nous de bon et de salutaire, c’est le privilege de la medecine de se l’attribuer. Tous les heureux succez qui arrivent au patient qui est soubs son regime, c’est d’elle qu’il les tient. Les occasions qui m’ont guery, moy, et qui guerissent mille autres qui n’appellent point les medecins à leurs secours, ils les usurpent en leurs subjects; et, quant aux mauvais accidents, ou ils les desavouent tout à fait, en attribuant la coulpe au patient par des raisons si vaines qu’ils n’ont garde de faillir d’en trouver tousjours assez bon nombre de telles. …” [Note the detection of the attribution problem.]

  On demandoit à un Lacedemonien qui l’avoit fait vivre sain si long temps: L’ignorance de la medecine, respondit il.

  Et Adrian l’Empereur crioit sans cesse, en mourant, que la presse des medecins l’avoit tué.

  Modern alternative medicine: Singh and Edzard (2008)—they had their skin in the game, as they were sued for it.

  Homeopathy and empirical evidence: Goldacre (2007). See also the highly readable Bad Science, Goldacre (2009).

  Modern evidence-based medicine: Manual in Sacket et al. (1998). Flaws of rationalistic methods, Silverman (1999), Gauch (2009), Sestini and Irving (2009).

  Icing: Collins (2008): “There is insufficient evidence to suggest that cryotherapy improves clinical outcome in the management of soft tissue injuries.” I could not find papers saying the opposite. What benefits are proffered seem so marginal it is not even funny.

  Convexity of blood pressure: Numbers from Welch et al. (2011).

  Jensen’s inequality and pulmonary ventilators: Brewster et al. (2005), Graham et al. (2005), Mutch et al. (2007).

  Paracelsus: Interesting character as a rebel; alas, seems to h
ave been hijacked by homeopathy advocates such as Coulter (2000). Biographies in Ball (2006), Bechtel (1970), Alendy (1937).

  Immortalization: Gray (2011).

  Stendhal: Le Rouge et le noir: “La besogne de cette journée sera longue et rude, fortifions-nous par un premier déjeuner; le second viendra à dix heures pendant la grand’messe.” Chapitre XXVIII.

  Specific Medical Topics

  Note that the concern of this author is not evidence, but rather absence of it and how researchers manage such a problem. The focus is in detecting missed convexities.

  Effectiveness of low-calorie sweeteners: One gets plenty of information by looking at studies by defenders with vested interests. De la Hunty et al. (2006) shows “advantages” to aspartame, with a meta-analysis, but focusing on the calorie-in calorie-out method, not overall weight gains. But reading it closely uncovers that the core is missing: “Some compensation for the substituted energy occurs but this is only about one-third of the energy replaced and is probably [emphasis mine] less than when using soft drinks sweetened with aspartame. Nevertheless these compensation values are derived from short-term studies.” Obviously, the paper was financed by a maker of aspartame. A better study, Anderson et al. (2012), though marred with conflict of interest (authors’ support from food companies), concludes: “there is no evidence that LCS (low calorie sweeteners) can be claimed to be a cause of higher body weight in adults. Similarly evidence supporting a role in weight management is lacking.” The last sentence is the only one that I can pay attention to as it is evidence “against interest.” Had there been benefits, we would have known about them. In other words, we are incurring iatrogenics of these sweets-without-calories without evidence, as of 2012, that they even work!

  Mithridatization and hormesis: In Pliny, Kaiser (2003), Rattan (2008), Calabrese and Baldwin (2002, 2003a, 2003b). Note that they miss the convexity argument or the insight about the departure from the norm—hormesis might just be reinstatement of normalcy.

  Fasting and hormesis: Martin, Mattson et al. (2006). Cancer treatment and fasting, Longo et al. (2008), Safdie et al. (2009), Raffaghelo et al. (2010)); on yeast and longevity under restriction, Fabrizio et al. (2001); SIRT1, Longo et al. (2006), Michan et al. (2010); review work in Blagosklonny et al. (2010).

  Definition of hormesis: Mattson (2008) for local definition, Danchin et al. (2011) for more complex-systems approach.

  Aging, longevity, and hormesis: An extremely rich research; Radak et al. (2005), Rattan (2008), Cypster and Johnson (2002) for the C-elegans; Gems and Partridge (2008), Haylick (2001), Masoro (1998), Parsons (2000); for inflammation and Alzheimer’s, Finch et al. (2001).

  Bone density and load: Dook (1997) for females, Andreoli et al. (2001) for more general athletes; Scott, Khan, et al. (2008) for general exercise. Aging for females: Solomon (1997), Rautava et al. (2007); Conroy et al. (1993) for young females.

  Bone density and bicycle riding: Nichols et al. (2003), Barry et al. (2008).

  Bone density and Olympic-style weightlifting: Some “weightlifting” studies mistake the resistance exercise on machines for real naturalistic weightlifting that stresses the skeleton. Conroy et al. (1993) is a more ecologically robust study because it focuses on weight.

  Thyroid: Earle (1975).

  Cholesterol: Non-naive look, Scanu and Edelstein (2008).

  Lewontin and life expectancy: Lewontin (1993). Got idea for the potential unreliability of the Lewontin estimation and was directed to the CDC data from some article on the Web I can’t remember.

  Outdoors not sports: Rose et al. (2008). Higher levels of total time spent outdoors, rather than at sports per se, were associated with less myopia and a more hyperopic mean refraction, after adjusting for near work, parental myopia, and ethnicity.

  “Neurobabble,” “brain porn” studies: Weisberg (2008), McCabe (2008), also “neuroscience and the law,” report by the U.K. Royal Society. Note that the writer Jonah Lehrer used brain porn quite effectively, building a narrative using some loose brain story, playing the narrative fallacy to the hilt—until he was caught creating both narrative and data to back it up.

  The pressure on dentists to generate revenues: “Dental Abuse Seen Driven by Private Equity Investments,” Sydney P. Freedberg, Bloomberg News, May 17, 2012.

  Significance: Simply, people in social science should not be using statistics any more than an accountant should be given a surgeon’s knife. The problem of misunderstanding significance affects professionals. See McCloskey and Ziliak (1996), Ziliak and McCloskey (2008), Soyer and Hogarth (2011), Kahneman and Tversky (1971), Taleb and Goldstein (2012).

  Practitioners and theoreticians in mathematical finance failing to understand an elementary notion in statistics in spite of all the hype: Evidence in Taleb and Goldstein (2007).

  Missing nonlinearities of dose response: The case of radiation is rather stark, Neumaier et al. (2012). “The standard model currently in use applies a linear scale, extrapolating cancer risk from high doses to low doses of ionizing radiation. However, our discovery of DSB clustering over such large distances casts considerable doubts on the general assumption that risk to ionizing radiation is proportional to dose, and instead provides a mechanism that could more accurately address risk dose dependency of ionizing radiation.” Radiation hormesis is the idea that low-level radiation causes hormetic overreaction with protective effects. Also see Aurengo (2005).

  Statins and convexity: For instance, with statin drugs routinely prescribed to lower blood lipids, although the result is statistically significant for a certain class of people, the effect is minor. “High-risk men aged 30–69 years should be advised that about 50 patients need to be treated for 5 years to prevent one [cardiovascular] event” (Abramson and Wright, 2007).

  Statins side effects and (more or less) hidden risks: Side effects in musculoskeletal harm or just pain, Women, Speed et al. (2012). General assessment, Hilton-Jones (2009), Hu Chung et al. (2012). Roberts (2012) shows another aspect of convexity of benefits, hence harm in marginal cases. Fernandez et al. (2011) shows where clinical trials do not reflect myopathy risks. Blaha et al. (2012) shows “increased risks for healthy patients.” Also, Reedberg and Katz (2012); Hamazaki et al.: “The absolute effect of statins on all-cause mortality is rather small, if any.”

  Harlan Krumholz, Forbes, April 29, 2011:

  Problem is that drugs that improve blood test results may not lower risk. For example, many drugs that reduce LDL or raise HDL or lower blood sugar or blood pressure, do not, against all expectations, lower risk—and in some cases they increase risk.

  This is particularly true when considering treatment options to prevent a future event such as a heart attack. Unfortunately, for many drugs that affect risk factors, studies that investigate whether patients benefit are either not done or delayed. This is the case with ezetimibe, a Merck agent that reduces LDL. Because the study that will include information about patient outcomes will only be completed when ezetimibe comes off patent, we will not know how it actually affects risk for a few more years. This billion dollar drug’s approval and sales have been solely based on its effect on a blood test.

  For the fibrates, though, we are more fortunate. There are studies of patient outcomes, and fenofibrate, the Abbott drug, has been tested twice in large studies. In both, the drug failed to reduce the risk of the patients taking it even as it very effectively lowered their triglyceride levels. Most recently, in a $300 million trial by the National Institutes of Health, no benefit was shown for the Abbott drug when it was combined with a statin—compounded by a suggested harm for women. The former concern is sufficiently high to have prompted the FDA to convene an advisory committee to review the findings.

  Back: McGill (2007); iatrogenics surgery or epidural, Hadler (2009), Sayre (2010).

  Doctor’s strikes: There have been a few episodes of hospital strikes, leading to the cancellation of elective surgeries but not emergency-related services. The data are not ample, but can give us insights
if interpreted in via negativa mode. Extracting the effect of elective surgery, Argeseanu et al. (2008).

  Diabetes and pharmacological treatments (ACCORD study): The ACCORD study (Action to Control Cardiovascular Risk in Diabetes) found no gain from lowering blood glucose, or other metrics—it may be more opaque than a simple glucose problem remedied by pharmacological means. Synthesis, Skyler et al. (2009), old methods, Westman and Vernon (2008).

  Discussions of diabetes and diet: Taylor (2008), reversal in Lim et al. (2011), Boucher et al. (2004), Shimakuru et al. (2010); diabetes management by diet alone, early insights in Wilson et al. (1980). Couzin, “Deaths in Diabetes Trial Challenge a Long-Held Theory,” Science 15 (February 2008): 884–885. Diabetes reversal and bariatric (or other) surgery: Pories (1995), Guidone et al. (2006), Rubino et al. 2006.

  Autophagy for cancer: Kondo et al. (2005).

  Autophagy (general): Danchin et al. (2011), Congcong et al. (2012).

  Jensen’s inequality in medicine and workout: Many such as Schnohr and Marott (2011) got close to dealing with the fact that extreme sprinting and nothing (as a barbell) outperforms steady exercise, but missed the convexity bias part.

  Art De Vany and Jensen’s inequality: Art De Vany, private correspondence: “Tissue gains are increasing but convex with nutrient intake (the curve is rising, but at a diminishing rate). This has to be the case for the point of origin to be a steady state solution. This implies that weight gain, including fat, is higher at the average intake than it is on a varying intake of the same calories and nutrients. Muscle and fat compete for substrate, so a fatter person will shift nutrient partitioning toward muscle because body fat induces insulin resistance in muscle. Insulin operates in a pulsate release and is far more effective with that pattern than with the chronic elevation induced by six meals a day. On the downside, where fat and muscle are lost, the curve is negatively sloped but declines at a diminishing rate (concave). This means you lose more fat feeding intermittently than continuously. The loss at the average intake (six per day keeps the variation of the average small) is less than the loss at the same intake but one that varies between a small intake and a large one. A more subtle point: you lose more weight when you eat at the average than intermittently, but that is because you lose more muscle in chronic deprivation than intermittent deprivation. Intermittent eating yields a superior body composition.”