Chapitre 1
When Certainty Becomes a Liability
Have you ever confidently predicted an outcome only to be completely wrong? Perhaps you were certain your favorite team would win, or that a job interview went brilliantly, only to face disappointment. What if these failures weren't just bad luck, but symptoms of a crucial cognitive skill you've never developed? This is the premise behind Dylan Evans' groundbreaking work on risk intelligence-our ability to accurately gauge probabilities in an uncertain world. Unlike traditional intelligence measures, risk intelligence isn't about knowing more facts; it's about knowing the limits of your knowledge. The book has become required reading in finance, military planning, and intelligence communities, with Warren Buffett reportedly calling it "the most important skill no one taught me." Even Elon Musk referenced it when discussing Tesla's risk assessment protocols, noting that "understanding probability is the foundation of rational decision-making."
Chapitre 2
The Essence of Risk Intelligence
Risk intelligence is fundamentally about estimating probabilities accurately. It's not just relevant for gamblers or stock traders-we use it daily when deciding whether to carry an umbrella, trust a news source, or make a career change. The truly risk intelligent person knows precisely how much they know and how much they don't. They're able to quantify uncertainty and assign realistic probability estimates to different outcomes, maintaining appropriate confidence levels based on available information.
Consider several professionals who demonstrate varying levels of risk intelligence: Kathryn, a detective who assigns precise confidence levels to her theories based on available evidence, scoring each piece of information from 1-10 and adjusting her overall case confidence accordingly. Jamie, a banker who carefully calibrates loan risk assessments by examining not just credit scores but also market conditions, industry trends, and historical default rates in similar circumstances. Dr. Chen, a physician who expertly balances treatment benefits against potential side effects, always considering multiple scenarios and their likelihood. These individuals demonstrate high risk intelligence by matching their confidence to their actual knowledge.
Contrast them with examples of poor risk intelligence: Diane, who consistently overestimates her understanding of relationship dynamics, making repeated dating mistakes based on false assumptions. Jeff, who chronically underestimates his military expertise despite years of experience, missing opportunities to lead important missions. Sarah, an investor who ignores market volatility data and puts all her savings into a single stock, displaying dangerous overconfidence. These cases illustrate how risk intelligence affects outcomes across diverse domains.
This skill matters enormously because we live in a probabilistic world where few things are certain. Critical decisions-from medical treatments to national security policies-must be made under uncertainty. The consequences of poor risk intelligence can be devastating, as illustrated by multiple historical examples: the 2008 financial crisis, where overconfident bankers misjudged mortgage default probabilities; the Challenger disaster, where engineers' risk assessments were overruled; and numerous medical misdiagnoses where doctors were too certain of their initial impressions.
What makes risk intelligence particularly fascinating is that it doesn't correlate strongly with traditional intelligence measures. Studies show that people with high IQs often display poor probability judgment. Research with horse racing handicappers demonstrates this disconnect-their ability to estimate race outcomes showed no correlation with IQ scores but improved dramatically with experience in their specific domain. Similarly, weather forecasters develop excellent calibration through constant feedback, while many brilliant academics show poor ability to estimate research completion times.
The good news? Unlike IQ, risk intelligence can be significantly improved through deliberate practice and awareness of our cognitive biases. Specific techniques include keeping a calibration journal, regularly reviewing past predictions, seeking disconfirming evidence, and explicitly stating confidence levels for important decisions. Organizations can improve collective risk intelligence by encouraging open discussion of uncertainties, maintaining prediction tournaments, and creating environments where expressing doubt is valued rather than punished. By understanding when we're likely to be overconfident or underconfident, we can calibrate our judgments more accurately and make better decisions in an uncertain world.
Chapitre 3
The Twilight Zone of Knowledge
Imagine your mind as a room illuminated by a single light bulb. Some objects stand in bright light-things you know with certainty, like your name or what you had for breakfast. Others remain in complete darkness-things about which you know absolutely nothing, like the exact population of a remote village in Mongolia. But between these extremes lies a vast twilight zone where most of our knowledge exists-areas where we have partial information but significant uncertainty.
The challenge of risk intelligence is accurately judging how much we truly know about these partially illuminated subjects. Do we have 30% of the relevant information? 70%? Our ability to make this assessment determines whether we'll be appropriately confident or dangerously overconfident.
Most people struggle with this twilight zone. Research shows we're particularly uncomfortable with ambiguity-the state of partial knowledge where certainty isn't possible. This "ambiguity intolerance" manifests as a psychological aversion to uncertainty that drives us toward false certainty. We'd rather have any answer, even a wrong one, than acknowledge the limits of our knowledge.
This discomfort creates opposing psychological forces. The "need for closure" pulls us toward certainty, making any answer preferable to ambiguity. Meanwhile, the "need to avoid closure" pulls us toward uncertainty, making no answer seem satisfactory. People with high risk intelligence balance these forces perfectly, while most of us skew toward one extreme or the other.
When the need for closure dominates, our probability estimates cluster at the extremes-0% and 100%-creating overconfidence. When the need to avoid closure prevails, our estimates huddle around 50%, indicating underconfidence. These tendencies can be measured through agreement with statements like "I hate to change my plans at the last minute" versus "I like to have friends who are unpredictable."
These psychological tendencies explain why many people engage in "worst-case thinking," substituting imagination for analysis and fear for reason. Dick Cheney's infamous "one percent doctrine"-treating a 1% chance of catastrophe as certainty-exemplifies this fallacy. This thinking ignores cost-benefit analysis and leads to terrible decisions, as seen after the Three Mile Island accident when fear prevented nuclear plant construction for thirty years, leading to more harmful coal and oil plants instead.
Similarly, the "all-or-nothing fallacy" skews probability estimates toward extremes by framing concepts like proof, knowledge, and belief in binary terms. This manifests in phrases like "You can't prove that" or "We cannot predict what will happen," which set impossibly high thresholds of absolute certainty. This fallacy renders everyday concepts unusable and has serious consequences, as seen in the MMR vaccine controversy where parents refused immunization despite overwhelming safety evidence.
Learning to feel comfortable with uncertainty-developing what poet John Keats called "negative capability"-is crucial for risk intelligence. This means embracing the twilight zone rather than fleeing from it.
Chapitre 4
Cognitive Illusions and Risk Blindness
Our minds employ numerous cognitive shortcuts-heuristics-that once served evolutionary purposes but now often undermine our risk intelligence. Like optical illusions that persist even when we know they're illusions, these psychological mechanisms can deceive us despite contradicting reality.
The availability heuristic causes us to estimate probabilities based on how easily examples come to mind. While this worked well for our ancestors, modern media distorts this correlation by disproportionately reporting dramatic events. Studies show media reporting bias (like reporting every ecstasy death but only 1 in 250 paracetamol deaths) skews our risk perception. Even imagining events can cause "imagination inflation," making them seem more likely to occur or to have occurred in the past.
Wishful thinking-allowing desires to influence beliefs-violates the principle that probabilities should be independent of outcomes. Research consistently shows an optimism bias where people overestimate positive outcomes (job offers, exam scores, marriage success) and underestimate negative ones. Even depressed individuals exhibit "depressive realism" but may still make overly optimistic predictions by failing to account for how their depression will affect outcomes.
This optimism bias fueled the pre-2007 asset bubble, with lenders and borrowers convinced property prices would rise indefinitely. To counter this bias, organizations can develop corrective procedures like the UK government's "optimism bias uplifts" for infrastructure projects, or we can personally adjust estimates-doubling time estimates for projects or halving revenue projections in business plans.
Confirmation bias-our tendency to embrace supporting evidence while ignoring contradictory data-severely undermines risk intelligence. Francis Bacon recognized this in 1620, noting how the mind "draws all things else to support and agree with it" once an opinion forms. Peter Wason's famous sequence experiment (2,4,6) demonstrated this bias, as students typically tested only confirming hypotheses.
Koriat's groundbreaking research showed that risk intelligence dramatically improves when people deliberately seek reasons why they might be wrong, not just supporting evidence. Surprisingly, only participants asked to provide contradicting reasons significantly improved their calibration. This suggests seeking diverse, opposing viewpoints is crucial-yet increasingly difficult in our digital age where "filter bubbles" algorithmically reinforce existing beliefs.
Hindsight bias-our tendency to claim we "knew it all along" after events unfold-devastates risk intelligence by preventing us from learning from mistakes. Fischhoff and Beyth demonstrated this in 1972 during Nixon's China visit; participants misremembered their predictions, inflating probabilities for events that occurred and deflating those that didn't. This bias robs us of genuine surprise-that mental jolt essential for learning. Recording predictions systematically offers the best defense against this self-deception.
Perhaps most insidious is the mind-reading illusion-our misplaced confidence in reading others' thoughts and detecting deception. Research consistently shows our lie-detection abilities barely exceed chance, yet our confidence remains stubbornly high. Even worse, training programs like the widely-used Reid technique can actually decrease accuracy while increasing confidence. As Stephen King chillingly illustrates in "A Good Marriage," this misplaced certainty can mask profound ignorance about those closest to us: "His insanity was like an underground sea... You could stroll through [the flowers] and never know the madwater was there."
Chapitre 5
The Social Dynamics of Overconfidence
While overconfidence undermines risk intelligence, it persists because it confers social benefits-people often mistake confidence for competence. This social dynamic may have evolutionary roots, as our ancestors likely trusted reassuring, confident leaders over more measured ones.
Social pressure to appear confident can undermine risk intelligence by discouraging careful deliberation. We often equate quick responses with intelligence ("quick-witted"), while those who pause to think may be perceived as less intelligent. Research using the Brief Loquaciousness and Interpersonal Responsiveness Test (BLIRT) shows that "blirtatious" people-those who respond quickly and effusively-initially impress others as more intelligent and competent. However, this advantage fades over time as people realize their exuberance often exceeds their insight.
Social pressure also undermines risk intelligence through our tendency to follow crowds. When people with similar views gather, they form self-reinforcing groups where each person's conviction strengthens others' while contrary evidence is blocked-creating conditions for the "madness of crowds" Charles Mackay described in 1841. His accounts of economic bubbles, like the Dutch tulip mania of the 1630s, remain relevant today. During this speculative frenzy, tulip bulbs briefly became the world's most expensive objects before the market collapsed.
How we communicate about risk significantly impacts our risk intelligence. Many well-intentioned efforts actually muddy the waters rather than clarify them. The Homeland Security color-coded threat system, implemented after 9/11 and abandoned in 2011, exemplifies this problem-it provided "little practical information," never used its lowest threat levels, and remained at "Yellow" (elevated risk) for six years.
Ambiguous language severely undermines risk communication. The credit rating agencies' use of letter grades (AAA, AA+) instead of precise numerical probabilities exemplifies this problem. Only when we transform these ambiguous labels into numbers can we see just how wildly optimistic their ratings were before the financial crisis. This contributed to CalPERS winning court backing to proceed with fraud suits against Moody's, S&P, and Fitch for "wildly inaccurate" ratings.
Legal standards of proof suffer from the same vagueness problem. Terms like "beyond reasonable doubt" and "balance of probabilities" are interpreted wildly differently by judges and jurors. Studies reveal that judges interpret "beyond reasonable doubt" as meaning about 89% certainty, while jurors averaged 83%. For civil cases, judges interpreted "balance of probabilities" as requiring 61% certainty, while jurors set a much higher threshold of 75%. This creates an illusion of communication that threatens due process.
Chapitre 6
The Power of Numerical Thinking
In 1682, Swiss mathematician Jacob Bernoulli revolutionized probability by representing it numerically between zero and one-a radical concept at the time when likelihood was expressed only through vague verbal labels. This innovation enabled the application of mathematical tools to probability, spawning an entire field that continues to evolve today.
The early Lloyd's of London insurance market illustrates the contrast; 17th century insurers relied solely on intuitive risk intelligence, pricing policies case-by-case without the mathematical tools or mortality tables we now take for granted. The development of probability theory and statistics over centuries represents a gradual scientific revolution that continues accelerating with computing power.
Modern statistical methods now challenge expert intuition across diverse fields-from Orley Ashenfelter's Bordeaux wine price prediction formula based on just three weather variables to computerized horse race handicapping systems that outperform human experts. Frank Singer capitalized on this trend, raising $800,000 to develop his own system in Hong Kong. After two years developing a complex model with over 100 variables, Singer achieved 30% profits, sometimes earning tens of millions yearly.
Despite their sophistication, computer models have serious limitations. The 2008 financial crisis partly stemmed from overreliance on models that created "idiot savants" with false authority. Similarly, during the UK's 2001 foot-and-mouth disease outbreak, poorly validated models using flawed data drove the culling of 10 million animals. Dr. Paul Kitching criticized how "seductive graphs" from modelers unfamiliar with the virus guided policy without expert input. The database itself was riddled with errors, including farms supposedly "situated in the North Sea."
Mastering probability theory is neither necessary nor sufficient for risk intelligence. Many skilled handicappers rely on intuition rather than formal probability calculations, while mathematically gifted "nerds" often lack judgment in estimating probabilities. Even in casinos, where probability theory seems most applicable, real-world conditions differ from textbook ideals. As Nassim Nicholas Taleb notes, "computable risks are largely absent from real life!"
Epistemic feelings-gut feelings about what we know-blur the traditional distinction between reason and emotion. These feelings are crucial for risk intelligence, facilitating communication between conscious and unconscious components of probability estimation. High risk intelligence depends on two factors: well-calibrated epistemic feelings (which accurately reflect your knowledge level) and the ability to translate these feelings into numbers.
Chapitre 7
Decision-Making Under Uncertainty
The challenge isn't just generating accurate probability estimates but knowing how to act on them. One simple approach is setting probability thresholds for action-deciding to take an umbrella only when rain probability exceeds 65%, or military commanders bombing a building only when there's over 80% certainty it houses insurgents. This method proved effective in Bangladesh's flood warning system, where warnings were issued only when flooding probability exceeded 80%.
Bet sizing represents another approach to using probabilities in decision-making. Expert gamblers recognize their finite resources and size bets appropriately, using methods like the Kelly criterion which ensures bets never exceed a certain fraction of one's bankroll. Another complementary strategy is making bet sizes proportional to confidence levels, as seen in blackjack card counting systems where players adjust bets based on running counts.
To properly size bets, we must accurately assess our confidence levels-this requires well-calibrated epistemic feelings. Like reading a thermometer, we need precise probability notches from 0-100%. While some argue against single-number probabilities as falsely precise, using probability ranges creates the same problem at a deeper level. Probability estimates are like physical measurements-they have inherent imprecision but remain useful.
The 100 percent rule states that probabilities assigned to mutually exclusive possibilities must sum to 100%. This helps calibrate estimates by weighing possibilities against each other. In a murder mystery experiment, participants not instructed about this rule consistently assigned probabilities totaling more than 100% to suspects. As suspects are eliminated, the remaining suspects' probabilities must increase proportionally.
Bayes's theorem provides a precise mathematical formula for how beliefs should be updated when new information arrives. Discovered by Reverend Thomas Bayes in 18th-century England, the theorem calculates the posterior probability of a hypothesis after considering new evidence. While early studies suggested people don't naturally follow Bayesian principles, recent research shows we're more Bayesian when dealing with familiar contexts. Griffiths and Tenenbaum found that people intuitively account for different statistical distributions when predicting movie run times versus box-office earnings.
Chapitre 8
The Rational Gambler's Approach to Life
Expected utility theory provides a mathematical framework for rational decision-making by analyzing choices as gambles with potential gains and losses. Just as Galileo's inclined plane experiments isolated gravity for study, gambling strips decision-making to its essence. The theory quantifies all outcomes-even intangibles like happiness and health-into "utility points" and calculates expected value by multiplying potential outcomes by their probabilities.
Every risky choice can be analyzed through four essential elements: the status quo (current situation), potential gain (reward if successful), potential loss (cost of failure), and chance of winning (probability of success). By converting all outcomes to utility points and calculating expected value, we can determine whether a gamble is worth taking.
Thirteen-year-old Hannah Jones' decision to initially refuse a heart transplant demonstrates how expected utility calculations can clarify life-or-death choices. By weighing the potential benefits (a functioning heart, longer life) against the risks (painful procedures, possible failure, dying away from family), Hannah made what she considered a rational choice. Six months later, when doctors determined the operation carried less risk, she recalculated and chose to proceed with the successful transplant.
Expected utility theory can be applied to major political decisions like the 2003 Iraq invasion. A rational approach would require listing potential benefits (removing Saddam Hussein, establishing military bases, securing oil reserves, spreading democracy) and assigning probabilities and utilities to each outcome. Similarly, potential costs must be assessed (casualties, financial costs, erosion of global support). The theory doesn't reveal an objectively optimal course, but forces transparency in reasoning.
The same approach can help anticipate terrorist actions by modeling their decision-making process. A terrorist leader might calculate benefits (American casualties, economic damage, fear generation, recruitment) against costs (increased US support, loss of operatives, financial costs). Current terrorism risk assessment often neglects terrorists' values and beliefs, assuming irrationality rather than different utility calculations.
Expected utility calculations in healthcare face serious concerns. The standard gamble method assumes people have good intuition about probabilities and doesn't account for variations in risk appetite. Most troubling, this approach assumes people can predict their feelings about health conditions, despite evidence showing "durability bias"-people consistently overestimate how long emotional reactions will last. Studies show even those with locked-in syndrome (near-total paralysis) often adapt and report happiness, with 47 of 65 patients in one study professing happiness and only 7% expressing a wish for euthanasia.
Chapitre 9
The Limits of Knowledge
Accurately assessing the limits of our knowledge is fundamental to risk intelligence. The danger lies not just in what we don't know, but in failing to recognize that gaps in our knowledge exist at all-what Donald Rumsfeld famously called "unknown unknowns." These are answers to questions we haven't even thought to ask, like the revolutionary theories of relativity and quantum mechanics that physicist Albert Michelson failed to anticipate when he declared physics nearly complete in 1894.
When estimating probabilities, we must consider not just the ratio of what we know to what we know we don't know, but also account for information we don't even realize we're missing. This fundamental limitation means we can never provide completely reliable probability estimates outside idealized scenarios. Real-world problems often involve "Outside Context Problems" or "black swans"-unexpected events with massive impact that couldn't be anticipated.
The Dunning-Kruger effect shows how incompetence creates a double burden: poor performance coupled with inability to recognize that poor performance. Paradoxically, as people become more competent in a domain, they become more aware of their limitations-converting unknown unknowns into known unknowns. This can lead the most skilled to underrate their abilities while the least skilled remain confidently ignorant. As Darwin noted, "Ignorance more frequently begets confidence than does knowledge."
Expert gamblers demonstrate this awareness of their blind spots, understanding the boundaries of their knowledge. J.P. McManus exemplifies this trait-a billionaire who built his fortune first on horse betting then currency trading, yet remains brutally honest about his limitations. When playing backgammon, he would make deliberate mistakes to test opponents, stopping if they played well. Like other expert gamblers, he knew when not to bet and maintained detailed records of wins and losses to learn from mistakes.
While much attention focuses on unknown unknowns, philosopher Slavoj Zizek highlights the equally important "unknown knowns"-information we possess but fail to use when solving problems because we don't recognize its relevance. These are knowledge fragments trapped in mental silos, disconnected from problems where they might prove useful. Risk intelligence involves liberating these facts from their information prisons by recognizing unexpected connections.
Physicist Enrico Fermi developed a technique for transforming unknown knowns into known knowns through estimation problems. By breaking seemingly impossible questions (like "How many piano tuners are in Chicago?") into manageable subproblems, we discover we know more than we realized. This method involves establishing upper and lower bounds, then calculating reasonable estimates. Companies like Microsoft and Goldman Sachs now use "Fermi questions" in interviews to test candidates' ability to leverage existing knowledge for novel problems.
Even perfect risk intelligence cannot guarantee success in a probabilistic world. Machiavelli claimed chance governs half our actions, comparing fortune to a violent river that overwhelms everything in its path. We must accept chance's irreducible nature while making the best probabilistic decisions possible. As Damon Runyon noted, "The race is not always to the swift, nor the battle to the strong-but that's the way to bet."
