Key Takeaways
Calibrate your confidence, not merely your knowledge.
The real skill. Risk intelligence is the ability to estimate probabilities in line with the evidence, including recognizing when knowledge is thin. It is not the same as IQ, factual expertise, or appetite for risk. A cautious person and a bold gambler can both be risk intelligent if they judge the odds accurately. In the book's opening examples, detective Kathryn avoids declaring a suspect a liar too quickly, while Diane repeatedly predicts near-certain success for new relationships that soon end.
Measure the match. A calibration curve compares stated confidence with what proves true across many judgments. If a forecaster assigns 70 percent to many events, about 70 percent should occur. An online test taken by 14,294 participants produced an average RQ score of about 64, though the self-selected sample should not be mistaken for a population-wide measure.
What's valuable here is the separation of judgment quality from personality. Confidence is not inherently a virtue, and caution is not inherently wisdom. Calibration makes the question testable: do confidence levels correspond to outcomes over repeated cases? The approach also suggests a useful distinction between a person's overall style and performance in a particular domain. A single score can conceal both. Any assessment should therefore be treated as a diagnostic starting point, not a permanent label or a substitute for examining the decisions that matter.
Practice forecasts where feedback is frequent, specific, and unavoidable.
Repetition teaches odds. Weather forecasters have long been required to estimate rain probabilities, and they receive clear feedback soon after each forecast. When forecasters assigned a 90 percent chance of rain, rain occurred about 90 percent of the time. In one study, doctors who estimated a 90 percent chance that a patient had pneumonia were correct only about 15 percent of the time. The difference is not simply intelligence: forecasters repeatedly answer a narrow question, while medical diagnoses are varied and outcomes may be hard to verify.
Train the judgment. Royal Dutch Shell gave geologists estimates from earlier drilling projects, then showed them which wells actually found oil. Their forecasts improved: a 40 percent estimate came to correspond to roughly four successful wells in ten. The practical lesson is to make predictions measurable, record them, and compare them with results.
This is a strong account of learning as a designed feedback loop. Expertise grows not just through exposure, but through repeated decisions whose outcomes can correct an internal model. That principle applies to hiring, project planning, and clinical work, where feedback is often delayed or selectively visible. Calibration also needs enough cases: one forecast can be unlucky, while a pattern across many forecasts is informative. Training should therefore reward honest estimates and accurate updating, rather than punishing every miss as if uncertainty itself were incompetence.
Stay in probability's gray zone when certainty is a comfort reflex.
Uncertainty has two traps. Some people rush toward closure, treating uncertain claims as certainly true or false. Others avoid commitment so completely that they cannot act. Risk intelligence means staying between those poles and expressing degrees of belief instead of forcing every question into a yes-or-no answer. The book calls the discomfort with ambiguous situations ambiguity intolerance. It can make people prefer a definite answer, even a bad one, to a measured estimate.
Fear can erase odds. The so-called one percent doctrine treats a frightening one-in-a-hundred possibility as if it were certain when choosing a response. That discards the probability and the costs of the response itself. A serious threat may justify action, but the chance of harm and the chance of harm from intervention both belong in the calculation.
The ability to tolerate ambiguity is not passivity. It is a way to keep judgment open long enough for evidence to matter, while still making decisions under imperfect knowledge. This is especially important when vivid danger drives attention toward worst-case outcomes. Yet numerical estimates can also create a false sense of mastery if their assumptions are hidden. The useful discipline is not to replace emotion with arithmetic, but to make the uncertainty and the trade-offs visible enough to debate.
Test vivid fears against base rates and overlooked benefits.
Vivid is not frequent. The availability heuristic is the mental shortcut of estimating likelihood by how easily examples come to mind. News images of plane attacks or dramatic kidnappings can make rare events feel common; ordinary hazards that attract little coverage can seem negligible. Imagination can have a similar effect: picturing an event may make it feel more likely, even when the image is not evidence that it will happen.
Count the full cost. After airport security tightened, some travelers drove instead of flying. One estimate attributed 2,300 additional road deaths between September 2001 and October 2003 to that shift, because driving was more dangerous than flying. The point is not that airport precautions had no value. It is that a visible response can reduce one risk while increasing another, so the comparison must include consequences on both sides.
The key move is to ask what is missing from attention. A vivid threat deserves consideration, but vividness is a property of memory, not a reliable measure of frequency. The same scrutiny should apply to familiar policies: their visible activity can make benefits feel certain and indirect costs disappear. Risk comparison is most useful when it avoids a false choice between panic and dismissal. Some rare hazards warrant preparation, but the response should be judged against realistic alternatives, including the risks created by doing nothing or doing too much.
Make your strongest belief search for the evidence against it.
Agreement comes easily. Confirmation bias is the tendency to seek information that supports a belief while neglecting evidence that could disprove it. In psychologist Peter Wason's number puzzle, people saw 2, 4, 6 and tested sequences that fit their guess, missing the simpler rule that any ascending sequence qualified. They were confident, but many had not tried to falsify their theory.
Contradiction improves judgment. In experiments by Asher Koriat and colleagues, asking people to give a reason against their chosen answer improved calibration more than asking only for reasons in its favor. The Iraq WMD case illustrates why this matters beyond puzzles: a source's claims can sound compelling while the source itself is unreliable. Strong evidence and credible evidence are different questions. Before settling on a conclusion, identify what would change the estimate and actively look for it.
A counterargument is most useful when it has a real chance to alter the conclusion, rather than serving as a ritual disclaimer. This distinction matters in polarized debate, where each side can quote an opposing view only to dismiss it. A better test is whether the strongest contrary evidence has been represented fairly and whether its credibility has been assessed separately from its emotional force. This method does not require treating all claims as equally plausible. It requires resisting the habit of granting one's preferred claim an easier standard of proof.
Record predictions before outcomes rewrite what you supposedly knew.
Memory edits confidence. Hindsight bias makes past events seem more predictable after they have happened. In a study of entrepreneurs whose businesses failed, participants had originally estimated an average 80 percent chance of success, but later recalled their estimates as about 50 percent. That revision makes failure look anticipated and strips the mistake of its lesson.
Keep a forecast log. Write down a clear prediction, its probability, and the date by which it can be judged. Avoid elastic claims such as expecting a generally good outcome. When the date arrives, record whether the prediction came true, then compare the result with the original probability. The record makes it harder to improve one's past judgment in memory and easier to notice recurring overconfidence, underconfidence, or faulty assumptions.
A forecast log turns self-criticism into evidence rather than a mood. It also separates a bad outcome from a bad decision: a reasonable 70 percent forecast can fail, while a reckless 20 percent forecast can succeed by luck. That distinction is central to learning under uncertainty. The method works best when predictions are chosen before the outcome is obvious and when missed forecasts are reviewed without blame. If only embarrassing failures are recorded, the log becomes another biased story rather than a useful measurement system.
Replace risk adjectives with probabilities others can audit.
Words hide disagreement. Terms such as likely, guarded, or serious possibility do not have one stable numerical meaning. In one reported case, two analysts understood the phrase that a cease-fire could break within a week as roughly a 30 percent chance and an 80 percent chance. Both thought they agreed. A numerical estimate would have exposed the gap.
Certainty can mislead courts. Fingerprint evidence is not infallible, yet it has often been presented as categorical. In Scotland, policewoman Shirley McKie was accused of leaving a print at a murder scene; outside experts concluded the print was not hers, and a jury acquitted her of perjury. The case illustrates the stakes of presenting expert judgment as certainty. Numerical estimates do not eliminate error, but they reveal how much uncertainty remains and allow claims to be checked against outcomes.
Quantification is valuable here as a tool for accountability and communication, not as decoration. A number forces a speaker to distinguish weak evidence from strong evidence and gives listeners a chance to challenge the estimate. Still, a precise percentage can be misleading if the underlying evidence, reference group, or method is unclear. Better communication therefore pairs numbers with reasons and limitations. The aim is not to turn every human judgment into a seemingly objective statistic, but to make disagreement explicit rather than hiding it in shared-sounding language.
Update beliefs from prior odds, not the latest dramatic clue.
Start with prevalence. Bayes's theorem describes how to revise a belief when new evidence arrives. A positive test result matters, but its meaning depends partly on how common the condition was before testing. In the book's example, HIV prevalence among English men is set at 0.2 percent, the test detects 99 percent of cases, and 5 percent of uninfected people test positive. A positive result raises the estimated chance of infection to only about 4 percent, not nearly 100 percent, because false positives greatly outnumber true positives when the condition is rare.
Use a reference class. When estimating a project deadline, start with the time required by comparable past projects, not just the appealing details of the new one. This is reference class forecasting. Also check that probabilities assigned to mutually exclusive outcomes add up to 100 percent.
The base-rate example is memorable because it shows how a powerful piece of evidence can still leave a low absolute probability when the event is rare. The same logic applies to screening, fraud alerts, and other systems that generate many warnings. The challenge is selecting a relevant reference class: too broad a group obscures important differences, while too narrow a group may offer too few cases. The best estimate is often provisional, with assumptions stated plainly and revised when evidence supports a better comparison.
Compare expected gains and losses before taking consequential risks.
Make trade-offs explicit. Expected utility is a method for comparing a risky choice with the status quo. Estimate the possible gains and losses, assign each a value based on the decision-maker's priorities, multiply each by its probability, and combine the results. A positive total favors taking the gamble under those assumptions; a negative one favors not taking it. The method can clarify decisions involving money, health, or time, not only casino bets.
A decision can be rational without being easy. Thirteen-year-old Hannah Jones refused a heart transplant because she judged the treatment's burdens and uncertainties against the possibility of more healthy life. She later changed her mind when doctors thought the operation had become less risky. Her case shows why estimates must be revisable. Thresholds can guide action, and bet sizing can limit exposure, but neither makes uncertain outcomes safe.
Expected utility provides structure, not an oracle. Its strength is that it surfaces hidden assumptions about probability, value, and opportunity cost. Its weakness is that people can disagree sharply about the value assigned to life, pain, time, or social consequences, and a single total can conceal those conflicts. In public decisions, scenario analysis can help expose interactions that a simple sum misses. The method is most defensible when its inputs remain visible, contestable, and open to revision as circumstances change.
Map what you know, then leave room for what you missed.
Knowledge has blind spots. The book distinguishes known facts, recognized gaps, and unknown unknowns: relevant matters that have not even been identified as questions. Physicist Albert Michelson expected physics to need only minor refinements, overlooking emerging problems that helped lead to relativity and quantum mechanics. A confident inventory of known questions can still omit the most important question.
Estimate instead of freezing. Fermi problems, named for physicist Enrico Fermi, break an unanswerable-looking question into smaller estimates. To approximate the number of piano tuners in a city, estimate its population, piano ownership, and how many instruments one tuner can service. Keep the assumptions visible and revise them with new information. Good risk judgment improves decisions; it does not control chance or guarantee that an estimate will be right.
This closing principle balances intellectual humility with practical action. Awareness of ignorance can prevent false certainty, but it can also become an excuse to avoid estimating anything. Decomposition offers a middle path: make the best available estimate, expose its assumptions, and identify which missing information could change it. The approach is especially useful when exact data are unavailable, though rough estimates should not be mistaken for measured facts. Its value lies less in producing a perfect answer than in making reasoning inspectable and improvable.
Analysis
Risk Intelligence is a practical argument for treating confidence as something that can be tested. Its central move is to shift attention from whether a person is knowledgeable to whether their expressed certainty tracks what they know. The book builds that case through forensic errors, medical diagnoses, financial models, forecasting studies, and gambling. The varied examples reinforce a single lesson: uncertainty is ordinary, but institutions often disguise it with confident language, vivid narratives, or authoritative-looking tools.
The strongest contribution is its emphasis on calibration and feedback. A forecast becomes useful not merely because it is numerical, but because it can be compared with later outcomes. That principle makes the book's advice unusually operational: state an estimate, define what would count as success, preserve the prediction, and review the error. The distinction between risk intelligence and risk appetite is equally important. Knowing the odds does not determine whether a person should accept them; values and tolerance for loss still matter.
The method also has limits. Calibration is clearest across repeated, comparable judgments, not one-off decisions where the outcome may be unknowable or the situation unique. A score from a general knowledge test may not transfer to medical, legal, or financial expertise, and a well-calibrated estimate can still be based on a poor model of what matters. Numerical precision can even lend weak assumptions an undeserved authority. Expected utility adds structure, but its usefulness depends on how gains and losses are framed and valued.
The book is most persuasive when treated as a discipline of transparency rather than a promise of rational mastery. Probability estimates cannot remove political incentives, institutional hierarchies, or the emotional appeal of certainty. They can make those forces easier to detect by showing where evidence ends and judgment begins. The deepest lesson is not that every decision can be calculated cleanly. It is that decisions improve when uncertainty is expressed, tested, and allowed to change minds.
Review Summary
Risk Intelligence receives mixed reviews, with ratings ranging from 1 to 5 stars. Positive reviews praise its insightful exploration of decision-making and risk assessment, highlighting the book's balance of academic knowledge and practical examples. Critics find it rambling and lacking in clear frameworks for improving risk intelligence. Some readers appreciate the mathematical concepts and thought-provoking ideas, while others find it boring or politically biased. The book's unique approach to measuring risk intelligence through probability assessments intrigues many readers, despite its occasional lack of focus.
People Also Read
Glossary
Risk intelligence
Accurate probability judgmentThe ability to estimate probabilities in a way that matches the evidence and the limits of one's knowledge. It applies to both opportunities and dangers. It differs from factual expertise, general intelligence, and risk appetite, which describes how willing someone is to accept risk.
RQ
Risk intelligence scoreA numerical summary of performance on a risk intelligence test. The book's test asks people to assign probabilities to statements whose truth is known, then scores how closely their confidence matches the results. An RQ score is a headline measure, not a complete description of a person's judgment in every field.
Calibration curve
Confidence compared with outcomesA graph showing whether events assigned a given probability occur at about that frequency. For example, among many judgments rated 70 percent likely, roughly 70 percent should prove true for perfect calibration. The curve can reveal overconfidence or underconfidence that a single overall score cannot explain.
K factor
Use of intermediate probabilitiesAn indicator used alongside the book's RQ test to show whether someone made meaningful attempts to judge uncertainty. It gives credit for estimates between 0, 50, and 100 percent, rather than relying exclusively on those three categories. A low K factor makes an apparently strong RQ score less informative.
Ambiguity intolerance
Discomfort with uncertain situationsA tendency to experience novelty, complexity, or uncertainty as uncomfortable or threatening. It can encourage black-and-white judgments, premature conclusions, or avoidance of judgment altogether. The book presents it as a barrier to estimating probabilities with appropriate nuance.
Availability heuristic
Recall used to judge likelihoodA mental shortcut in which people estimate how likely an event is by how easily examples come to mind. It can work when memory reflects actual experience, but media coverage and vivid imagery distort the relationship. Dramatic rare events may feel common, while ordinary hazards remain easy to overlook.
Imagination inflation
Imagining raises perceived likelihoodThe tendency for imagining an event to increase confidence that it is likely to occur or may have occurred in the past. Because imagined and experienced events can become confused in memory, repeated visualization can affect predictions and recollections without adding evidence.
All-or-nothing fallacy
Treating belief as binaryThe mistake of treating proof, knowledge, or belief as if each requires absolute certainty. This makes ordinary degrees of confidence seem illegitimate and encourages either categorical claims or refusal to judge. The book argues that probability estimates express useful shades of belief between certainty and ignorance.
Reference class forecasting
Forecasting from comparable casesA method for estimating a future outcome by examining how long or how often similar cases took or occurred in the past. For a new project, it means starting with the record of comparable projects rather than relying only on the distinctive details and optimistic plan for the current one.
Expected utility
Probability-weighted value of outcomesA framework for comparing choices by assigning values to possible gains and losses, multiplying each value by its probability, and combining the results. The values can represent money, health, time, or other priorities. The calculation clarifies assumptions, but it depends on estimates and subjective judgments about what outcomes are worth.
Epistemic feelings
Feelings tied to knowingMental feelings such as certainty, doubt, and recognition that accompany thinking and help indicate how much a person believes they know. In the book's account, good probability judgments depend both on these feelings being well calibrated and on the ability to translate them into numerical estimates.
100 percent rule
Exclusive outcomes total 100%A constraint that probabilities assigned to mutually exclusive possibilities must sum to 100 percent. If one suspect is eliminated from a murder inquiry, the probability assigned to the remaining suspects must increase accordingly. The rule helps keep estimates internally consistent as evidence changes.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.