Key Takeaways
Treat an unknowable future as uncertainty, not a roulette wheel
Some unknowns are different. Resolvable uncertainty can be looked up or described by a stable probability distribution, as with a fair coin. Radical uncertainty is different: the possibilities may be unclear, the rules may change, or the consequences may depend on what people believe and do. In such situations, assigning precise odds can imply knowledge no one has. The authors argue that this is the ordinary condition of economic, political, and personal decisions.
History makes the point. Columbus sailed without knowing that the Americas lay between Europe and Asia. In 2007, a Goldman Sachs executive described market moves that, under his firm’s model, should have been fantastically rare. The repeated extremes were not evidence of several cosmic flukes. They showed that the model was a poor account of the world it claimed to measure.
The distinction is most useful when it changes how a decision is made, not merely how uncertainty is named. A forecast can be precise within a model and still be unreliable outside it. This is a familiar problem in engineering: a simulation earns trust through testing against the physical system it represents. Social systems are harder because people react to forecasts and policies, changing the system itself. Still, refusing false precision should not become an excuse to stop measuring. Where data are meaningful, they can constrain judgment; where they are not, honesty about the gap is itself practical information.
First decide whether you face a puzzle or a mystery
Puzzles have rules. A puzzle has a defined question, a limited set of possibilities, and an answer that can in principle be checked. A mystery lacks one or more of these features. It may be impossible to list all possible outcomes, or even to agree on what the question means. The distinction helps determine whether calculation can settle the issue or whether judgment is unavoidable.
NASA and business show the contrast. The MESSENGER probe travelled 4.9 billion miles before reaching Mercury on a path predicted years earlier, because planetary motion follows well-understood, stable laws. By contrast, computer companies repeatedly failed to foresee how personal computing would develop. Their technical problems were often puzzles; the future of their industry was a mystery. Calling a mystery a puzzle does not make it solvable. It only disguises the uncertainty.
This distinction offers a practical first move: specify what would count as an answer and how it could be checked. If no one can do that, adding calculations may only decorate disagreement. Yet puzzles and mysteries are not permanent categories. The extinction of the dinosaurs, once unresolved, became more puzzle-like as evidence accumulated. The useful stance is therefore neither blind confidence nor permanent agnosticism. It is to ask what is currently knowable, what new evidence could change that, and which parts of the decision remain matters of interpretation.
Use probabilities only when the outcomes and rules are genuinely defined
Precision needs a foundation. Probability works well when the possible outcomes and the process producing them are sufficiently clear. The Monty Hall game is a good example: a contestant chooses among three boxes, and a host who knows where the prize is opens an empty one. Because the host’s action carries information, switching gives the contestant a better chance. The problem is fully specified, so the calculation can be checked by repeating the game.
Real questions often are not. In the two-child puzzle, “one child is a girl” can mean different things depending on how that fact was learned, producing different answers. In the Sally Clark trial, a doctor multiplied the odds of two infant deaths as if the events were independent, despite the possibility of shared causes. The number looked exact; the assumptions were not.
Probability is not the enemy here. The target is probability used beyond its warrant. The Monty Hall example shows how formal reasoning can correct intuition when the rules are explicit; the Clark case shows how a correct calculation under false assumptions can distort a judgment. In practice, the most important statistical question may be whether the model’s categories fit the case at hand. This also explains why uncertainty should not be collapsed into a single confidence score: a number may summarize a person’s confidence without establishing that the underlying event has a measurable frequency.
Diagnose the situation before choosing a solution
Start with the context. The authors’ recurring decision question is not simply “What should be done?” but “What is happening here?” Diagnosis means understanding the situation, the relevant actors, and the constraints before ranking actions. This differs from beginning with a menu of choices and calculating which one scores highest.
Obama’s decision illustrates the difference. Before authorizing the 2011 raid on the Abbottabad compound, President Obama heard sharply different assessments of whether Osama bin Laden was there. Some advisers expressed confidence; others were far less sure. Obama rejected the idea that those probabilities could settle the matter. He listened, considered the risks of the operation and its alternatives, then made the decision himself. The choice could not be handed to a model or an intelligence official. The responsibility belonged to the president.
Diagnosis is a defense against solving the wrong problem with impressive efficiency. It resembles clinical reasoning: symptoms must be interpreted in context before treatment makes sense. It also limits the seductive habit of turning every decision into a spreadsheet exercise. But diagnosis is not a guarantee of wisdom. Leaders can misread a situation, and an insistence on more context can become an excuse for delay. The discipline is to gather the most relevant knowledge, distinguish what is known from what is conjecture, and then accept responsibility for acting before certainty arrives.
Choose workable actions when optimization requires knowledge nobody has
Optimization assumes a complete map. Standard economic models often portray people as listing every possible action, predicting each result, and selecting the best one. In a radically uncertain world, people do not know all the options or consequences, so this ideal is not a realistic description of decision-making. The authors distinguish this axiomatic model of rationality from practical judgment shaped by experience and context.
Herbert Simon called the alternative satisficing. It means searching for an option that is good enough to meet the need, rather than proving it is the best imaginable choice. Gary Klein’s studies of firefighters and other experienced responders found that they often recognize a familiar pattern, consider a workable response, and act, rather than compare a long list of alternatives. Such methods can outperform a formal calculation built on assumptions that do not fit reality.
A good-enough choice is not automatically a good choice. Experience-based judgment can be fast and powerful, but it can also carry habits from a past situation into a new one. The strongest reading of this idea is not to romanticize intuition, but to use it where the environment is complex and time or information is limited, then make it answerable to evidence and feedback. Formal optimization remains valuable in well-defined tasks. The mistake is treating success in those tasks as proof that the same machinery can optimize a life, a company, or a national policy.
Build decisions around coherent stories, then expose them to challenge
People make sense through narratives. A narrative is an account that connects facts and actions into an explanation of what is happening. A useful one should be credible in light of experience and coherent within itself. It need not be a forecast or a complete account of reality; it helps people interpret imperfect information and coordinate action.
Legal cases show both the value and danger. A jury cannot convict simply because a statistic makes one possibility seem rare. It needs an explanation that fits the evidence and holds up against alternatives. In the O. J. Simpson trial, DNA evidence mattered, but it could not establish the whole story by itself. Stories can also mislead: a persuasive account may hide missing evidence or assumptions. The authors’ standard is not “trust a good story,” but use a story to organize the evidence and then test it.
Narrative reasoning makes scattered facts usable, especially when no complete probability model exists. It also exposes what a bare number leaves out: motive, sequence, relationships, and context. The risk is that coherence can feel like truth. People are adept at making events fit a compelling plot, and a smooth explanation may exclude possibilities simply because they are hard to imagine. A disciplined narrative therefore needs friction: competing explanations, contrary evidence, and explicit acknowledgment of what remains unknown. The courtroom’s adversarial structure captures that value, though it cannot guarantee that the most persuasive story is the most accurate one.
Protect the future you need without pretending to predict it
A reference narrative sets the aim. This is the realistic account of how a person, household, or organization expects things to go. Risk is the possibility that this account will fail in an important way. For a family, it may include a secure retirement; for a bank, the ability to meet obligations and continue operating. The task is not to eliminate uncertainty, but to make plans robust across plausible changes.
Diversification is one practical defense. A portfolio’s exposure depends on how its investments relate to one another, not just on the apparent risk of each asset alone. But historical correlations cannot guarantee future protection. The authors contrast robust planning with the elaborate bank risk models that failed before the 2007-08 crisis. They also warn that pension systems seeking near-certainty through long-range assumptions can produce costly rules that weaken the security they were meant to provide.
This approach shifts attention from forecasting the exact future to preserving the capacity to respond. That is a valuable link between finance and resilience planning: options, buffers, and varied exposures can matter more than a precise central estimate. Still, “robustness” needs a clear purpose. A plan resilient to one disruption may be costly or fragile against another, and not every contingency can be protected against. The practical question is which failures would most damage the reference narrative, what early signs would reveal them, and which choices keep room to adapt.
Use models as tools, never as reality
A model should clarify a decision. Small-world models simplify reality to isolate a mechanism, as Ricardo’s example of England and Portugal used cloth and wine to explain comparative advantage. Their value is insight, not a claim to reproduce the whole economy. A model becomes dangerous when its assumptions are hidden, its numbers are invented, or its output is treated as a decision by itself.
More detail can mean less understanding. British transport appraisals assigned precise monetary values to travellers’ time decades into the future, while the Edinburgh tram’s forecasting exercise used simulations built from assumed inputs. A simpler model of HIV spread, by contrast, highlighted the importance of how many different partners an infected person had, a factor obscured in a more elaborate model. Good modelling identifies key drivers, tests assumptions, and explores scenarios rather than manufacturing certainty.
The key distinction is between a model that disciplines thought and one that performs authority. A stripped-down model can reveal which assumptions matter; a complex one can hide them behind technical detail. Yet simplicity is not a virtue by itself. A model that omits a critical mechanism can be as misleading as one that buries it. Its usefulness depends on the question, the evidence, and whether users can inspect how changing an assumption changes the result. Model outputs should inform judgment, not relieve decision-makers of it.
Treat intelligence as collective: invite expertise and dissent
No single person holds the whole picture. Human achievement depends on groups exchanging specialized knowledge, from building aircraft to making public decisions. Collective intelligence is not the average opinion of a crowd; it is the combination of distinct information and expertise, organized so that it can shape action.
The Cuban Missile Crisis shows why challenge matters. After the Bay of Pigs invasion failed, President John F. Kennedy changed how he sought advice during the 1962 crisis. He arranged for advisers supporting different responses to critique one another’s arguments and sometimes stayed away from meetings so they would speak freely. This helped test the competing accounts of a naval quarantine and an airstrike. The authors contrast that process with groupthink, where fear of disagreement lets a weak narrative pass unexamined.
Collective intelligence requires more than assembling smart people. Status, incentives, and fear can silence the knowledge a group most needs. The Kennedy example suggests that leaders can shape the quality of judgment by designing disagreement into the process, rather than treating consensus as proof. But dissent alone is not enough: endless debate can postpone responsibility, while a confident leader can cherry-pick only the objections that suit a preferred conclusion. Effective groups need both open challenge and clear decision rights, so that learning from others does not become a way to evade the final choice.
Analysis
The book’s deepest argument is not that numbers are useless, but that their authority depends on the world they describe. Probability is powerful in settings with defined outcomes and stable generating processes. The authors’ critique begins when that success is generalized to economies, organizations, and political decisions, where categories shift, people react to predictions, and important possibilities may not yet be imaginable. Their alternative is a practical method: diagnose the situation, build an explanatory narrative from distributed knowledge, test it against challenge, and protect important aims against several plausible futures.
This account is especially persuasive when it distinguishes calculable uncertainty from an underspecified problem. The failures of financial risk models, pension projections, and transport appraisals show how precise outputs can conceal fragile assumptions. The stories of Obama, Kennedy, and experienced firefighters make a complementary point: sound action often depends on contextual judgment, not an exhaustive calculation. The book also avoids a simple anti-model stance. Its examples of planetary motion, probability puzzles, comparative advantage, and epidemiology show that formal tools can be illuminating when properly scoped.
The argument is less secure when it risks making narrative judgment look more reliable than quantitative analysis. Stories can organize evidence, but they can also rationalize a leader’s prior beliefs, reward charismatic explanation, and discount base rates that genuinely matter. Statistical models can expose patterns no individual can see, even when the model is imperfect. The practical standard should therefore be plural: use numbers to discipline claims, narrative to interpret context, and challenge to reveal where either method is failing. Radical uncertainty does not excuse vagueness or abandon learning. It changes the aim from pretending to know the optimum to making decisions that remain defensible and adaptable despite incomplete knowledge.
Review Summary
Radical Uncertainty receives mixed reviews, with an average rating of 3.90 out of 5. Many readers appreciate the book's critique of overreliance on mathematical models in economics and decision-making under uncertainty. The authors' expertise and engaging writing style are praised. However, some find the book repetitive and overly long. Critics argue that the main ideas could have been conveyed more concisely. Despite these drawbacks, many readers consider it an important contribution to the field, offering valuable insights into dealing with complex, unpredictable situations.
People Also Read
Glossary
Radical uncertainty
Uncertainty without calculable oddsA condition in which knowledge is too incomplete to define all relevant outcomes or assign meaningful probabilities to them. The processes involved may change over time, depend on human beliefs and actions, or concern situations that have no close precedent. It differs from a known risk such as a fair coin toss.
Puzzles and mysteries
Solvable versus ill-defined problemsA puzzle has clear rules, a limited set of possible answers, and a solution that can be checked. A mystery is vague, incomplete, or unique enough that no objectively correct solution may be available. The distinction helps determine whether calculation can settle a question or whether judgment and interpretation remain necessary.
Viniar problem
Mistaking model odds for realityThe error of treating probabilities calculated inside a model as if they described the real world. A model may imply that an event is extraordinarily unlikely, but that conclusion depends on the model accurately representing the world. If its assumptions fail, an apparent statistical extreme may instead reveal model failure.
Reference narrative
Expected course of eventsA person’s or organization’s account of how things are realistically expected to go. It gives a context-specific meaning to risk: risk is the possibility that important parts of this account will fail. Planning can then focus on protecting the underlying aim and maintaining room to adapt, rather than predicting every possible outcome.
Satisficing
Finding a workable optionA decision approach associated with Herbert Simon in which a person searches for an option that is good enough to meet a need, rather than proving that it is the best possible option. It is useful when the available choices, consequences, or conditions cannot be fully known, making exhaustive optimization impossible or misleading.
Evolutionary rationality
Adaptive judgment in contextA way of understanding human reasoning as adapted to the complex, social environments people face, rather than as imperfect calculation against idealized axioms. Intuition, emotion, pattern recognition, and cooperation can help people act under uncertainty. These capacities are not infallible, but their usefulness depends on context rather than conformity to a universal optimization rule.
Communicative rationality
Reasoning expressed through discussionThe reasoning people use to explain, justify, and challenge judgments in communication with others. It helps groups compare interpretations, share knowledge, and coordinate action. The explanation offered for a decision does not necessarily reproduce the mental process that produced it, but discussion can test and improve the account guiding a collective choice.
Small world
A simplified, defined problemA setting in which the relevant choices, outcomes, and rules can be specified well enough to permit calculation. Games of chance and carefully constructed economic models are examples. A small-world solution can illuminate part of a real problem, but should not be mistaken for a complete description of a complex, changing world.
FAQ
What is Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King about?
- Explores decision-making limits: The book examines how traditional quantitative models and forecasts often fail to capture the complexity and unpredictability of real-world decision-making.
- Focus on radical uncertainty: It introduces the concept of radical uncertainty, where future events cannot be assigned meaningful probabilities or fully anticipated.
- Emphasizes narratives and judgment: Kay and King argue for the importance of narratives, practical knowledge, and collective intelligence in navigating uncertainty in economics, business, and policy.
Why should I read Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Challenges conventional wisdom: The book critiques the overreliance on optimization, models, and forecasts in economics, business, and finance, showing their limitations in real-world situations.
- Offers practical decision-making insights: Readers gain a richer framework for making decisions under uncertainty, emphasizing judgment, communication, and adaptability.
- Broad interdisciplinary relevance: Drawing from history, psychology, law, and economics, the book is valuable for policymakers, business leaders, and anyone interested in how decisions are made.
What are the key takeaways from Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Numbers don’t tell all: Quantification and models can create a false sense of understanding, often missing the true complexity of real-world problems.
- Narratives are essential: Humans use stories to interpret unique events, communicate decisions, and guide action when probabilities are unavailable.
- Models have limits: Small-world models are useful but incomplete; many real-world events cannot be assigned meaningful probabilities or predicted reliably.
- Collective intelligence matters: Effective decision-making relies on communication, challenge, and cooperation among diverse perspectives, not just isolated optimization.
What is "radical uncertainty" as defined in Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Unknowable future: Radical uncertainty refers to situations where the future is not just unknown but fundamentally unknowable, making it impossible to list all possible outcomes or assign meaningful probabilities.
- Contrast with risk: Unlike risk, which involves measurable variability and known probabilities, radical uncertainty deals with ambiguity, vagueness, and ill-defined problems.
- Implications for models: Models that assume known probabilities or optimization are often misleading under radical uncertainty, as they pretend to knowledge that does not exist.
How do John Kay and Mervyn King distinguish between "risk" and "uncertainty" in Radical Uncertainty: Decision-Making Beyond the Numbers?
- Risk as measurable variability: Risk involves situations where outcomes and their probabilities are known or can be estimated, allowing for quantification and pricing.
- Uncertainty as unknowable: Radical uncertainty involves unknown or unknowable outcomes and probabilities, making it impossible to assign reliable probabilities or optimize decisions.
- Consequences of confusion: Treating radical uncertainty as risk leads to overconfidence in models and can create systemic vulnerabilities, as seen in financial crises.
How do John Kay and Mervyn King define and use "narratives" in decision-making in Radical Uncertainty: Decision-Making Beyond the Numbers?
- Narratives as sense-making tools: Narratives help humans interpret complex, uncertain situations where probabilities are unavailable, providing coherence and credibility.
- Communication and social context: Narratives are essential for persuading others, eliciting cooperation, and making collective decisions under uncertainty.
- Evolving and challengeable: Good narratives are open to challenge and evolve over time, much like scientific paradigms, ensuring adaptability and resilience.
What is the difference between "puzzles" and "mysteries" in decision-making according to Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Puzzles are well-defined: Puzzles have clear rules and a single correct solution, solvable by probabilistic or logical reasoning (e.g., games of chance, engineering problems).
- Mysteries are ill-defined: Mysteries lack clear definitions, have no objectively correct answers, and involve vagueness and indeterminacy (e.g., geopolitical events, business strategy).
- Different approaches required: While puzzles can be addressed with optimization and probability, mysteries require narrative framing, judgment, and acceptance of ambiguity.
How do John Kay and Mervyn King critique the use of probabilities and optimization in economics and decision theory in Radical Uncertainty: Decision-Making Beyond the Numbers?
- Inapplicability in large worlds: Probabilistic models and optimization assume well-defined options and known probabilities, which rarely exist in real economic or political contexts.
- Model-based probabilities vs. reality: The authors highlight the danger of confusing model-based probabilities with real-world probabilities, as seen in the 2007–08 financial crisis.
- Limits of axiomatic rationality: Axiomatic definitions of rationality, which underpin expected utility maximization, fail to capture how real people make decisions under radical uncertainty.
What is the difference between "axiomatic rationality" and "evolutionary rationality" in Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Axiomatic rationality: Defined by adherence to logical axioms and maximization of expected utility, suitable for small, well-defined problems.
- Evolutionary rationality: Describes how humans have evolved cognitive and social abilities to cope with complex, uncertain worlds through heuristics, narratives, and cooperation.
- Practical decision-making: Evolutionary rationality better explains real-world behavior, including the use of rules of thumb, satisficing, and adaptive judgment rather than strict optimization.
How does Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King view the role of human intelligence versus artificial intelligence in decision-making under uncertainty?
- Different strengths: Humans excel at dealing with ill-defined, unique, and radically uncertain problems through narrative and judgment, while AI excels at well-defined, repeatable puzzles.
- Limits of AI: Even advanced AI systems rely on defined rules and repeated play, and cannot handle unique, complex political or economic decisions.
- Complementarity: AI should augment human decision-making, not replace it, as good judgment and context understanding remain uniquely human capabilities.
What practical advice or methods does Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King offer for living and making decisions in a radically uncertain world?
- Embrace uncertainty: Accept that perfect knowledge is impossible and that uncertainty is pervasive and unavoidable.
- Use narratives and communication: Build and revise stories collaboratively to make sense of complex situations and guide decisions.
- Focus on robustness and adaptability: Develop strategies and policies that are resilient to multiple possible futures rather than optimizing for a single predicted outcome.
What are some real-world examples of model misuse discussed in Radical Uncertainty: Decision-Making Beyond the Numbers by John Kay and Mervyn King?
- Financial risk models: Value at risk models underestimated risks before the 2007–08 crisis due to reliance on historical data and assumptions of stationarity.
- Transport project appraisals: The UK’s WebTAG model used invented numbers and ignored broader impacts, leading to misleading conclusions about infrastructure investments.
- Fisheries and health models: Complex models in Canadian fisheries and WHO HIV projections failed due to oversimplification or neglect of key behavioral factors, demonstrating the dangers of overreliance on quantitative models.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.