Two Minds That Changed How We Think About Thinking
The partnership between Daniel Kahneman and Amos Tversky is the most consequential intellectual collaboration in the history of psychology, possibly in the history of social science. In roughly fifteen years of sustained joint work between 1969 and 1983, they produced a body of research that overturned the dominant model of human rationality in economics, launched the field of behavioral economics, won the Nobel Prize in Economic Sciences (awarded to Kahneman in 2002; Tversky died in 1996, and the prize isn’t awarded posthumously), and generated a framework for understanding human judgment that now permeates contemporary medicine, law, public policy, finance, and military strategy.
Michael Lewis’s The Undoing Project tells the story of how this collaboration happened, and why it produced what it produced. Lewis approaches it as a journalist who covered the early application of Kahneman and Tversky’s ideas in sports analytics (in his earlier book Moneyball) and got curious about where the ideas had come from. What he found was a story that’s simultaneously intellectual history, psychological portrait, and love story — the last being the most surprising element. Kahneman and Tversky’s relationship was one of the most intense and productive intellectual partnerships in scientific history, and it ended in a painful estrangement whose causes neither man could clearly articulate and neither fully recovered from.
The book operates on two levels simultaneously, inseparably. At the intellectual level, it explains the substance of the heuristics and biases research program — the specific ways human judgment departs from the predictions of rational actor models — in language accessible to readers with no background in psychology or economics. At the human level, it tells the story of two extraordinary individuals whose characters, histories, and relationship dynamics shaped the research in ways no purely intellectual account of the findings could capture. Understanding how Kahneman and Tversky worked together, what each brought to the collaboration, and how their complementary minds produced findings neither could have reached alone, turns out to be essential to understanding why the findings look the way they do.
Amos Tversky: The Mind That Couldn’t Stop
Amos Tversky appears in Lewis’s account as perhaps the most intellectually gifted person the various witnesses to his life had ever encountered — a man of such fluency in abstract reasoning, such comfort with mathematical formalism, and such social magnetism that his colleagues genuinely struggled to decide whether his intelligence was more impressive than his personality or the other way around. Born in Israel in 1937, Tversky served in the Israeli military and was repeatedly decorated for physical courage before starting the academic career that would make him one of the most influential psychologists of the twentieth century.
The courage he showed physically he showed just as much intellectually. No interest in defending or elaborating existing theories — his mode was to identify the most confident assumption in a field and find the experiment that would reveal whether it was wrong. Contemptuous of complexity for its own sake, with a gift for the simple, clean experimental design that could answer a fundamental question with minimal apparatus. His collaborators consistently describe his pace of thinking as alarming — conversations with Tversky were an exercise in trying to keep up with a mind that had reached the conclusion before most people had finished understanding the question.
His social world was as intense as his intellectual one. He was the center of every room he entered — not through effort or affectation but through the quality of his attention, absolute when engaged and absolutely absent when not. People who had his attention felt like the most interesting person in the world. People who didn’t felt his indifference acutely. That pattern created intense loyalty in his close collaborators and a difficult environment for anyone at the margin of his attention. Kahneman, who was neither indifferent to other people’s regard nor indifferent to Tversky’s, was his perfect complement.
Daniel Kahneman: The Mind That Wouldn’t Rest
If Tversky’s defining characteristic was confidence, Kahneman’s was doubt. Lewis portrays him as constitutionally self-critical — someone for whom certainty was a sign of insufficient reflection rather than mastery. Born in Tel Aviv in 1934, Kahneman spent part of his childhood hiding from the Nazis in occupied France, an experience he’s reflected on publicly in ways that reveal how deeply the arbitrary terror of racial persecution, the capricious decisions of people in power, shaped his lifelong interest in the mechanics of human judgment.
Kahneman came to psychology with a question: why do people believe what they believe? That question had practical urgency early in his career, when as a young IDF officer he was tasked with evaluating candidates for combat leadership — using an interview procedure whose predictive validity, when he measured it, turned out to be essentially zero. Discovering that confident expert predictions based on structured observation could be no more accurate than random chance was a formative empirical encounter that shaped his research program for the next five decades.
Where Tversky moved through problems on the force of intellectual confidence, Kahneman moved through them by systematically interrogating his own intuitions. Exquisitely sensitive to the feeling of certainty as a potential sign of error — the comfortable sense that a judgment was correct was, for Kahneman, as likely to mean the judgment had bypassed critical examination as that it had passed it. That sensitivity made him an ideal partner for Tversky’s confident breakthroughs: Kahneman supplied the persistent “but what about…” questioning that refined the confident insight into an empirically testable, strong finding.
The Heuristics and Biases Program
The research Kahneman and Tversky produced together identified the specific mental shortcuts — heuristics — humans use to make judgments under uncertainty, and the systematic errors — biases — these heuristics produce in specific circumstances. Not an attack on human reasoning as fundamentally flawed. Heuristics are intelligent adaptations to the computational constraints of the human mind: they produce good-enough answers most of the time with far less effort than exhaustive analysis would require. The problem is that they produce systematically wrong answers in specific, common circumstances — and the confidence we feel in our heuristic judgments doesn’t track their accuracy.
The representativeness heuristic is judging the probability of a hypothesis by how closely it resembles a prototype. Asked whether someone described as meticulous, detail-oriented, and introverted is more likely a librarian or a farmer, people say librarian — even though farmers vastly outnumber librarians, so the base rate strongly favors farmer. The description matches the librarian prototype better than the farmer prototype, and representativeness weighs that match against the base rate in ways that reliably produce wrong probability estimates.
The availability heuristic is estimating an event’s frequency by how easily examples come to mind. Events that get heavy media coverage — plane crashes, terrorist attacks — feel more available than statistically more common events — car accidents, heart disease — and get judged more probable relative to those less-available events than their actual frequencies warrant. Availability is a decent proxy for frequency in most real-world environments, where more frequent events tend to be more commonly encountered and therefore more available. It fails when availability and frequency come apart — when media coverage, emotional salience, or recent personal experience makes rare events disproportionately vivid.
The anchoring heuristic is estimating values by starting from an initial reference point and adjusting — insufficiently, leaving final estimates clustered around the anchor even when the anchor has zero rational relationship to the true value. Kahneman and Tversky demonstrated it with a spinning wheel that was obviously random: participants were asked whether the percentage of African nations in the United Nations was higher or lower than the wheel’s number, then asked for their actual estimate. Participants who saw a high number from the wheel gave higher estimates than those who saw a low number, despite everyone knowing the wheel was random. The arbitrary anchor contaminated the estimate in ways participants couldn’t fully correct for even knowing the anchoring effect was operating.
Prospect Theory: How We Actually Experience Gains and Losses
- First, people evaluate outcomes relative to a reference point — typically the status quo or an anticipated level — rather than in absolute terms. A gain isn’t experienced as “having more” in any objective sense but as “more than the reference point.” A loss isn’t “having less” objectively but “less than the reference point.” This reference dependence means the same objective outcome can register as a gain or a loss depending entirely on the reference point framing it — with enormous implications for how prices, negotiations, and changes should be designed and communicated.
- Second, the value function is asymmetric around the reference point in the direction of loss aversion. Losses get felt about twice as intensely as equivalent-magnitude gains. The pain of losing $100 exceeds the pleasure of gaining $100, which means any gamble with equal odds of gaining or losing the same amount feels bad in expectation even though it’s financially neutral. This explains why people accept lower expected returns in exchange for reduced variance, why they hold losing investments longer than rational analysis warrants, and why they work harder to prevent losses than to achieve equivalent gains.
- Third, the probability weighting function is nonlinear in ways that produce the specific risk preference patterns Kahneman and Tversky documented empirically. People overweight small probabilities — treating a 1% chance of a large outcome as more significant than its expected value implies — and underweight moderate to high probabilities. This distortion explains why people simultaneously buy lottery tickets (small probability of large gain, overweighted) and buy insurance (small probability of large loss, overweighted), which looks contradictory under expected utility theory but is entirely consistent with prospect theory’s nonlinear probability weighting.
The intellectual pinnacle of the Kahneman-Tversky collaboration is prospect theory, the descriptive model of decision under uncertainty they developed as an alternative to expected utility theory — the standard economic model of how rational agents make choices involving risk. Prospect theory is the technical foundation of behavioral economics, and it earned the Nobel Prize primarily because it provided, for the first time, a mathematically precise account of how real humans differ from Econs in evaluating risky choices.
Three key features of prospect theory describe three systematic departures from expected utility theory.
The Friendship as Research Method
Lewis’s most original contribution is his account of how the collaboration actually worked — and his argument that the nature of their friendship wasn’t incidental to the quality of the research but constitutive of it. They wrote everything together. Not the division-of-labor collaboration academics typically mean by joint work, where one writes a section and the other edits — genuine co-writing, where neither produced text without the other present to react, contest, and refine. Every idea that made it into the final product had survived the real-time interrogation of a mind of equivalent power with no stake in its survival.
The collaboration worked because of the complementarity Lewis documents throughout. Tversky’s confidence and Kahneman’s doubt weren’t merely personality differences — they were different epistemic stances producing different intellectual contributions. Tversky generated possibilities at extraordinary speed and selected among them with confident intuition. Kahneman tested the selected possibility against every alternative framing, every potential objection, every way the claim might be wrong. Together they produced ideas that were both creative and rigorous — generated with Tversky’s fertility, tested with Kahneman’s tenacity.
It also worked because of what each brought from different intellectual backgrounds. Tversky was a mathematical psychologist who thought naturally in formal models — his instinct was to capture a psychological phenomenon in a precise mathematical description that generated testable predictions. Kahneman was a perceptual psychologist trained to carefully design experiments revealing the structure of subjective experience. Together they could formalize an idea into a testable model and design the clean experimental test for whether the formalization was correct. Neither could have done this as well alone.
The Painful End and Its Lessons
The dissolution of the partnership is the most painful part of Lewis’s account, and the most humanly instructive. No specific event or betrayal caused it — no falling out, no discovery of duplicity, no competing interests forcing a choice. It came from the gradual accumulation of asymmetric recognition their work received, and from what that asymmetry did to a friendship whose equality had been essential to its function.
As the heuristics and biases program became famous in the early 1980s, the academic world increasingly credited the work to Tversky — the more charismatic, more mathematically imposing, more conventionally brilliant of the two. Kahneman found himself introduced as Tversky’s collaborator rather than an equal partner. Invitations that should have gone to both went to Tversky alone. Awards the work deserved went to Tversky on behalf of the collaboration, rendering Kahneman invisible. Tversky, for his part, accepted the recognition without adequately insisting on Kahneman’s equal contribution — not from malice, his friends insisted, but from a combination of vanity and social insensitivity that was the dark side of his extraordinary confidence.
Kahneman experienced this asymmetric recognition as a loss he couldn’t speak about directly, because speaking about it felt ignoble. Not resentment of Tversky’s brilliance — Kahneman genuinely believed Tversky was more brilliant and said so publicly. Not resentment of the work itself — he was proud of what they’d made together. This was a profound fairness violation in a domain — intellectual credit — where fairness norms run strong but informal, and where the injured party has no legitimate remedy except to keep producing work that eventually establishes the contribution independently. The estrangement grew out of a fairness wound that festered because it couldn’t be spoken, not from any rational calculation that the friendship wasn’t worth preserving.
The Legacy: What the Research Changed
Kahneman and Tversky’s research program changed how consequential decisions get made across multiple domains, and the specific changes illuminate what the findings actually mean for practice. In medicine, the heuristics and biases framework provided the first rigorous account of why experienced physicians make systematic diagnostic errors, and a framework for designing decision support tools that reduce those errors by counteracting specific heuristics. The medical judgment research Kahneman’s collaboration with Paul Meehl and others produced showed that actuarial prediction — statistical models based on measurable patient characteristics — consistently outperforms clinical judgment wherever sufficient outcome data exists, a finding that’s driven the development of clinical decision support tools across medicine.
In finance and investment, the prospect theory findings provided the theoretical foundation for a comprehensive re-examination of investor behavior and market pricing. The behavioral finance research program that Thaler, Shleifer, DeBondt, and others built on the Kahneman-Tversky foundation has produced decades of evidence that asset prices reflect not just fundamental information but the systematic biases and heuristics individual investors apply. That evidence has changed how sophisticated institutional investors think about markets, how financial regulation gets designed, and how financial products get built and disclosed to reduce the exploitation of behavioral tendencies.
In public policy, the insights about defaults, framing, and reference dependence have informed a new generation of policy designs — the nudge architecture — that achieve behavior change by modifying choice environments rather than relying on information and incentives alone. The applications span retirement savings, healthcare enrollment, energy conservation, tax compliance, and organ donation — domains where the gap between what people say they want and what their default behavior produces is large, and where modest architectural changes produce measurable improvements.
Thinking Fast and Slow: The Popularization
Kahneman’s Thinking, Fast and Slow (2011), published a decade after his Nobel Prize, is the most comprehensive popular account of the heuristics and biases research program — and it sold millions of copies, reaching an audience far beyond any academic psychology book. Lewis’s account treats it as the culmination of Kahneman’s effort to synthesize the work in a way that gives full credit to Tversky while making the findings accessible to a broad public. The book introduced the System 1 / System 2 framework — fast intuitive processing versus slow deliberate processing — as an organizing metaphor for the findings, and that framework has since entered the general vocabulary of educated discourse about decision-making.
The System 1 / System 2 framework isn’t itself a research finding from the heuristics and biases program — it’s an interpretive structure Kahneman uses to explain why heuristics operate as they do. System 1, the fast intuitive system, produces heuristic judgments automatically and with high confidence; System 2, the slow deliberate system, can check and correct System 1’s outputs but is cognitively costly and often doesn’t get engaged. Most of the biases Kahneman and Tversky documented are System 1 products that System 2 fails to catch — not because System 2 is incapable of catching them in principle, but because it’s lazy, disengaged, or simply doesn’t know to be suspicious of the confident output it’s being asked to endorse.
The framework helps practitioners because it gives a vocabulary for the experience of decision-making and a heuristic for when to distrust your intuitions: when the conditions producing System 1 errors are present — time pressure, novel domain, emotional arousal, complexity exceeding intuitive pattern-matching capacity — engage System 2 deliberately even when System 1 is producing a confident answer. The practical challenge is that System 1 is designed to keep System 2 from getting engaged when it isn’t needed, which also tends to prevent engagement when it is needed. Building habits of deliberate cognitive check — the pre-mortem, the reference class forecast, structured decomposition of complex estimates — is the practical work of training System 2 to engage exactly when it matters most.
Implications for Judgment and Resilience
The availability heuristic creates specific resilience vulnerabilities. Under stress, the most available mental models — the most recent analogous experience, the most emotionally vivid precedent — dominate judgment in ways that may not track the actual probability distribution of the current situation. The leader who lived through a specific type of organizational crisis ten years ago will see the current crisis through that earlier lens even when the structural differences matter. Building resilient judgment means building practices that deliberately expand the range of analogies considered rather than defaulting to the most available one.
Loss aversion creates a specific pattern of risk preference under adversity: people take larger risks to avoid losses than to achieve equivalent gains, which means adversity itself changes the risk calculus in ways that can produce over-bold action when caution is warranted. The leader or individual in a loss position — behind on targets, facing a declining situation, at risk of significant penalty — will see the available options through a systematic skew toward accepting risks they’d refuse in a gain-domain context. Knowing this pattern allows for deliberate correction: assessing options from a loss position, ask explicitly whether the risk preference would hold if the same option were framed as a gain instead of a loss.
The clinical takeaway
The Undoing Project is the best available narrative account of how two extraordinary minds produced the most important findings in behavioral science. Lewis writes with his characteristic ability to make complex ideas vivid through story — never presented abstractly, always through the specific experiments, arguments, and human moments in which they got developed. The result teaches behavioral economics without feeling like a textbook and illuminates a friendship without feeling like a biography.
The research Kahneman and Tversky produced together has changed how human judgment gets understood at the deepest level. Their fundamental finding — that systematic errors in human judgment come not from stupidity or irrationality but from the heuristics intelligent minds use to work through a world of limited information and bounded time — is both humbling and liberating. Humbling because it implies confidence isn’t a reliable indicator of accuracy, that intuition can be confidently wrong, that expertise in a domain confers no immunity to the biases that domain is subject to. Liberating because the biases are predictable, their conditions of occurrence identifiable, and the practices that reduce their influence known and learnable.
The story of Kahneman and Tversky is also a story about intellectual courage — the courage to follow an observation wherever it leads, to challenge the dominant framework of your field when the evidence demands it, and to maintain the quality of your work despite the professional risks heterodoxy always carries. Both men took those risks throughout their careers, and the world’s understanding of human judgment is immeasurably richer for it. The friendship that made the work possible, and the painful dissolution both men mourned without either fully being able to describe it, is the most human part of a story about what it means to think as carefully as it’s possible to think about how poorly humans think.
Framing Effects: How Description Shapes Decision
One of the most consistently demonstrated findings in the Kahneman-Tversky research program — and one of the most directly threatening to standard economic theory — is the framing effect: people respond differently to identical information depending on how it’s described, violating the rational actor assumption that preferences should be independent of irrelevant changes in description. The classic demonstration is the Asian disease problem: participants are told an unusual disease is expected to kill six hundred people, and asked to choose between two programs. Framed in terms of lives saved (Program A saves two hundred people; Program B has a one-third chance of saving all six hundred and a two-thirds chance of saving no one), most choose the certain option. Framed in terms of deaths (Program C: four hundred people will die; Program D: one-third chance of zero deaths, two-thirds chance of six hundred deaths), most choose the risky option. Mathematically identical outcomes across frames.
Preferences reverse completely depending on whether the framing emphasizes gains (lives saved) or losses (deaths).
This finding shouldn’t be possible in a world populated by rational agents. Rational agents evaluate outcomes by their actual consequences, not by how those consequences are described. Preferences reversing on nothing but the description of the same outcomes directly demonstrates people aren’t rational agents in the relevant sense, and that the preferences they express depend on the framing rather than on any stable underlying value system the framing merely reveals. Profound implications for every domain where decisions get made on presented information: policy design, medical informed consent, financial advising, negotiation, marketing. Whoever controls the frame controls a significant portion of the decision.
Lewis conveys the significance of this finding through narrative rather than formal presentation, showing how Kahneman and Tversky’s conversations about the framing paper were themselves exercises in the kind of productive disagreement that characterized their collaboration at its best. Their arguments about what the framing result meant — a failure of rationality, a feature of evolved psychology, or something else entirely — get recounted with enough specificity to convey both the intellectual content and the quality of the relationship that generated it. This is what the book does better than any systematic presentation of the findings: it makes the intellectual process visible in a way that makes the ideas more memorable and the researchers more human.
Sports Analytics and the Bias Discovery
Lewis opens the book with Daryl Morey, the Houston Rockets’ general manager who built a career using statistical analysis to challenge the cognitive biases of NBA player evaluation — and who later discovered that the specific biases he was fighting (over-weighting recent performance, prototype-matching “looks like an NBA player,” availability-driven assessment of draft prospects) had been named and described by Kahneman and Tversky twenty years before the Moneyball revolution applied similar logic to baseball. This framing device is one of Lewis’s most effective narrative choices: it grounds the abstract psychology in a concrete, competitive domain where the costs of cognitive bias are measurable in wins and losses, and it provides a through-line from theoretical research to real-world application that makes the science feel immediately relevant rather than academically distant.
The sports analytics connection is more than a narrative device, though. It illustrates a pattern running through every Kahneman-Tversky application: the people exploiting the biases the research described — for profit, for competitive advantage, for better policy outcomes — often didn’t know the research existed. They were discovering the same patterns empirically in their specific domains that Kahneman and Tversky had documented systematically in the laboratory. The Moneyball scouts systematically undervaluing on-base percentage and overvaluing physical appearance were exhibiting representativeness and availability heuristics. The medical professors resistant to statistical diagnosis algorithms were exhibiting the overconfidence and availability biases the clinical judgment research documented. The investors consistently more willing to buy recent winners and sell recent losers than the evidence warranted were exhibiting representativeness and the hot hand fallacy. The same cognitive mechanisms producing the same systematic errors, across domains with no direct connection to each other or to the psychological research that identified the mechanisms.
What Kahneman Said Alone: The Continuation After Tversky
Lewis does not shy away from the uncomfortable question of what the collaboration’s fracture meant for the quality of both men’s work afterward. Tversky’s output, after the collaboration’s intensity diminished, shifted toward more technical probability theory and less of the behavioral observation that had driven the heuristics and biases program. Kahneman’s output after Tversky’s death included Thinking, Fast and Slow — arguably his most comprehensive and most influential single work, and in some sense the monument to the collaboration rather than an independent continuation of it.
What each man could do alone versus what they could do together is one of the most intellectually interesting aspects of Lewis’s account, and he doesn’t answer it definitively — too honest about the counterfactual problem to pretend he knows what either would have produced in the other’s absence. What the account does suggest is that the collaboration wasn’t simply the sum of two complementary skill sets. Something more: a relationship that generated a shared sensibility and a quality of productive disagreement neither could fully replicate with anyone else. The lesson isn’t that great work requires great partnerships — plenty of great solo intellectual achievement exists. It’s that when great partnerships do occur, they produce something qualitatively different from what either partner can produce alone, and that difference is often more valuable than either partner’s solo contribution.
Additional Strengths: What Lewis Gets Exceptionally Right
Lewis’s treatment of the biographical context for the research is one of the book’s most important contributions. Understanding that Tversky’s confidence was shaped partly by his experience as an Israeli paratrooper decorated for bravery under fire — that his decisiveness grew out of a culture that demanded it under conditions where hesitation cost lives — makes his intellectual style more comprehensible and less mysterious than it looks in pure scientific biography. Understanding that Kahneman’s anxiety was shaped partly by his childhood as a Jew in occupied Paris, where misreading a stranger’s intentions could mean life or death, makes his relentless second-guessing more than a personality quirk. Lewis is good at this kind of biographical grounding, and uses it without overextending into pseudo-psychological determinism.
The account of Tversky’s death from cancer — the way he chose to keep it private, to work through it, to protect the people around him from the burden of their own anticipatory grief — is one of the most affecting passages Lewis has written, and it’s effective precisely because of the restraint with which it’s presented. Tversky, characteristically, approached his own death with the same decisive clarity he approached everything else, making peace with what was coming without the prolonged ambivalence his opposite, Kahneman, might have found necessary. Lewis doesn’t editorialize about this. He presents it as the final expression of Tversky’s character, and the presentation is sufficient.
The Simulation Heuristic and Counterfactual Thinking
Among the less-celebrated findings of the Kahneman-Tversky program is the simulation heuristic — judging the probability of an event by how easily it can be mentally simulated or imagined. Construct a mental scenario readily and the event feels probable; if the simulation requires effort or produces implausible intermediate steps, the event feels unlikely. A close relative of the availability heuristic, but operating through ease of construction rather than ease of retrieval.
The simulation heuristic interacts powerfully with counterfactual thinking — the mental construction of alternative versions of past events. The person who narrowly misses their flight feels greater regret than the person who misses it by a wide margin, because the near-miss activates an easily-constructed simulation of the world where the outcome was different, and that simulation feels particularly real. The investor who nearly sold an investment before it collapsed feels worse than the investor who never considered selling, because the counterfactual world where they acted on their near-decision is easily simulated. The regret isn’t proportional to the actual magnitude of the loss. It’s proportional to how easily the alternative world can be simulated.
Lewis depicts Kahneman’s particular fascination with this finding in the context of grief and bereavement. The bereaved parent who lost a child in an accident thinks obsessively about the small changes — the different route, the earlier departure, the decision not to go out at all — that would have prevented the tragedy, because those counterfactuals are easily simulated and therefore feel particularly real. The grief gets amplified by the mental simulation of the world that almost was, a world the simulation heuristic makes feel more accessible and more real than any consideration of actual probability would warrant. The psychological function of the simulation heuristic in grief is to create a sense of agency — the feeling that the outcome was avoidable — which is simultaneously a source of intense guilt and a defense against the existential terror of pure randomness.
For organizational decision-making, the simulation heuristic and its counterfactual twin explain why near-miss experiences often carry larger impacts on subsequent behavior than actual losses of similar magnitude. The near-miss activates a vivid simulation of the catastrophic outcome that almost occurred, creating precautionary responses the abstract knowledge of comparable risks wouldn’t generate. Near-miss reporting systems in aviation and nuclear power exploit this property deliberately: by making near-miss events visible, they create the simulation experience that activates precautionary responses without requiring the actual disaster that would produce equivalent behavioral impact through direct experience.
Regression to the Mean and the Causal Illusion
One chapter in The Undoing Project focuses on Kahneman’s discovery of regression to the mean in a military training context — illustrating how behavioral economics insights emerge at the intersection of statistical reasoning and human psychology. Kahneman was teaching statistics to Israeli air force flight instructors and claimed praise is more effective than punishment for improving flight performance. The instructors pushed back: in their experience, trainees praised for exceptional performances typically performed worse next time, while trainees criticized for poor performances typically improved. Looked, to them, like proof that criticism worked and praise didn’t.
Kahneman realized the instructors were observing the purely statistical phenomenon of regression to the mean and reading it as a causal story. Extreme performances — very good or very bad — are partly skill, partly random variation. On the next attempt, the random component varies independently, so extreme performances on average get followed by less extreme ones. The very good performance is followed, on average, by something somewhat less good; the very bad performance by something somewhat less bad. This regression happens whether the instructor praises or criticizes — it isn’t caused by the response to the performance. It’s a statistical property of repeated measurements with a random component.
The instructors had consistently observed praise followed by decline and criticism followed by improvement — both exactly what regression to the mean predicts — and built a causal story where their responses caused the changes. This causal illusion is ubiquitous wherever people observe sequences of events containing regression to the mean. Managers who praise exceptional performance and criticize poor performance will observe both responses appearing counterproductive, for purely statistical reasons. Coaches managing elite athletes will observe that interventions after extraordinary performances appear to produce declines regardless of what those interventions are. Medical practitioners who intervene when patients are at their worst will observe patients improving after treatment even when the treatment has no efficacy, because the worst-case presentation on average gets followed by a less extreme one regardless of treatment.
The regression to the mean insight is one of the most practically important contributions of the statistical reasoning program Kahneman and Tversky advanced. Failing to recognize it produces not just incorrect causal attributions but incorrect interventions: organizations that punish exceptional performance declines and reward exceptional performance improvements are responding to statistical noise rather than the actual performance trajectory their interventions should target. Understanding regression to the mean is a prerequisite for valid causal inference from sequential performance data — which describes essentially all organizational performance measurement.
The Linda Problem and Conjunction Fallacy
Among Kahneman and Tversky’s most famous and most debated findings is the conjunction fallacy, demonstrated through what became known as the Linda problem. Participants get a description of Linda — a woman who, as a college student, was deeply concerned with social justice, participated in anti-nuclear demonstrations, and majored in philosophy — and rank the probability of several statements about her current situation, including “Linda is a bank teller” and “Linda is a bank teller who is active in the feminist movement.”
The vast majority rank the conjunction — bank teller AND feminist — as more probable than the simple statement. A logical impossibility: the probability of two events occurring together can’t exceed the probability of either alone. The probability that Linda is a bank teller and a feminist can’t exceed the probability she’s a bank teller, because every bank-teller-and-feminist is a bank teller, but not every bank teller is a feminist. Yet the representativeness of the description — the match between it and the prototype of an active feminist — makes the conjunction feel more probable than the bare statement.
The Linda problem generated substantial academic controversy, with critics arguing participants were interpreting the task differently than intended — that “Linda is a bank teller” got heard as “Linda is a bank teller and nothing else,” making the conjunction interpretation pragmatic rather than logical error. Kahneman and Tversky responded with variations controlling for these alternative interpretations, and still got the conjunction fallacy from substantial majorities, including statistically sophisticated participants who should have caught the logical constraint.
The practical importance of the conjunction fallacy lies in what it implies about how complexity affects perceived probability. Detailed, coherent scenarios feel more probable than sparse ones, even though additional detail can only reduce the probability of the scenario as a whole. Persuasive narratives — detailed and internally coherent — get judged more probable than sparse factual statements about the same situation, which is one reason compelling stories move beliefs more than equivalent statistical evidence. The mechanism that makes narratives compelling is the same one that produces the conjunction fallacy: the representativeness of a narrative’s details activates the simulation heuristic in ways that override the logical probability constraints careful statistical reasoning would impose.
The Undoing Project and Its Relevance to How We Make Decisions Today
The heuristics and biases research program Kahneman and Tversky launched has been extended, challenged, refined, and occasionally replication-tested over fifty years of subsequent research. Not every specific finding has survived full scrutiny unchanged — the replication crisis in social psychology has affected some areas of behavioral research, and the boundary conditions and effect sizes of specific findings have been revised. But the core architecture of the program — identifying representativeness, availability, anchoring, loss aversion, and reference dependence as systematic features of human judgment — has held up extraordinarily well across decades, methods, and cultures.
That resilience reflects the fact that Kahneman and Tversky weren’t documenting laboratory curiosities. They were documenting fundamental features of a cognitive architecture evolution built for a very different environment than the one we currently inhabit. The heuristics they identified are intelligent responses to the computational limitations of the human mind: good-enough answers most of the time, with limited cognitive investment, in the environments they were calibrated for. The biases they document are the specific circumstances where that calibration fails — where a heuristic that works well in the ancestral environment produces systematic errors in the modern one.
This evolutionary framing doesn’t excuse the errors — real errors, real consequences — but it does explain why awareness of them doesn’t eliminate them. You can’t correct for the anchoring heuristic by simply deciding not to be anchored, any more than you can correct for the Müller-Lyer illusion by deciding to see the lines as equal. The heuristic operates below conscious deliberation. So the practical work of improving decision quality isn’t primarily correcting the heuristic in the moment. It’s designing decision processes, organizational systems, and personal habits that reduce the circumstances where the heuristic produces errors — or that build correction mechanisms that engage after the heuristic has done its work, rather than trying to prevent it from operating at all.
What Lewis Adds: The Human Story Behind the Science
Michael Lewis’s distinctive contribution in The Undoing Project isn’t the explanation of the science — Kahneman himself does that better in Thinking, Fast and Slow — it’s the human story the science grew from. Lewis spent time with Kahneman and with people who knew Tversky, and reconstructed the texture of the collaboration from multiple perspectives. What emerges is a portrait of how great scientific work actually happens — not through solitary genius, not through institutional support, but through the chemistry of a specific human relationship that let each person become more than they could be alone.
The relationship worked because of what each person provided that the other needed. Tversky needed a collaborator who wouldn’t be intimidated by his intellectual confidence, who’d push back when the confident idea was wrong, who’d pursue the methodological rigor his speed sometimes sacrificed in the interest of getting to the next question. Kahneman needed a collaborator who could generate ideas faster than he could subject them to doubt, who could maintain momentum when Kahneman’s self-criticism threatened to stall the work, and whose social fluency could open the academic doors Kahneman’s introverted style made hard for him to open alone.
The estrangement that ended the collaboration is Lewis’s cautionary tale about what happens when the inequality latent in any asymmetric relationship becomes visible. Tversky and Kahneman weren’t equally gifted — Kahneman said so himself, publicly. But they were equally indispensable to what they produced together, and the academic world’s habit of crediting joint work to the more visible collaborator violated a fairness norm Kahneman couldn’t articulate without seeming to claim an equality he’d explicitly denied. The resulting wound was something neither man knew how to address, and they let the friendship cool rather than force a direct confrontation that would have required both to say things they weren’t sure how to say.
This human detail matters beyond the gossip value that all stories of intellectual fallout carry. It’s a case study in how the cognitive and emotional architecture Kahneman and Tversky spent their careers documenting played out in their own relationship. The availability heuristic made Tversky’s brilliance vivid and Kahneman’s less so. The representativeness heuristic made Tversky the better fit for the prototype of the intellectual star. Loss aversion made acknowledging the asymmetry unbearable for Kahneman, and the endowment effect made the friendship’s value invisible to Tversky until it was gone. The people who understood human judgment most deeply were as subject to its systematic failures as anyone else — which is, finally, the deepest lesson their work has to offer.
Related: Thanks for the Feedback Summary
References
Editorial StandardsCorrectionsMedical DisclaimerAbout Our ContentAffiliate DisclosureSite Map