Thinking, Fast and Slow

by Daniel Kahneman

The prejudice against algorithms is magnified when the decisions are consequential. Meehl remarked, “I do not quite know how to alleviate the horror some clinicians seem to experience when they envisage a treatable case being denied treatment because a ‘blind, mechanical’ equation misclassifies him.” In contrast, Meehl and other proponents of algorithms have argued strongly that it is unethical to rely on intuitive judgments for important decisions if an algorithm is available that will make fewer mistakes. Their rational argument is compelling, but it runs against a stubborn psychological reality: for most people, the cause of a mistake matters. The story of a child dying because an algorithm made a mistake is more poignant than the story of the same tragedy occurring as a result of human error, and the difference in emotional intensity is readily translated into a moral preference.

Fortunately, the hostility to algorithms will probably soften as their role in everyday life continues to expand.

When can you trust an experienced professional who claims to have an intuition?

The model of intuitive decision making as pattern recognition develops ideas presented some time ago by Herbert Simon, perhaps the only scholar who is recognized and admired as a hero and founding figure by all the competing clans and tribes in the study of decision making. I quoted Herbert Simon’s definition of intuition in the introduction, but it will make more sense when I repeat it now: “The situation has provided a cue; this cue has given the expert access to information stored in memory, and the information provides the answer. Intuition is nothing more and nothing less than recognition.” This strong statement reduces the apparent magic of intuition to the everyday experience of memory. We marvel at the story of the firefighter who has a sudden urge to escape a burning house just before it collapses, because the firefighter knows the danger intuitively, “without knowing how he knows.” However, we also do not know how we immediately know that a person we see as we enter a room is our friend Peter. The moral of Simon’s remark is that the mystery of knowing without knowing is not a distinctive feature of intuition; it is the norm of mental life.

Emotional learning may be quick, but what we consider as “expertise” usually takes a long time to develop. The acquisition of expertise in complex tasks such as high-level chess, professional basketball, or firefighting is intricate and slow because expertise in a domain is not a single skill but rather a large collection of miniskills. Chess is a good example. An expert player can understand a complex position at a glance, but it takes years to develop that level of ability. Studies of chess masters have shown that at least 10,000 hours of dedicated practice (about 6 years of playing chess 5 hours a day) are required to attain the highest levels of performance. During those hours of intense concentration, a serious chess player becomes familiar with thousands of configurations, each consisting of an arrangement of related pieces that can threaten or defend each other. Learning high-level chess can be compared to learning to read. A first grader works hard at recognizing individual letters and assembling them into syllables and words, but a good adult reader perceives entire clauses. An expert reader has also acquired the ability to assemble familiar elements in a new pattern and can quickly “recognize” and correctly pronounce a word that she has never seen before. In chess, recurrent patterns of interacting pieces play the role of letters, and a chess position is a long word or a sentence. A skilled reader who sees it for the first time will be able to read the opening st...

Klein and I quickly found that we agreed both on the nature of intuitive skill and on how it is acquired. We still needed to agree on our key question: When can you trust a self-confident professional who claims to have an intuition? We eventually concluded that our disagreement was due in part to the fact that we had different experts in mind. Klein had spent much time with fireground commanders, clinical nurses, and other professionals who have real expertise. I had spent more time thinking about clinicians, stock pickers, and political scientists trying to make unsupportable long-term forecasts. Not surprisingly, his default attitude was trust and respect; mine was skepticism. He was more willing to trust experts who claim an intuition because, as he told me, true experts know the limits of their knowledge. I argued that there are many pseudo-experts who have no idea that they do not know what they are doing (the illusion of validity), and that as a general proposition subjective confidence is commonly too high and often uninformative.

the confidence that people have in their intuitions is not a reliable guide to their validity. In other words, do not trust anyone—including yourself—to tell you how much you should trust their judgment.

If subjective confidence is not to be trusted, how can we evaluate the probable validity of an intuitive judgment? When do judgments reflect true expertise? When do they display an illusion of validity? The answer comes from the two basic conditions for acquiring a skill: an environment that is sufficiently regular to be predictable an opportunity to learn these regularities through prolonged practice When both these conditions are satisfied, intuitions are likely to be skilled. Chess is an extreme example of a regular environment, but bridge and poker also provide robust statistical regularities that can support skill. Physicians, nurses, athletes, and firefighters also face complex but fundamentally orderly situations. The accurate intuitions that Gary Klein has described are due to highly valid cues that the expert’s System 1 has learned to use, even if System 2 has not learned to name them. In contrast, stock pickers and political scientists who make long-term forecasts operate in a zero-validity environment. Their failures reflect the basic unpredictability of the events that they try to forecast.

human learning is normally efficient. If a strong predictive cue exists, human observers will find it, given a decent opportunity to do so. Statistical algorithms greatly outdo humans in noisy environments for two reasons: they are more likely than human judges to detect weakly valid cues and much more likely to maintain a modest level of accuracy by using such cues consistently.

At the end of our journey, Gary Klein and I agreed on a general answer to our initial question: When can you trust an experienced professional who claims to have an intuition? Our conclusion was that for the most part it is possible to distinguish intuitions that are likely to be valid from those that are likely to be bogus. As in the judgment of whether a work of art is genuine or a fake, you will usually do better by focusing on its provenance than by looking at the piece itself. If the environment is sufficiently regular and if the judge has had a chance to learn its regularities, the associative machinery will recognize situations and generate quick and accurate predictions and decisions. You can trust someone’s intuitions if these conditions are met.

It is difficult to reconstruct what it was that took us years, long hours of discussion, endless exchanges of drafts and hundreds of e-mails negotiating over words, and more than once almost giving up. But this is what always happens when a project ends reasonably well: once you understand the main conclusion, it seems it was always obvious.

The inside view is the one that all of us, including Seymour, spontaneously adopted to assess the future of our project. We focused on our specific circumstances and searched for evidence in our own experiences. We had a sketchy plan: we knew how many chapters we were going to write, and we had an idea of how long it had taken us to write the two that we had already done. The more cautious among us probably added a few months to their estimate as a margin of error. Extrapolating was a mistake. We were forecasting based on the information in front of us—WYSIATI—but the chapters we wrote first were probably easier than others, and our commitment to the project was probably then at its peak. But the main problem was that we failed to allow for what Donald Rumsfeld famously called the “unknown unknowns.” There was no way for us to foresee, that day, the succession of events that would cause the project to drag out for so long. The divorces, the illnesses, the crises of coordination with bureaucracies that delayed the work could not be anticipated. Such events not only cause the writing of chapters to slow down, they also produce long periods during which little or no progress is made at all. The same must have been true, of course, for the other teams that Seymour knew about. The members of those teams were also unable to imagine the events that would cause them to spend seven years to finish, or ultimately fail to finish, a project that they evidently had thought was very fea...

if the reference class is properly chosen, the outside view will give an indication of where the ballpark is, and it may suggest, as it did in our case, that the inside-view forecasts are not even close to it.

Amos and I coined the term planning fallacy to describe plans and forecasts that are unrealistically close to best-case scenarios could be improved by consulting the statistics of similar cases Examples of the planning fallacy abound in the experiences of individuals, governments, and businesses. The list of horror stories is endless.

The diagnosis of and the remedy for the planning fallacy have not changed since that Friday afternoon, but the implementation of the idea has come a long way. The renowned Danish planning expert Bent Flyvbjerg, now at Oxford University, offered a forceful summary: The prevalent tendency to underweight or ignore distributional information is perhaps the major source of error in forecasting. Planners should therefore make every effort to frame the forecasting problem so as to facilitate utilizing all the distributional information that is available. This may be considered the single most important piece of advice regarding how to increase accuracy in forecasting through improved methods. Using such distributional information from other ventures similar to that being forecasted is called taking an “outside view” and is the cure to the planning fallacy. The treatment for the planning fallacy has now acquired a technical name, reference class forecasting, and Flyvbjerg has applied it to transportation projects in several countries. The outside view is implemented by using a large database, which provides information on both plans and outcomes for hundreds of projects all over the world, and can be used to provide statistical information about the likely overruns of cost and time, and about the likely underperformance of projects of different types. The forecasting method that Flyvbjerg applies is similar to the practices recommended for overcoming base-rate neglect: Identify an...

When forecasting the outcomes of risky projects, executives too easily fall victim to the planning fallacy. In its grip, they make decisions based on delusional optimism rather than on a rational weighting of gains, losses, and probabilities. They overestimate benefits and underestimate costs. They spin scenarios of success while overlooking the potential for mistakes and miscalculations. As a result, they pursue initiatives that are unlikely to come in on budget or on time or to deliver the expected returns—or even to be completed. In this view, people often (but not always) take on risky projects because they are overly optimistic about the odds they face. I will return to this idea several times in this book—it probably contributes to an explanation of why people litigate, why they start wars, and why they open small businesses.

The outside view was much easier to ignore than bad news in our own effort.

I hope I am wiser today, and I have acquired a habit of looking for the outside view. But it will never be the natural thing to do.

The planning fallacy is only one of the manifestations of a pervasive optimistic bias. Most of us view the world as more benign than it really is, our own attributes as more favorable than they truly are, and the goals we adopt as more achievable than they are likely to be. We also tend to exaggerate our ability to forecast the future, which fosters optimistic overconfidence. In terms of its consequences for decisions, the optimistic bias may well be the most significant of the cognitive biases. Because optimistic bias can be both a blessing and a risk, you should be both happy and wary if you are temperamentally optimistic.

two aspects of choice that the standard model of indifference curves does not predict. First, tastes are not fixed; they vary with the reference point. Second, the disadvantages of a change loom larger than its advantages, inducing a bias that favors the status quo. Of course, loss aversion does not imply that you never prefer to change your situation; the benefits of an opportunity may exceed even overweighted losses. Loss aversion implies only that choices are strongly biased in favor of the reference situation (and generally biased to favor small rather than large changes).

The question of when an approach or a movement got its start is often difficult to answer, but the origin of what is now known as behavioral economics can be specified precisely. In the early 1970s, Richard Thaler, then a graduate student in the very conservative economics department of the University of Rochester, began having heretical thoughts. Thaler always had a sharp wit and an ironic bent, and as a student he amused himself by collecting observations of behavior that the model of rational economic behavior could not explain. He took special pleasure in evidence of economic irrationality among his professors, and he found one that was particularly striking. Professor R (now revealed to be Richard Rosett, who went on to become the dean of the University of Chicago Graduate School of Business) was a firm believer in standard economic theory as well as a sophisticated wine lover. Thaler observed that Professor R was very reluctant to sell a bottle from his collection—even at the high price of $100 (in 1975 dollars!). Professor R bought wine at auctions, but would never pay more than $35 for a bottle of that quality. At prices between $35 and $100, he would neither buy nor sell. The large gap is inconsistent with economic theory, in which the professor is expected to have a single value for the bottle. If a particular bottle is worth $50 to him, then he should be willing to sell it for any amount in excess of $50. If he did not own the bottle, he should be willing to pay...

The fundamental ideas of prospect theory are that reference points exist, and that losses loom larger than corresponding gains. Observations in real markets collected over the years illustrate the power of these concepts. A study of the market for condo apartments in Boston during a downturn yielded particularly clear results. The authors of that study compared the behavior of owners of similar units who had bought their dwellings at different prices. For a rational agent, the buying price is irrelevant history—the current market value is all that matters. Not so for Humans in a down market for housing. Owners who have a high reference point and thus face higher losses set a higher price on their dwelling, spend a longer time trying to sell their home, and eventually receive more money.

The psychologist Paul Rozin, an expert on disgust, observed that a single cockroach will completely wreck the appeal of a bowl of cherries, but a cherry will do nothing at all for a bowl of cockroaches.

Loss aversion refers to the relative strength of two motives: we are driven more strongly to avoid losses than to achieve gains. A reference point is sometimes the status quo, but it can also be a goal in the future: not achieving a goal is a loss, exceeding the goal is a gain. As we might expect from negativity dominance, the two motives are not equally powerful. The aversion to the failure of not reaching the goal is much stronger than the desire to exceed it.

we refuse to cut losses when doing so would admit failure, we are biased against actions that could lead to regret, and we draw an illusory but sharp distinction between omission and commission, not doing and doing, because the sense of responsibility is greater for one than for the other. The ultimate currency that rewards or punishes is often emotional, a form of mental self-dealing that inevitably creates conflicts of interest when the individual acts as an agent on behalf of an organization.

Theoretical beliefs are robust, and it takes much more than one embarrassing finding for established theories to be seriously questioned.

Italy and France competed in the 2006 final of the World Cup. The next two sentences both describe the outcome: “Italy won.” “France lost.” Do those statements have the same meaning? The answer depends entirely on what you mean by meaning. For the purpose of logical reasoning, the two descriptions of the outcome of the match are interchangeable because they designate the same state of the world. As philosophers say, their truth conditions are identical: if one of these sentences is true, then the other is true as well. This is how Econs understand things. Their beliefs and preferences are reality-bound. In particular, the objects of their choices are states of the world, which are not affected by the words chosen to describe them. There is another sense of meaning, in which “Italy won” and “France lost” do not have the same meaning at all. In this sense, the meaning of a sentence is what happens in your associative machinery while you understand it. The two sentences evoke markedly different associations. “Italy won” evokes thoughts of the Italian team and what it did to win. “France lost” evokes thoughts of the French team and what it did that caused it to lose, including the memorable head butt of an Italian player by the French star Zidane. In terms of the associations they bring to mind—how System 1 reacts to them—the two sentences really “mean” different things. The fact that logically equivalent statements evoke different reactions makes it impossible for Humans to b...

Would you accept a gamble that offers a 10% chance to win $95 and a 90% chance to lose $5? Would you pay $5 to participate in a lottery that offers a 10% chance to win $100 and a 90% chance to win nothing? First, take a moment to convince yourself that the two problems are identical. In both of them you must decide whether to accept an uncertain prospect that will leave you either richer by $95 or poorer by $5. Someone whose preferences are reality-bound would give the same answer to both questions, but such individuals are rare. In fact, one version attracts many more positive answers: the second. A bad outcome is much more acceptable if it is framed as the cost of a lottery ticket that did not win than if it is simply described as losing a gamble. We should not be surprised: losses evokes stronger negative feelings than costs. Choices are not reality-bound because System 1 is not reality-bound.

An experiment that Amos carried out with colleagues at Harvard Medical School is the classic example of emotional framing. Physician participants were given statistics about the outcomes of two treatments for lung cancer: surgery and radiation. The five-year survival rates clearly favor surgery, but in the short term surgery is riskier than radiation. Half the participants read statistics about survival rates, the others received the same information in terms of mortality rates. The two descriptions of the short-term outcomes of surgery were: The one-month survival rate is 90%. There is 10% mortality in the first month. You already know the results: surgery was much more popular in the former frame (84% of physicians chose it) than in the latter (where 50% favored radiation). The logical equivalence of the two descriptions is transparent, and a reality-bound decision maker would make the same choice regardless of which version she saw. But System 1, as we have gotten to know it, is rarely indifferent to emotional words: mortality is bad, survival is good, and 90% survival sounds encouraging whereas 10% mortality is frightening. An important finding of the study is that physicians were just as susceptible to the framing effect as medically unsophisticated people (hospital patients and graduate students in a business school). Medical training is, evidently, no defense against the power of framing.

A comment I heard from a member of the audience after a lecture illustrates the difficulty of distinguishing memories from experiences. He told of listening raptly to a long symphony on a disc that was scratched near the end, producing a shocking sound, and he reported that the bad ending “ruined the whole experience.” But the experience was not actually ruined, only the memory of it. The experiencing self had had an experience that was almost entirely good, and the bad end could not undo it, because it had already happened. My questioner had assigned the entire episode a failing grade because it had ended very badly, but that grade effectively ignored 40 minutes of musical bliss. Does the actual experience count for nothing? Confusing experience with the memory of it is a compelling cognitive illusion—and it is the substitution that makes us believe a past experience can be ruined. The experiencing self does not have a voice. The remembering self is sometimes wrong, but it is the one that keeps score and governs what we learn from living, and it is the one that makes decisions. What we learn from the past is to maximize the qualities of our future memories, not necessarily of our future experience. This is the tyranny of the remembering self.

Decisions that do not produce the best possible experience and erroneous forecasts of future feelings—both are bad news for believers in the rationality of choice. The cold-hand study showed that we cannot fully trust our preferences to reflect our interests, even if they are based on personal experience, and even if the memory of that experience was laid down within the last quarter of an hour! Tastes and decisions are shaped by memories, and the memories can be wrong. The evidence presents a profound challenge to the idea that humans have consistent preferences and know how to maximize them, a cornerstone of the rational-agent model. An inconsistency is built into the design of our minds. We have strong preferences about the duration of our experiences of pain and pleasure. We want pain to be brief and pleasure to last. But our memory, a function of System 1, has evolved to represent the most intense moment of an episode of pain or pleasure (the peak) and the feelings when the episode was at its end. A memory that neglects duration will not serve our preference for long pleasure and short pains.

Daniel Gilbert and Timothy Wilson introduced the word miswanting to describe bad choices that arise from errors of affective forecasting. This word deserves to be in everyday language. The focusing illusion (which Gilbert and Wilson call focalism) is a rich source of miswanting. In particular, it makes us prone to exaggerate the effect of significant purchases or changed circumstances on our future well-being.