The Invisible Gorilla: And Other Ways Our Intuitions Deceive Us

by Christopher Chabris

  According to journalist Bob Woodward, in late 2002 President George W. Bush had doubts about launching an invasion of Iraq, so he asked CIA director George Tenet directly about the strength of the evidence that Saddam Hussein possessed unconventional weapons. Tenet said, “It’s a slam dunk case!” Bush repeated, “George, how confident are you?” Tenet’s reply: “Don’t worry, it’s a slam dunk!” Weeks into the war, White House spokesman Ari Fleischer expressed “high confidence” that weapons of mass destruction would yet be found. As of this writing, they are still missing, and an exhaustive government investigation has concluded that they were not there to be found.24

  Why does confidence have such a hold on us? Why do we have such an overwhelming and often unnoticed inclination to take another person’s outward confidence as an honest signal of his or her inner knowledge, ability, and resolve? As we have seen, the most incompetent among us tend to be the most overconfident, yet we still rely on confidence as an indicator of ability.

  Sometimes the Cream Doesn’t Rise to the Top

  Imagine being asked to work together with three other people—call them Jane, Emily, and Megan—to solve challenging math problems. You don’t know who in your group is good at math; you have only your (imperfect) knowledge of your own abilities. On the first problem, Jane is the first to suggest an answer, and Emily chimes in with her own thoughts. Megan is initially quiet, but after a while, she comes up with the correct answer and explains why the other answers were wrong. This happens several times, so it becomes clear to all that Megan is good at solving problems like these. The group comes to defer to her as their de facto leader, and it does very well on its task. In an ideal world, group dynamics would always work this way: The cream would rise to the top; all members would contribute their unique knowledge, skills, and competence; and group deliberation would lead to better decisions. But the reality of group performance too often diverges from this ideal.

  Chris once interviewed a U.S. government intelligence agent about group decision processes. The agent described a method his group sometimes used to arrive at a shared estimate for an unknown quantity: The members went around the room, each giving his or her own estimate, in descending order of seniority.25 Imagine the false sense of consensus and confidence that cascades through a group when one person after another confirms the boss’s original guess. Although each staff member could have offered an independent, unbiased, uninfluenced opinion by secret ballot, the chances of that happening in practice are virtually nil. The very process of putting individuals together to deliberate before they reach a conclusion almost guarantees that the group’s decision will not be the product of independent opinions and contributions. Instead, it will be influenced by group dynamics, personality conflicts, and other social factors that have little to do with who knows what, and why they know it.

  Rather than producing better understanding of abilities and more realistic expressions of confidence, group processes can inspire a feeling akin to “safety in numbers” among the most hesitant members, decreasing realism and increasing certainty. We think that this reflects another illusion people have about the mind—the misguided intuition that the best way for a group to use the abilities of its members in solving a problem is to deliberate over the correct answer and arrive at a consensus.

  Suppose you are in a group of people asked to estimate an unknown quantity, such as the number of jelly beans in a large jar. You might think the best approach would be to discuss the options with other members until you agree on an estimate, but you’d be wrong. A different strategy consistently outperforms all others: With no prior discussion, each person should write down his or her best estimate, and then the group should simply average together all the independent estimates.26 We asked Richard Hackman, a Harvard professor and expert on the psychology of groups, if he had ever heard of a group spontaneously deciding to use this procedure, as opposed to immediately launching into discussion and debate.27 He had not.

  Of course, in some contexts, the overconfidence derived from group consensus has value. In the midst of a military battle, nervous, low-confidence soldiers can draw strength from their comrades and leaders and take necessary risks—including the ultimate risk, their own lives—that they might choose not to take if they were alone in the decision. But the illusion of confidence can have tragic consequences when independent analysis and judgment of the highest quality are required. And just like individuals, groups appear to be totally unaware that they have this tendency to overstate their collective abilities.

  Cameron Anderson and Gavin Kilduff at Berkeley’s Haas School of Business actually conducted the math problem–solving experiment we asked you to imagine.28 They formed groups of four students who had never met one another and asked them to solve math questions from the GMAT, a standardized test used for admission to graduate business schools. An advantage of using math problems as the group task was that Anderson and Kilduff could objectively measure how well each group member performed by assessing (from the videos) how many good and bad solutions each member suggested. And they could compare group members’ perceptions of each person’s math competence with an objective measure of the person’s actual competence: scores on the math section of the SAT college admission test.

  Anderson and Kilduff recorded all of the group interactions on video and reviewed them later to determine who the group leaders were. They also asked outside observers to make the same determination, and they polled the members of each group to see whom they thought had taken the leadership roles. All parties identified the same people as group leaders. The important question was what factors determined which of the four members of each group became its leader. In the hypothetical example we used to start this section, the cream rose to the top and the best mathematician, Megan, emerged as the go-to group member.

  As you’ve probably anticipated, in the actual experiment, the group leaders proved to be no more competent than anyone else. They became leaders by force of personality rather than strength of ability. Before starting the group task, the participants completed a short questionnaire designed to measure how “dominant” they tended to be. Those people with the most dominant personalities tended to become the leaders. How did the dominant individuals become the group leaders even though they were no better at math? Did they bully the others into obeying, shouting down meek but intelligent group members? Did they campaign for the role, persuading others that they were the best at math, or at least the best at organizing their group? Not at all. The answer is almost absurdly simple: They spoke first. For 94 percent of the problems, the group’s final answer was the first answer anyone suggested, and people with dominant personalities just tend to speak first and most forcefully.

  So in this experiment, group leadership was determined largely by confidence. People with dominant personalities tend to exhibit greater self-confidence, and due to the illusion of confidence, others tend to trust and follow people who speak with confidence. If you offer your opinion early and often, people will take your confidence as an indicator of ability, even if you are actually no better than your peers. The illusion of confidence keeps the cream blended in. Only when confidence happens to be correlated with actual competence will the most able person rise to the top.

  The Trait of Confidence

  Psychologists use the term trait to describe a general characteristic of a person that influences his or her behavior in a wide variety of situations. In Anderson and Kilduff’s study of group leadership, dominance was taken to be a trait—people scoring high on the researchers’ dominance test were thought to assert control and assume power across a wide range of situations. Similarly, if you score high on a test of extraversion, you are probably more outgoing than the average person, and your tendency to approach and engage with other people will manifest itself more often than not.

  Personality traits don’t determine your behavior all the time—many other factors, especially ones particular to the situation you are in, have powerf
ul influences as well. An extraverted person who knows nothing about Star Trek might be more shy at a science-fiction convention than an introvert who attends these events all the time. However, extraverted people tend toward more social engagement in the absence of other overriding situational factors. By default, they are more socially gregarious than are introverted people.

  Confidence itself doesn’t show up in most of the lists of traits compiled by psychologists. It isn’t one of the so-called “big five” dimensions, which include neuroticism, extraversion, openness to experience, agreeableness, and conscientiousness. Confidence is related to, but is not the same as dominance, and even dominance isn’t typically measured in studies of personality. We think that differences among people in their tendency to express confidence are vitally important for understanding how we make decisions and influence one another. So do these differences exist? Is confidence a trait?

  The “con” part of “con man,” “con artist,” and “con game” is short for confidence. The original “confidence man” was a grifter in the 1840s named William Thompson, who had the audacity to approach strangers on the streets of Manhattan and simply ask them to hand over their watches. Attempting this gambit required Thompson to somehow gain the confidence of his marks; amazingly, he was able to do this while explicitly asking them, “Have you confidence in me to trust me with your watch until tomorrow?”29

  The most inherently confident person in history might have been Frank Abagnale, who was portrayed by Leonardo DiCaprio in Steven Spielberg’s movie Catch Me If You Can. Abagnale started early: While still a high school student, he successfully impersonated a high school teacher, and he conned his father out of $3,400. By the age of eighteen, pretending to be a Pan Am pilot, he had tricked the airline into letting him fly over one million miles as a “deadhead”—riding in unsold seats or as a guest in the cockpit. He expertly forged checks worth millions of dollars. When he was finally arrested in France, at the age of twenty-one, he was wanted in twelve countries. After being tried and serving time in France and Sweden, he was extradited to the United States, where he repeatedly escaped and eluded authorities, on one occasion by pretending to be an undercover investigator looking into allegations of poor conditions for prisoners. Eventually he was recaptured and convicted. As part of a deal with American prosecutors, he agreed to assist the FBI in future investigations of other frauds in exchange for early parole. The diversity, ease, and precociousness of his con games attest to his ability to display levels of confidence that people expect to see only in those who are telling the truth.30

  Chris and some of his colleagues wondered whether confidence is a stable trait, as the careers of Abagnale and Thompson suggest.31 They conducted a simple experiment to find out. Subjects were asked to answer a series of challenging true/false trivia questions, such as “the O.J. Simpson murder trial ended in 1993” (false—it ended in 1995), and to express their confidence in each answer as a percentage (between 50% and 100%). On this test, most people express considerable overconfidence: They get about 60 percent of the answers correct, but their average confidence is about 75 percent.

  The critical element in the design of this experiment was the creation of two trivia tests that were equally difficult but included entirely different questions. Each subject first completed one of the tests, and then several weeks later, completed the other one. Remarkably, just by knowing how confident someone was on the first test, it was possible to predict how confident they would be on the second test. Of those people who scored in the top half on confidence in the first test they took, 90 percent scored in the top half on the second test. Yet confidence did not predict accuracy; the more confident people were no more accurate than the less confident people. Confidence also was unrelated to intelligence.

  Other experiments have shown that confidence is a general trait: People who are highly confident of their skills in one domain, such as visual perception, also tend to be highly confident of their skills in other domains, such as memory.32 In short, confidence appears to be a consistent quality that varies from one person to the next, but has relatively little to do with one’s underlying knowledge or mental ability. One thing that does appear to influence confidence is genes. According to a recent study by a group of economists in Sweden, identical twins are more similar to each other in how confident they are of their own abilities than are fraternal twins.33 Since identical twins have essentially the same genes, but fraternal twins are no more similar genetically than ordinary siblings, confidence must have at least some genetic basis. Your confidence isn’t entirely determined by your genetic makeup, but it’s not entirely independent of it. As it turns out, Frank Abagnale’s father was also a con man; he lost the family home in a failed tax-fraud scheme.

  Why David Took on Goliath

  In August 2008, the tiny nation of Georgia provoked a military conflict with its northern neighbor Russia over two provinces whose separatist movements were being encouraged and supported by the Russian government. Georgia’s army was overwhelmed after less than one week of fighting, and Russia was left in control of the provinces. All that Georgia obtained from the war was some sympathy among Western governments. Incredibly, Georgia’s leaders actually believed their forces would quickly overtake the key points in South Ossetia and Abkhazia and that, once entrenched, they would be able to repel Russian counterattacks. “Several Georgian officials said that night that seizing South Ossetia would be militarily easy,” according to the New York Times. “Some administration officials said the Georgian military had drawn up a ‘concept of operations’ for a crisis in South Ossetia that called for its army units to sweep across the region and rapidly establish such firm control that a Russian response could be pre-empted.”34

  The Georgians were woefully overconfident in provoking war with the second-strongest military power in the world. In his book Overconfidence and War, political scientist Dominic Johnson of Princeton University analyzes a range of military turning points, from World War I to Vietnam to Iraq, and although he doesn’t use our terminology, he makes the point that almost any country that voluntarily initiates a war and then loses must have suffered from the illusion of confidence, since negotiation is always an option.35 When Mikheil Saakashvili was elected president of Georgia in 2004, he was only thirty-six years old. He stocked the government with loyal ministers who were also in their thirties and lacked military experience but sympathized with their leader’s views about the importance of reclaiming the breakaway regions from Russian influence. Over the next four years they managed to convince themselves that it was a good idea to fight an army that outnumbered theirs by twenty-five to one. It’s not hard to imagine how a group of like-minded government officials could take a set of opinions that none of them held with great confidence individually and aggregate them, by deliberating among themselves and reinforcing one another’s public statements, into a high-confidence conclusion.36

  Chris and his colleagues at Harvard tried to capture this process of confidence inflation in groups through an experiment. They began by giving seven hundred people one of the true/false trivia tests we just described. As usual, people thought they knew more than they did, averaging 70 percent confidence in their answers even though their actual scores averaged only 54 percent correct. Chris’s team used each person’s confidence on the first test to select members for three different types of two-person groups: groups with two high-confidence members, groups with two low-confidence members, and groups with one high-and one low-confidence member. Each pair then visited the laboratory, where they worked together on the second trivia test—one that was just as difficult but had different questions from the first one. The members of each group could share their thoughts, deliberate about the best answer, and make a collective judgment about how likely they were to be right.

  Our intuition tells us that groups should be more accurate and less overconfident than individuals. When two people come up with different answers to a trivia question, one of them mu
st be wrong. Such discrepancies should lead to two changes. First, they should spur further discussion, which should sometimes result in more accurate answers. Second, they should provide a signal to both individuals that their certainty in their own opinion may be too high, so the group’s collective certainty should decrease when there is disagreement.

  At least for this sort of trivia task, though, two heads weren’t better than one: The groups were no better at solving the trivia questions than were the individuals. But being part of a group did swell the heads of the subjects. Even though they were no more accurate, they were more confident!37 Confidence increased the most for pairs composed of two low-confidence people. The members of groups like this apparently reinforced each other, leading to an 11-percent increase in confidence despite no improvement in performance. This experiment illustrates why the Georgian government’s high-confidence decision to provoke war with Russia did not necessarily stem from the overconfident beliefs of any one individual. The people making these decisions might each have had low confidence, perhaps so low that they would not have given the order by themselves. In a group, however, their confidence could have inflated to the point where what were actually risky, uncertain actions seemed highly likely to succeed.