65Mindwise

Nicholas Epley, Mindwise: How We Understand What Others Think, Believe, Feel, and Want, Knopf, 2014

Linked to 36 chapters: supports 21 · contradicts 4 · mechanism 18 · same claim 5 · pairing 5

Epley studies the sense by which one person reads another, and he calls it a sixth sense: the ability to infer from a face, a posture and a sentence what someone else thinks, believes, feels and wants. He does not argue that the sense is broken. It beats chance almost everywhere it has been tested, and nothing else the brain does makes coordination between strangers possible. He argues that confidence in it outruns its accuracy, and that the gap between the two is where marriages, negotiations and wars go wrong. The case is built in four moves. First he puts mind reading to the test, including on the mind we are surest of, and finds the accuracy modest and the confidence high. Then he takes the two errors that come before any reading: failing to engage the sense, which makes another person look relatively mindless, and engaging it where there is nothing to read, which puts minds into alarm clocks, hurricanes and gods. Then he takes the three tools the sense uses once engaged — your own mind, what you know about the other person's group, and what you can watch them do — and names the mistake each produces. He ends with the remedy, which is not the one people expect. Reading faces does not work, and imagining yourself in someone's position does not work. Asking them does.

An overconfident sense

The book opens with George W. Bush saying he could read Vladimir Putin by looking him in the eye, and with the evidence that impressions do form that fast. Fifty milliseconds of a face is enough to produce a judgment of competence, and in one experiment the candidate who looked more competent after such a glance won about 70 per cent of the time. Speed breeds confidence. Asked whose mind they would read if they could, five hundred Americans mostly named a spouse, a boss or a neighbour, and wanted to know what that person thought of them.

That is one of the harder things to know. A set of experiments with a common design has people in a group predict how each other member will rate them, and compares prediction with rating. Predicting the group's average view of you works about as well as predicting a son's height from his father's; predicting any single person's view of you barely beats guessing.

The last row is Epley's own work with Tal Eyal. People who have looked at their own face every morning of their lives cannot predict how a stranger will rate a photograph of it; the predictions bore no relation whatever to the ratings. Deception is no better. A survey of hundreds of experiments found people separating truths from lies 54 per cent of the time when chance is 50.

Familiarity helps less than it feels. William Ickes reports that strangers read each other's moment-to-moment thoughts with about 20 per cent accuracy and that close friends and married couples raise this to about 35. The trouble is the number that goes with it. In a study modelled on a television game show, dating partners predicted each other's answers about self-worth correctly 44 per cent of the time and believed they had got 82 per cent right. The longer a couple had been together the more they thought they knew, and length of relationship was not related to accuracy at all. People judging videotaped claims about HIV status were right 52 per cent of the time and confident 70 per cent of the time, and more information raised only the confidence.

Epley then turns the test on the one mind that seems exempt. In 1933 the sociologist Richard LaPiere drove ten thousand miles across the United States with a young Chinese couple, seeking service at 184 restaurants and 67 hotels. They were refused once. Six months later he wrote to the same establishments asking whether they would accept Chinese guests, and more than nine in ten said no. Honours students two months from a deadline predicted thirty-four days realistically and forty-nine at worst; they took fifty-five.

What introspection misses, Epley says, is construction. The mind is not an iceberg with its bulk below the surface and joined to the top; it is a house, whose finished rooms you can describe exactly and whose building you never saw. You know what you feel and want and believe. When you say why, you are guessing. "Introspection is blind to construction." In one experiment people chose the more attractive of two photographs, were handed back the one they had rejected by sleight of hand, and only 27 per cent noticed; the rest explained a choice they had not made as fluently as anyone explained a real one. The consequence is naïve realism: colour is constructed inside the head yet seems to sit on the apple, and one's own view of a political question likewise seems to be simply the view of it, so that whoever disagrees must be biased.

How we dehumanize

On 2 May 1879, in a Nebraska courtroom, Standing Bear of the Ponca rose to argue that he was a person. The government had moved 752 Poncas six hundred miles to the Indian Territory, where a third died in the first year, including his son. Carrying the boy's bones home, he was arrested, and General George Crook, unable to disobey the order to return him, arranged instead to be sued as the government's representative. Standing Bear held out his hand and said that if it were pierced it would hurt, and the blood would be the same colour as the court's. Judge Dundy ruled that he was a person under the law.

Dehumanisation, in Epley's definition, is failing to recognise the fully human mind of another person, and it does not require hatred. Its ordinary cause is distance, physical and psychological, and it works by leaving the sense switched off. Two things switch it on. The first is the senses: two people share attention, one body imitates the other, and the imitation produces a shade of the experience. Disable the imitation, with a pen held between the lips or with Botox, and the ability to read what another person feels falls measurably. War shows the dependence on presence most starkly: interviews with American soldiers in the Second World War found only 15 to 20 per cent able to fire at a visible enemy, and at Rosebud Creek in 1876 Crook's own men fired 25,000 musket balls to inflict ninety-nine casualties.

The second trigger is inference, and it has an address. Acupuncturists watching videos of needles entering a hand show almost nothing in the regions that fire when one is hurt oneself; instead they engage the medial prefrontal cortex, the region used for reasoning about what someone else is thinking. A patient, Epley notes, does not want a doctor's empathy but a doctor's inference. That region is engaged less for people who are psychologically distant. In one scanner study, American students looking at photographs of homeless people showed less of it than for any other group and more of the pattern produced by disgusting objects, and outside the scanner rated the same people as less intelligent, less articulate and less emotional.

The result is what Epley calls the lesser minds effect: whatever is not read is assumed to be thinner. People rate their own future as less predictable than a roommate's. Members of distant groups are judged less able to feel shame, pride or remorse, which is why their apologies are discounted as insincere. The workplace version concerns motivation. Epley's Chicago students report every year that meaning and pride matter more to them than to their classmates, who are in it for the money; of 242 junior American officers who had left the service, 73 per cent rated pay the least important reason and described bureaucracy instead.

The cost is also personal. Epley and Juliana Schroeder asked Chicago commuters which would make the better journey: solitude, talking to the stranger beside them, or whatever they normally did. Commuters predicted that conversation would be worst and solitude best. Randomly assigned to do those things, commuters from the same population reversed the order, with no loss of reported productivity.

How we anthropomorphize

The opposite error is the same machinery running where there is nothing to read. Ray Nagin explained Hurricane Katrina by saying God was angry with America. Analysts describe a market as pleased or as trying to recoup its losses and then trade on the inference. A car's front end reads as a face, and the designer of the Dodge Charger said plainly that a mean one was the goal. Anthropomorphism and dehumanisation, Epley argues, are two sides of one coin: the same sense, engaged too easily in one case and not at all in the other. His dog makes the point. In one experiment the guilty look appeared in response to the owner's disapproval whether or not the dog had done anything wrong; the guilt was in the owner's head.

The question is therefore not which things really have minds but what pulls the trigger, and Epley names three: perception, explanation and connection.

Perception works on similarity, and above all on eyes. Fake eyes are assumed to deter birds, but Martin Stevens backed some mealworm feeders with eyespots and others with squares of equal contrast, and found birds deterred by conspicuousness rather than by eyes; the resemblance fools people, not predators. On people it works very well. Psychology professors at Newcastle paying for coffee at an honesty box paid about three times more in the weeks when the notice above it showed eyes than in the weeks when it showed flowers. Movement matters more than appearance, and the effective speed is roughly one's own: robots made much faster or slower read as machines.

Explanation works on unpredictability. The physicists who ran the Tevatron called it the Tev and said it was not happy with them, but only on the days it failed; a roboticist said personality arrives with steering that is a little loose. In a survey of nearly nine hundred listeners to a car repair programme, the less reliable a car was the more its owner reported it having beliefs, desires and a personality, and a survey of computer users found the same. Epley and his collaborators described identical gadgets to different people as predictable or unpredictable — an alarm clock that runs away from you either on a setting you choose or on its own account — and the unpredictable versions were rated as having intentions, free will and feelings. In the scanner, those gadgets engaged the same medial prefrontal region used for people. A mind is the one explanation that fits anything. A film made in the 1940s of a large triangle, a small triangle and a circle moving about was described by all but one of thirty-four viewers as a story of motives, because as bare motion it makes no sense at all.

Connection is the third, and in the car survey the strongest: how much someone liked their car predicted its apparent mind almost twice as well as how often it broke down. A photograph of an undergraduate's favourite television character produces the effect a real observer produces, better performance on easy tasks and worse on hard ones, while a photograph of their least favourite character from the same programme does not. People who report chronic loneliness rate gadgets, pets and the universe as more minded, and making people feel lonely in an experiment raises their reported belief in God.

This is where the book's machines appear. Sixty per cent of people who own a robotic vacuum cleaner give it a name, and nobody names the upright one. Users hated the Microsoft Word assistant because its face and manner made it a colleague with no social skills, and armies fit war robots with eyes and skin because a machine that looks back is more frightening. Epley refuses to treat any of this as stupidity. Recognising a mind in a machine is the same ability that recognises a mind in a person, and of the two errors only one is dangerous: treating a gadget as minded looks silly, while failing to see a mind that is there is the ingredient of inhumanity.

The trouble of getting over yourself

Eleven days before the 2008 election, a campaign volunteer named Ashley Todd reported being assaulted by a man who carved a B into her cheek. The letter was backwards, as it would be in a mirror. She had done it herself, and she had done it from her own point of view.

Egocentrism is not outgrown, Epley argues; it is corrected. He and his colleagues sat parents and children at the Boston Children's Museum on opposite sides of a grid of boxes, some screened so that the instructor could not see what the mover could. Asked to move the small truck, adults took the egocentric object — the smallest one, which only they could see — 25 per cent of the time and children 50 per cent. The eye tracker told a different story: both groups looked first at the egocentric object, at the same latency and with the same frequency. What separated them was the speed of the correction, not the absence of the reflex.

Two kinds of egocentrism follow, and Epley separates them by where the difference lies. In the neck problem two people attend to different things. In the lens problem they attend to the same thing and interpret it differently.

The neck problem produces results that look silly and are common. Married couples asked what share of the housework each is responsible for produce sums well over 100 per cent, and, less obviously, do the same for messing up the house and starting arguments: you notice your own contributions, good and bad. Among 699 Harvard students working in study groups, the overclaiming grew with group size until groups of eight were claiming nearly 140 per cent of the work between them. Asked first to write down what each other member had contributed, they came back close to the arithmetic. The same error pointed inward is the social spotlight: students made to wear a Barry Manilow T-shirt into a room of strangers estimated that half would name the face, and 23 per cent could.

The lens problem is harder, because a lens is invisible from behind. Epley's students, rating questionable business practices, divide anywhere from 6 to 94 per cent, and never estimate that fewer than half the class agrees with them, so even the moral minority believes itself a majority.

Knowledge is the lens that has been studied most. Readers asked to count the six f's in a short block of capitals usually find three, because expertise turns the f in "of" into a v before it can be counted; first graders do better. Elizabeth Newton's unpublished dissertation is the clean case: people who tapped out well-known tunes on a table predicted that listeners would name them half the time, and listeners named 2.5 per cent. Everyone is an expert on one subject, which is why the photograph result comes out as it does: people predicting their own rating list the hair above the left eye and a smile narrower than usual, while the raters write things like average height and weight. When Epley instead asked people to predict how they would be judged by someone seeing the photograph three months later, accuracy rose from zero to .55, because the distant question forces the broad lens a stranger actually uses.

The less you know about a mind, the more of your own you pour into it. Epley and Justin Kruger had people deliver sincere and sarcastic sentences by telephone or by e-mail. Senders expected to be understood about 80 per cent of the time either way; receivers managed it on the telephone, were at chance by e-mail, and in both conditions believed they had got nine out of ten right. The same logic runs to the least answerable mind of all. Believers' estimates of God's beliefs track their own more closely than their estimates of other people's, and in a scanner, reasoning about God's beliefs could not be told apart from reasoning about one's own. When persuasive arguments about affirmative action moved people's own opinions, their estimates of God's opinion moved in step and their estimates of other people's did not move at all.

The neck can be turned on purpose; the lens cannot. Asking people to put themselves in the evaluator's shoes did nothing at all for accuracy on the photographs. Epley's example of the limit is waterboarding, which a member of Congress judged not to be torture without having experienced it and which a Chicago radio host, having said the same, endured for seven seconds before calling it exactly that. You do not get past the lens by imagining harder. You get past it by being in the position, or by hearing from someone who has been.

Stereotypes, and how actions mislead

Learn that someone belongs to a group and you put down egocentrism and pick up a stereotype. Supervisors at a manufacturing plant in the 1950s held much the same view of what matters in a job as their workers did, and predicted their workers' answers as though guessing about another species: they switched to a tool that seemed to fit better and it made them worse.

Epley's defence of stereotypes rests on a demonstration with circles. Shown a set of circles and then a single test circle, people cannot say whether the test circle was in the set; what they do instead is accept, almost without exception, a circle of the set's average diameter that was never shown. The brain extracts the average of a group in under half a second and discards the members. Applied to people this is a serviceable instrument, and it gets the direction of most group differences right. What it gets wrong is the size.

Asked which of two distributions of national wealth they would rather be dropped into at random, 92 per cent of a thousand Americans chose the more equal one. Democratic and Republican voters differed by 3.5 percentage points, and Epley's 481 respondents predicted 35; rich and poor differed by 3 points against a predicted 40, men and women by 2 against a predicted 12. Measured differences between the sexes in reading emotions run to an effect size of about .2, and when Epley and Eyal asked people to estimate it they said about .7.

Epley borrows Robin Hogarth's term and calls the problem a wicked environment: the statistician is fine, the data are bad. The sample is small and unrepresentative, gathered from a narrow circle and from secondhand stories. Groups are defined by their differences, so what makes them alike is invisible by construction. And the cause of a difference cannot be seen, which is the failure Epley thinks does the damage, because a stereotype can be accurate while the explanation for it is entirely wrong. Cognitive decline with age is smaller in cultures that revere the elderly, and among people whose views of ageing were measured young and whose health was then tracked for thirty-eight years, 56 per cent of those with negative views had a major cardiovascular event against 18 per cent of those with positive views. Black students at Stanford did worse on a verbal test when the cover sheet asked them to state their race than when that line was absent.

Stereotypes are also dropped fast: viewers of a fifteen-second interview clearly used ethnic stereotypes and, after twelve minutes of the same person, showed no trace of them. What replaces the stereotype is the third tool, watching what a person does, and it has a failure of its own. Epley calls it by its usual name, the correspondence bias: the assumption that a mind matches the actions it produces. The demonstration is a rigged quiz. One volunteer writes ten hard questions out of their own knowledge and another has to answer them; contestants managed four of ten. The audience rated the questioners at the eighty-second percentile for general knowledge and the contestants at the forty-ninth, and the contestants agreed. Only the questioners, who knew how many questions they had discarded for not knowing the answers themselves, rated the two as equal.

The stakes are practical. Those who stayed in New Orleans during Hurricane Katrina were described by the head of Homeland Security as having chosen to disobey, and in a survey Americans reached for the words stupid, passive and inflexible; they were disproportionately poor, with larger families and no car. Convincing them was not the missing ingredient. A bus was. Policies built on the same mistake fail the same way. Paying students and teachers for performance, in trials costing some eighty million dollars, produced no improvement in achievement, attendance or graduation; the fixes Epley prefers act on the situation rather than on the will.

The habit starts early and is never removed, only overridden: three-year-olds say a rock is pointy so that animals can scratch on it, and the explanation returns in adults the moment you distract them with arithmetic. The bystanders who stepped over a man dying of a heart attack in a crowded store on the busiest shopping day of the year were not, Epley argues, greedier than anyone else. They were in a crowd, where an ambiguous event is read off the faces of others who are all reading it off yours. The six people who eventually helped him were nurses, who know what an emergency looks like.

How, and how not, to be a better mind reader

Two remedies are usually proposed, and neither survives the evidence. The first is learning to read bodies. When storytellers describe an emotional experience and observers track how they feel moment by moment, observers who can only hear the storyteller do significantly better than observers who can only see them: emotion travels on the voice. "Body language speaks, but only in whispers." The strong form of the idea fares worse. Out of 697 facial expressions recorded while people concealed or revealed an emotion, not one full microexpression appeared and fourteen partial ones did, half of them while the person was telling the truth. The Transportation Security Administration built a screening programme on the idea anyway. Of about two billion passengers passing through its airports between 2004 and 2008, some 152,000 were pulled aside and 1,083 arrested — under one per cent of those selected, no terrorists, and immigration offences accounting for forty per cent of the arrests.

The second remedy is perspective taking, and its weakness is that it combines the two tools that have already failed: a stereotype builds the other person's situation and your own mind runs it. If the stereotype is wrong, taking the perspective carefully makes the error worse. Epley's group simulated a fisheries dilemma in which four parties negotiated a shared catch. Those asked first to consider what each other party would think fair trusted each other less, took more, and collapsed the stock fastest. In a study of 104 couples, control partners predicted each other's attitudes with a correlation of .5 and an average error of 1.5 points on an eleven-point scale. Partners asked to write about a typical day in the other's life and then answer from inside it did slightly worse: .39, and an error of 1.7.

The same experiment had a third condition. One partner was allowed to interview the other with the questionnaire first, talking it through without writing anything down, and then predict from memory. Their error fell to about half the control group's. Asked afterwards how confident they were, the three groups barely differed: the group with the worst method was the most overconfident and the group with the best the least, but nobody's confidence tracked their accuracy.

Epley's public case is the American military's ban on open service by gay soldiers. In 2010, 1,167 retired officers used their experience to imagine the consequences of repeal and wrote to the President predicting damage to morale, discipline and cohesion. The Pentagon surveyed 115,052 soldiers and 44,266 spouses: 70 per cent expected no effect or a positive one, and of the 69 per cent who had already worked with a gay service member, 92 per cent said it had made no difference or had helped. A study a year after repeal found a non-event.

Epley then turns to the conditions under which people will actually tell you. Lying is rarer than assumed and most of it is done by a few chronic liars; what blocks the truth more often is fear of what follows it, which is why interrogators who establish rapport get more than those who apply pressure, and why Epley's answer to a five-year-old who lied reflexively was a cooling-off period and a standing offer of immunity. Ask what rather than why, as pollsters do, because the answer to why is a construction. And verify, as the talking stick does by making the next speaker restate the last speaker's position to that speaker's satisfaction before adding anything: its value is in what it does to listening.

The book's last section argues the other half, that being read requires opening up: when the University of Michigan's hospitals began disclosing medical errors to patients in 2001 and offering compensation, malpractice suits fell from thirty-nine a year to seventeen. The afterword returns to October 1962, when the two governments closest to destroying the world were communicating through intermediaries and reading an imagined adversary, and were pulled back by a private exchange of letters in which Khrushchev asked Kennedy to consider his position and then described it. Kennedy concluded that the hard-liners on each side fed on each other and that his counterpart's problem resembled his own. The agreement that followed established a direct line between the two capitals, and Epley's closing proposal is that everyone could use one.

Since 2014

The negative results have held better than most of the field. The 54 per cent figure for lie detection is among the most reproduced numbers in social psychology, and behaviour-based screening has gone on failing: the Government Accountability Office recommended in 2013 that Congress stop funding the airport programme Epley criticises, and the corps of behaviour detection officers was cut back. The finding that perspective taking does not raise accuracy was extended by Eyal, Steffel and Epley in 2018 across two dozen experiments covering strangers, spouses, facial expressions and lies, with improvement appearing only when people were given the other's perspective directly. The commuter-train result has been repeated in London and has grown into a literature on undersociality: people underestimate how much a stranger liked them after a conversation, and how welcome an unprompted message is.

Several of the supporting studies have not survived. The eyes-on-the-honesty-box effect did not hold up in larger pre-registered work, and meta-analysis has found no reliable effect of watching eyes on generosity. The facial feedback result, which Epley uses for the imitation step of his mechanism, failed a seventeen-laboratory replication in 2016, though a larger study in 2022 recovered a small effect for posed expressions. Stereotype threat's effect on test scores has been contested on grounds of publication bias.

What has aged into new relevance is the account of anthropomorphism. The three triggers Epley named with Adam Waytz and John Cacioppo have become the standard explanation, and conversational software pulls all three at once: it produces humanlike language, it behaves unpredictably enough to need explaining, and it arrives to people who want company. The lonely-users result is no longer a curiosity. His claim that the two errors are not morally symmetrical, and that failing to see a mind that is there is the worse one, is the sentence the work is now most often cited for.