Insight Editions · No. 28
Subliminal
How Your Unconscious Mind Rules Your Behavior
Leonard Mlodinow
Leonard Mlodinow explains how unconscious processes shape perception, memory, emotion, social judgment, and behavior, often guiding decisions before conscious awareness can influence them.
A study guide to Subliminal by Leonard Mlodinow. Shared free for personal reading; not for sale and not endorsed by the authors or publishers. Please read and buy the original.
Prologue
Mlodinow opens in June 1879, aboard a steamship from Boston to New York. The philosopher and scientist Charles Sanders Peirce has his gold watch stolen from his stateroom. He lines up the crew, talks to each man, and finds no clue. Then he forces himself to name a suspect anyway, picks one man on a hunch, and is right: after the ship docks he hires a Pinkerton detective, who recovers the watch from a pawnshop and traces it to exactly the man Peirce had guessed.
The anecdote would be folklore except that Peirce turned it into science. Five years later, with his student Joseph Jastrow, he ran an experiment on judging tiny differences in the weight of objects. Subjects felt so unsure that they believed they were guessing, yet they picked the heavier object on more than 60 percent of trials, well above chance. Peirce called this faculty "the loftiest of our merely instinctive powers." It is the book's first exhibit of knowing without knowing that you know. The design Peirce and Jastrow used, making people choose even when they insist they cannot tell, became the forced-choice method that recurs throughout the book, from blindsight to unconscious vision experiments. Mlodinow argues that Peirce, Wundt, Carpenter, Jastrow and William James, more than Freud, are the true ancestors of the modern science of the unconscious.
There is a broader lesson in the watch story too. Peirce did not reason his way to the thief; when he tried, he got nowhere. He succeeded only when he stopped deliberating and let an impression he could not justify guide him. The book will return to this tension often: the unconscious is frequently right for reasons we cannot state, and frequently wrong for reasons we cannot see, and the art is in knowing which situation you are in.
"These subliminal aspects of everything that happens to us may seem to play very little part in our daily lives. But they are the almost invisible roots of our conscious thoughts."
"Subliminal" comes from the Latin for "below the threshold." Jung, Freud and others speculated about this region for a century, but their tools (introspection, observing behavior, studying brain damage) could not look inside. Mlodinow's frame is a physicist's: just as quantum mechanics showed that the familiar world of everyday physics rests on a strange hidden layer, the new science of mind shows that everyday social life rests on an unseen layer of automatic processing.
The instrument that changed things is fMRI, which from the 1990s let researchers map the activity of brain regions while people perceive, decide and feel. He shows an image reconstruction by Jack Gallant's lab in which a computer, reading only brain activity, produces a rough picture of what a subject was looking at. The new field built on such methods is social neuroscience, and the book is a tour of its findings.
The Gallant-style reconstructions Mlodinow showed were blurry. Since 2023, groups pairing fMRI with diffusion image models and large language models have produced recognisable image reconstructions and continuous semantic decoders of language (for example Tang and Huth's 2023 decoder). These still need hours of per-person training data and the subject's cooperation, but the "mind reading" frontier has moved far past where the book left it.
He closes the prologue with Jung's point that dreams and myths express the heart as history expresses events, and promises a tour of the hidden forces beneath "willed, rational behavior," forces whose impact is much greater than most of us suppose.
F-C-B. Forced-choice beats introspected certainty. Consciousness is the press secretary, not the CEO. Brain imaging lets us watch the nonconscious vote before the conscious report. Recite: forced choice, confabulation, before.
Hold two nearly identical pens or mugs. Close your eyes. Have a friend hand you one, then the other, five times in random order. After each pair, guess which is heavier even when you feel you have no basis. Track hit rate. Most people beat chance on near-threshold differences they insist they cannot feel. That is Peirce's method without a laboratory.
Forced-choice is a measurement tool, not a decision ethic. Using "you must pick" to extract a preference from a tired customer, a panicked patient or a junior employee creates false certainty for the asker. The shadow of Peirce's watch is coercive choice architecture: the pick looks like knowledge when it is only a forced guess under pressure.
Depth: the forced-choice bet
Peirce's theft is the book's opening myth, but the science is the 1884 forced-choice weight experiment with Joseph Jastrow. Subjects judged which of two pressures was heavier even when they reported zero confidence, and hit rates stayed above chance (often cited near 60 percent versus 50). That design later became a backbone of psychophysics and of signal-detection work. Modern framing is less mystical: low d-prime with a strict criterion still yields above-chance forced choices. The nonconscious claim survives; the detective yarn is illustration, not evidence.
Historians note that Peirce's recovery of the watch rests partly on his own report decades later and on newspaper accounts that disagree on details. Keep the method; treat the crime story as colour.
The prologue's second move is historical lineage. Helmholtz's unconscious inference, Freud's dynamic unconscious, and the cognitive revolution's automaticity research are three different answers to the same question: what work is the mind doing that the owner cannot report? Mlodinow's bet, stated early, is that the third answer is the scientific one: computational, measurable, often benign, and far broader than repression.
Peirce had only hit rates. Today's labs have masking, priming, eye-tracking, TMS and fMRI. The bet is unchanged: if the subject cannot report a cue, yet behavior or neural activity tracks it, the processing is nonconscious. The prologue's promise is that the rest of the book will show that bet paying off for perception, memory, social judgment, emotion and self.
Charles Sanders Peirce: philosopher-scientist; watch anecdote and forced-choice experiment. Joseph Jastrow: student collaborator on the 1884 weight study. Carl Jung: epigraph source on subliminal roots of thought. Jack Gallant: fMRI image-reconstruction lab that Mlodinow uses as the modern instrument.
- Peirce's watch and the weight experiment show that people can perform above chance while feeling they are guessing.
- Mlodinow frames the unconscious as a hidden layer, like quantum physics under the everyday world.
- fMRI and social neuroscience are what made the "new unconscious" testable.
1. The New Unconscious
Chapters 1 to 4 cover perception, memory and social machinery as automatic systems that consciousness only partly sees.
| Story or study | What it shows |
|---|---|
| Miss Dinnerman the tortoise | We read intention into automatic behavior |
| Mother's Thursday call | Implicit frames feel like conclusions |
| Surname marriages | Implicit egotism |
| Stale popcorn, fonts | Context and fluency shape intake and effort |
| Rangel's wine, Pepsi paradox | Expectation changes experience |
| Tickers, sunshine | Unconscious effects show up in markets |
The chapter opens at home. At eighty-five, Mlodinow's mother inherits his son's Russian tortoise, Miss Dinnerman, and becomes convinced the animal is clever and purposeful because it keeps trying to escape its pen. Mlodinow's point: we read conscious intention into behavior that is largely automatic, in tortoises, in fruit flies, and especially in ourselves.
Habit that feels like reasoning
The real material is his mother's mind. In graduate school he phoned her every Thursday at eight. One Thursday he was out on a date and did not call; she phoned the police, the hospitals, and his friends, sure he was dead. To her, a missed call meant catastrophe, a reading shaped by her past as a Holocaust survivor. Her interpretations were not consciously reasoned; they were automatic, like grammar. We all carry such implicit frames, milder ones with luck, and we experience their outputs as obvious conclusions.
The new unconscious versus Freud
Mlodinow draws a firm line. Freud's unconscious was "hot and wet," seething with lust and anger, repressed and irrational. The new unconscious described by modern research is "kinder and gentler" and "more reality bound." It is not a dungeon of forbidden wishes; it is the automatic processing system that perceives, remembers, evaluates and steers, mostly for our benefit, and mostly without our knowing.
Small cues, big effects
The chapter then piles up experiments in which irrelevant factors move behavior and people do not know it:
- Names and marriage. Among the five most common U.S. surnames, Smiths marry Smiths three to five times as often as they marry Johnsons, Williamses, Joneses or Browns. The driver is implicit egotism: we are drawn to what resembles us, even a name.
- Stale popcorn. Moviegoers given free popcorn, some fresh and some fourteen days stale, in medium or huge buckets, ate more from bigger buckets even when it was stale. Doubling container size raises snack consumption 30 to 45 percent. Nobody reported "deciding" to eat more.
- Labels and fonts. Food described with appealing names tastes better; a recipe printed in a hard-to-read font seems harder to cook. This is the fluency effect: ease of processing is misread as ease or quality of the thing itself.
Why we cannot report the causes
Mlodinow stresses that when the popcorn eaters or the font readers were asked why they behaved as they did, they gave sensible, sincere, wrong answers. Research suggests we combine low ability and high confidence in understanding our own feelings: you may be sure you took a job for its challenge when its prestige moved you more, or that you like a partner for her sense of humor when her resemblance to you mattered more. The new unconscious is not just hidden; it is covered by a convincing story. This theme, introduced here, returns in full in Chapter 9.
The practical implication is uncomfortable for anyone who relies on asking people what they want. Stated preferences describe the conscious theory; behavior reveals the whole system. Good researchers, marketers and designers learn to watch what people do in realistic settings rather than only what they say in interviews.
Economics meets the unconscious
Caltech's Antonio Rangel is the chapter's guide to neuroeconomics. People will pay 40 to 61 percent more for junk food when the actual item is in front of them rather than a picture or text. In a British supermarket, four French and four German wines were shelved together with music playing: on French-music days 77 percent of wine sold was French; on German-music days 73 percent was German. Asked why they chose, shoppers almost never mentioned the music.
Rangel had volunteers in a scanner taste wines labeled only by price. They rated the $90 wine better than one marked as costing far less, but the two were the same wine. The brain's pleasure region (medial orbitofrontal cortex) actually responded more to the "expensive" one. In the Pepsi paradox, Pepsi wins blind tests while Coke wins labeled ones; people with damage to the ventromedial prefrontal cortex lose the brand effect and simply prefer Pepsi. Expectation is folded into experience, not added afterward.
Markets are made of people
If these effects are real, they should show up in aggregate, and they do:
| Cue | Effect reported |
|---|---|
| Pronounceable ticker (e.g. KAR vs RDO) | IPOs with easy codes outperformed in the first day, week and year (NYSE 1990 to 2004) |
| Pronounceable company names | Fluent names priced higher shortly after listing |
| Sunshine in New York | Perfectly sunny year: about 24.8% NYSE return; fully overcast: about 8.7% (1927 to 1990 data) |
| Sunny day in a restaurant | Waiters told it was sunny received bigger tips |
Mlodinow, author of The Drunkard's Walk, warns about data mining: sift enough data and patterns appear by chance. He presents the stock results as suggestive, supported by many studies, not as trading strategies.
Robust: fluency effects on judgment, portion-size effects on intake, and the price-placebo effect on wine ratings (medial OFC). Confounded: implicit egotism in surname and name-letter marriage data (ethnic and regional clustering). Weak or retracted: several food-labeling studies from Brian Wansink's Cornell lab; Wansink resigned in 2018 after 18 retractions. The original popcorn study is not among them, but treat lab-of-origin results with care. The weather-and-markets effect shrinks or vanishes in some later samples. Behavioral social priming as a class: Mac Giolla et al. 2024 meta-analysis of 70 close replications found independent-team effects near zero (d ≈ 0.002). Keep defaults, habits and fluency; retire the party tricks.
Classical economics
More cues the chapter piles on
Beyond popcorn and wine, Mlodinow walks through a denser stack of context effects:
- Surname magnetism. Among the five most common U.S. surnames, Smiths marry Smiths at three to five times the rate they marry Johnsons, Williamses, Joneses or Browns. Later demographers flagged ethnic and regional clustering as confounds; the broader name-letter / implicit egotism literature survives in weaker form.
- Hard fonts, hard tasks. A recipe in a difficult typeface is judged harder to cook; fluency is misread as a property of the world.
- Presence beats picture. Rangel's subjects paid 40 to 61 percent more for junk food when the item sat in front of them rather than appearing as text or image.
- Music and nationality. French music days: 77 percent of wine sold was French; German music days: 73 percent German; shoppers almost never named the music.
- Tickers and sunshine. Pronounceable IPO tickers outperformed hard ones in early windows (NYSE 1990 to 2004); sunny NYSE years averaged about 24.8 percent return versus about 8.7 percent for fully overcast years in a long historical sample. Mlodinow, as author of The Drunkard's Walk, flags data-mining risk: treat market curiosities as suggestive, not as strategies.
The unifying claim is not that people are fools. It is that Choice = Preference + Context + Noise, and introspection only sees Preference. assumes choices follow from conscious, consistent preferences. The new unconscious adds context terms: price, music, font, container size, sunshine. A better model is Choice = Preference + Context + Noise, where the person sees only the first term.
Depth: three cue families and their 2026 status
Mlodinow's "new unconscious" is not a seething cauldron of repressed wishes. It is the set of fast, parallel, capacity-cheap processes that handle perception, motor control, categorization and much of social judgment. Bargh's formulation (and the broader social-cognition tradition) treats these processes as goal-dependent but often unintentional: once a goal is active, related behaviors can fire without the person intending the specific act.
The chapter's demonstrations cluster into three families with very different replication fates:
- Behavioral priming (elderly words → slower walking; achievement words → persistence). Most damaged by the replication crisis.
- Experiential and evaluative cues (warm cup → warmer personality ratings; scent → store dwell time). Mixed; some sensory-affect links hold, many social generalizations do not.
- Mindlessness and placebic reasons (Langer's copier). More robust in the small-request regime; the "because" particle itself carries compliance weight.
Large multi-lab projects (Many Labs 1-3, Registered Replication Reports) found little or no effect for several classic social-priming paradigms. A 2024 meta-analysis of 70 close social-priming replications (Mac Giolla et al.) found independent teams produced a near-zero average effect (d ≈ 0.002), while replications that included an original author still showed a moderate effect (d ≈ 0.40). Embodiment findings (heavy clipboard → seriousness) also shrank. The safer 2026 reading: context and goals clearly shape automatic behavior; specific party-trick primes from 1990s-2000s papers should not be treated as established facts. Habit, framing and default effects remain robust; cocktail-party semantic priming of walking speed does not.
Miss Dinnerman: the tortoise that opens the intention-attribution joke. Antonio Rangel: Caltech neuroeconomist; wine-price and junk-food presence studies. Brian Wansink: Cornell food-lab source of several portion-size findings; resigned 2018 after mass retractions (original popcorn study not among them, but lab-of-origin caution applies).
CUE → CHANGE → CLAIM. A cue (word, smell, posture) you do not notice can change speed, liking or choice; the conscious mind then claims a tidy reason. Recite after every priming anecdote: What was the cue? What changed? What story did they invent?
Ask a colleague for a small, low-stakes favor with no reason ("Can I jump the printer queue?"). Later, ask another with a placebic reason ("Can I jump because I need to print?"). Note compliance. Langer's classic found a large jump for placebic "because" on small requests. Do not use this to manipulate; use it once to feel how thin the conscious filter is.
The same chapter that celebrates automaticity also describes the toolkit of dark patterns: scarcity banners, default ticks, warm lighting in showrooms, "because" copy in checkout flows. The dark side is when product teams treat every cue as a lever and every user as a subject. Ethical use requires disclosure, reversibility and a refusal to exploit cues the user cannot see.
- We over-attribute behavior to conscious intent, in tortoises and in ourselves.
- The new unconscious is adaptive automatic processing, not Freud's cauldron.
- Irrelevant cues (names, fonts, music, price, bucket size) change choices and even experience.
- People cannot report these influences, and invent reasons instead.
2. Senses Plus Mind Equals Reality
| Story or study | What it shows |
|---|---|
| Weber, Fechner, Wundt, Carpenter | Mind became measurable; two trains of action |
| Bandwidth numbers | Conscious channel is tiny |
| Patient TN | Blindsight guides faces and walking |
| Lt. Col. T. (1917) | Vision is modular |
| Binocular rivalry | Unseen images steer attention |
| Phonemic restoration | Hearing fills gaps from context |
From Kant to the two trains
The chapter traces the history of the idea. Kant doubted psychology could become a science because mental events cannot be weighed. Nineteenth-century Germans tried anyway: Ernst Weber and Gustav Fechner found that the smallest detectable change in a stimulus scales with its size (detect one gram against a reference; against a reference ten times heavier you need ten grams). Wilhelm Wundt founded the first psychology lab in Leipzig in 1879, without official funding. The British physiologist William Carpenter, in Principles of Mental Physiology (1874), described "two distinct trains of mental action," one conscious and one automatic. Peirce, the child prodigy who wrote a history of chemistry at eleven, supplied the first experimental evidence.
Today, Mlodinow says, Carpenter's two trains are two entire railway systems, each with many lines, constantly interacting. The unconscious system is more fundamental: it evolved first, for survival, and it handles the bulk of the work.
- Sensory input: about 11 million bits per second.
- Conscious processing: roughly 16 to 50 bits per second.
- Some estimates: we are conscious of about 5 percent of cognitive function.
- Deep concentration raises brain energy use by only about 1 percent over rest: most of the work is always on.
Language shows the system at work. "The cooking teacher said the children made good snacks" and "The cannibal said the children made good snacks" use the same words, but you resolve "made" instantly and differently, with no sense of having chosen.
Patient TN and blindsight
The chapter's centerpiece is TN, a fifty-two-year-old African doctor who suffered two strokes that destroyed the visual cortex in both occipital lobes. His eyes worked; he was blind. Researchers showed him circles and squares on a laptop and asked him to guess: he was at chance. Then they showed faces, angry or happy, and asked him to guess the expression. He scored well above chance, and an fMRI showed his amygdala responding to emotional faces he could not see.
Then a second team set up an obstacle course in a corridor. TN refused at first, then, with an escort promised to trail him, agreed, and walked down the hallway zigzagging around every box and chair without touching one, while insisting he could see nothing. This is blindsight: visual information reaching behavior through older, subcortical pathways (via the superior colliculus and thalamus) that bypass conscious vision.
Lieutenant Colonel T. and binocular rivalry
The first hint of multiple visual pathways came in 1917 from British army doctor George Riddoch. World War I bullets were slow enough to bore neat tunnels through brains, and 25 percent of penetrating wounds to British soldiers were head wounds, many in the occipital lobe. Lieutenant Colonel T., shot through the right occipital lobe, could not see static objects on his left (he missed the meat on the left of his plate) but could see motion there. Vision, it turned out, is assembled from separate modules.
Because blindsight patients are rare, Christof Koch found a way to create unconscious vision in healthy people in 2005, using binocular rivalry (continuous flash suppression): flash rapidly changing colorful mosaics to one eye and the other eye's image never reaches awareness. In one study, erotic images shown to the suppressed eye pulled attention toward their side: heterosexual men toward naked women, heterosexual women toward naked men, gay men toward naked men, even though everyone reported seeing only mosaics.
Mlodinow adds a personal story about trusting signals you cannot fully decode. At twenty, hiking on the Golan Heights just after the 1973 war, he walked into a farmer's field to look at a bird. A man on a tractor began yelling and gesturing; through broken Hebrew Mlodinow grasped that the field was mined. He says the episode cured him of distrusting instincts he could not justify logically: we are all a bit like TN, advised by an unconscious that sees more than we do.
All perception is illusion, usefully
The eyes deliver poor data. There is a blind spot where the optic nerve attaches, the eyes jiggle in microsaccades and move about 100,000 times a day, and peripheral vision is crude. The brain edits out the motion, fills in the blind spot, and presents a stable, sharp scene that is partly invented. Hearing works the same way.
Researchers recorded "The state governors met with their respective legislatures convening in the capital city," cut out 120 milliseconds (the first "s" in legislatures) and replaced it with a cough. Of twenty listeners, nineteen said nothing was missing, and the one who noticed named the wrong sound. Even when the whole syllable "gis" was obliterated, nobody could identify the gap. Context words that come later in the sentence can determine what you "heard" earlier. Hearing is a best guess, assembled and then presented as fact.
The chapter ends with illusions of shape and color, and with the rule they all teach: the unconscious interprets and enhances, and we experience its interpretation as the world. In shading illusions, two squares of identical gray look strikingly different because the brain corrects for an apparent shadow; you cannot stop seeing the difference even after you measure it. In contour illusions, the brain draws edges that are not on the page because a hidden shape is the most likely explanation of the fragments it sees. Knowledge does not switch the illusion off, which is the clearest sign that the processing happens below the level where knowledge operates.
Mlodinow's summary formula gives the chapter its title: senses plus mind equals reality. The senses supply sparse, noisy data; the mind supplies context, expectation and inference; what we call reality is the product. Most of the time the product is good enough to act on, and that is all evolution required.
Illusions that will not switch off
The chapter closes with shape and color illusions that survive measurement. Two squares of identical gray look different under an apparent shadow; you cannot unsee the difference after you measure it. Contour illusions make the brain draw edges that are not on the page because a hidden shape is the most likely explanation of the fragments. These are not party tricks bolted onto the chapter. They are the phenomenology of fill-in: knowledge sits too high in the stack to veto the construct.
Mlodinow's Golan Heights story (age twenty, post-1973, a mined field he almost walked into) is his personal vote for respecting unjustified alarms, then checking them. The unconscious is fast and often right about danger; it is not a court of appeal.
Blindsight is well established, though researchers still debate whether patients have degraded conscious vision rather than none. Phonemic restoration and filling-in are textbook. The specific erotic-image attention study (Jiang et al., 2006) is one of the more cited CFS results, but later work showed CFS can let some image information leak into partial awareness, so "fully unconscious" claims are now made more carefully. The 11 million bits figure is an engineering estimate, not a measurement, and the "5 percent conscious" figure has no firm basis.
Depth: construction as engineering, not mysticism
The bandwidth gap is the chapter's engineering claim: about 11 million bits/s of sensory inflow versus roughly 16 to 50 bits/s of conscious throughput. Those numbers are order-of-magnitude estimates from sensory physiology and from classic reading/listening rates, not a single experiment. What they buy is a design constraint: almost everything the mind does with sensory data must be automatic compression.
Blindsight (Weiskrantz; TN cases) is the existence proof that some of that compression reaches action without reaching report. The cortical route (retina → LGN → V1) supports conscious seeing; the older route (retina → superior colliculus → pulvinar → amygdala/motor) can still steer. Riddoch's 1917 motion-without-form cases showed the same modularity from war wounds. Continuous flash suppression (CFS) lets labs study similar dissociations in healthy subjects, with the caveat that later work found CFS can leak partial awareness, so "fully unconscious" claims are made more carefully than in 2006.
Phonemic restoration (Warren) and visual fill-in are everyday versions of the same rule: the product presented to consciousness is a best guess, not a raw feed. Knowledge does not switch illusions off; that is the clearest sign the processing sits below the level where beliefs operate.
TN: African physician with bilateral occipital strokes; obstacle-course blindsight. George Riddoch: WWI doctor; motion vision without form. Christof Koch: CFS / rivalry methods for unconscious vision in healthy subjects. Weber, Fechner, Wundt, Carpenter: nineteenth-century founders of measurable mind and "two trains."
SPARSE IN, RICH OUT. The senses deliver sparse, noisy bits; the mind fills, edits and presents a stable world. Recite: bandwidth gap → fill-in → useful illusion. Blindsight is the proof case: behavior without the report.
Cover one eye and find your blind spot: fixate a small mark on paper and move a second mark around until it vanishes (about 15 degrees temporal). Then note that you do not see a hole in the world. That fill-in is the chapter in one minute. Next, listen to a noisy phone call and notice how often you "hear" words that were never said; phonemic restoration works in ordinary life.
"Trust your unconscious vision" becomes dangerous when the fill-in is biased (stereotype, hope, fear). TN walked around boxes; eyewitnesses "see" the face they expected. Shadow: treating every hunch as blindsight-grade data, or dismissing every hunch as noise, both invert the chapter. The rule is respect, then check.
- The mind has two railway systems; the unconscious one is older and bigger.
- Blindsight proves vision can guide behavior with no awareness.
- Perception is constructed: blind spots, saccades and gaps in sound are filled in.
- Respect hunches, then check them; the unconscious is fast, not infallible.
3. Remembering and Forgetting
| Story or study | What it shows |
|---|---|
| Thompson and Cotton | Confident eyewitness, wrong man |
| John Dean and the tapes | Right gist, invented scenes |
| Münsterberg, von Liszt | Errors of 26 to 80 percent |
| Shereshevsky | Perfect memory blocks meaning |
| Sweet test, Bartlett | Memory regenerates themes |
| Penny, Simons films, balloon photo | We store little and accept plants |
The wrong man
In July 1984, in Burlington, North Carolina, a man broke into the apartment of Jennifer Thompson, a twenty-two-year-old Elon student, and raped her. Throughout the attack she deliberately studied his face so she could identify him. She escaped, picked Ronald Cotton from a photo array and then a live lineup, and testified with total confidence. Cotton was convicted. In prison he met Bobby Poole, who resembled him and reportedly bragged about the crime. At a retrial Thompson looked at both men from fifteen feet away and again pointed at Cotton. Eleven years after the crime, a DNA test on preserved evidence exonerated Cotton and matched Poole. Thompson's memory, formed under the best possible intentions, had been rewritten.
- About 75,000 police lineups a year in the U.S.; in 20 to 25 percent, witnesses pick a person police know is innocent (a filler).
- In mock-crime studies where the culprit is absent, more than half of witnesses still pick someone.
- Of hundreds of Innocence Project DNA exonerations, about 75 percent involved mistaken eyewitness identification.
John Dean's "tape-recorder" memory
Watergate offered a rare ground truth. White House counsel John Dean testified in 1973 with such detail that he was called "the human tape recorder." Then the Nixon tapes surfaced. The psychologist Ulric Neisser compared them. Of Dean's account of a 15 September 1972 Oval Office meeting, "hardly a word" matched: Nixon had not said the things Dean attributed to him. Yet Dean was right about the gist: there was a cover-up, and Nixon was part of it. He remembered the meaning and invented the scene.
Münsterberg and the staged quarrel
The first scientist to grasp this was Hugo Münsterberg, a Harvard psychologist recruited by William James. In 1907 his house was burgled and he gave police sworn testimony, then discovered that several details he had sworn to were false; he had absorbed things he overheard from police. He collected cases, including Professor Franz von Liszt's staged seminar quarrel in Berlin, where a student shouted, another defended the speaker, and a revolver appeared. Divided into fourteen components, the witnesses' reports had error rates from 26 to 80 percent. His best seller On the Witness Stand proposed that memory keeps gist, loses detail, fills gaps with expectation, and that we believe the result.
Mlodinow's summary of the reconstructive view: (1) the brain stores the general meaning of events well and details poorly; (2) at recall, missing details are filled in from expectations, later information and imagination; (3) the reconstruction feels like a memory and earns full confidence. Confidence measures vividness, not accuracy.
Why forgetting is a feature
The Russian journalist Solomon Shereshevsky, studied for thirty years by Alexander Luria, forgot almost nothing: he could recall long lists and tables years later. But he struggled to recognize faces (a face looked different with each expression) and to grasp the point of what people said, because exact words crowded out meaning. A memory that kept everything would lose the forest for the trees. Discarding detail is how we extract patterns.
Mlodinow asks readers to study a list: candy, sour, sugar, bitter, good, taste, tooth, nice, honey, soda, chocolate, heart, cake, eat, pie. Later: was "taste" on it? "point"? "sweet"? Most people correctly reject "point," accept "taste," and many also confidently "remember" sweet, which was never shown. In his own audiences about half claimed it. This is the Deese-Roediger-McDermott (DRM) paradigm: the brain stores the theme and regenerates a likely member.
Bartlett and the War of the Ghosts
Frederic Bartlett at Cambridge had subjects read a Native American folk tale, The War of the Ghosts, and retell it over weeks and months. The story grew shorter, more conventional and more English: canoes became boats, strange supernatural elements dropped out, puzzling events were given explanations. Memory, like culture passed along a game of telephone, smooths toward the familiar. Gestalt psychologists saw the same smoothing in memory for geometric shapes.
How little we notice
- The penny test. Nickerson and Adams asked Americans to pick the real penny among fifteen drawings. Most failed; only the top 5 percent could reproduce all eight key features of a coin seen thousands of times.
- Change blindness. In Daniel Simons's films, an actor at a desk walks off camera and a different actor continues the scene. About two-thirds of viewers did not notice the person had changed. Details we do not attend to are never stored.
Memories of things that never happened
Researchers obtained family photos and doctored one to show the subject as a child in a hot air balloon. After a few interviews, half of twenty subjects recalled the ride, some with details. Across such studies, 15 to 50 percent of people accept planted memories. People have been convinced they were lost in a mall, saved by a lifeguard, attacked by an animal; after a fake Disneyland ad featuring Bugs Bunny (a Warner Bros. character who could not have been there), many "remembered" meeting him, 62 percent recalled shaking his hand, and one remembered him holding a carrot. False memories have been induced even in infants, gorillas, pigeons and rats.
Mlodinow closes with tenderness: of his father, dead twenty years, he keeps only scraps, and his children will forget most of his hugs. That is not tragedy but design. The unconscious keeps what matters and lets the rest go; the danger is only that we mistake a reconstruction for a recording.
Planted pasts and the DRM engine
The balloon-photo studies (about half of twenty subjects "remembered" a childhood ride) and the Bugs Bunny at Disneyland ads (62 percent recalled shaking his hand in one report) show how photos and suggestions become autobiographical. False memories have been induced even in infants, gorillas, pigeons and rats. The DRM sweet lure is the lab-clean version of the same engine: store the theme, regenerate a likely member, award confidence.
Shereshevsky is the opposite warning. Perfect retention without gist extraction left him poor at faces and meaning. Forgetting detail is how intelligence finds patterns; the legal system just happens to need the details evolution did not prioritize.
The reconstructive model is among the best-replicated findings in psychology. Two refinements since 2012: (1) John Wixted and colleagues showed that a witness's initial, uncontaminated confidence in a fair, double-blind lineup is actually fairly diagnostic; the danger is confidence that grows after feedback, as Thompson's did. (2) The 2014 National Academy of Sciences report Identifying the Culprit led many U.S. jurisdictions to adopt double-blind, sequential or recorded-confidence lineups. The Bugs Bunny figure varies by study (later analyses argue some "memories" were acceptance rather than recollection).
Depth: the legal and lab stakes of reconstructive memory
Thompson-Cotton is not an anecdote about a careless witness. Thompson studied the attacker's face on purpose, picked Cotton from photos and lineup, and reaffirmed the ID at retrial against Poole. DNA still cleared Cotton. The lesson Mlodinow needs is harsh: sincere, effortful encoding does not freeze a face. Feedback, repeated retrieval and lineup procedures can inflate confidence while accuracy stays flat or falls.
Neisser's reading of John Dean adds the complementary rule: gist can be right while scenes are invented. Dean was correct that Nixon was in a cover-up; almost no quoted line matched the tapes. Münsterberg's 1907 program (On the Witness Stand), Bartlett's War of the Ghosts, the DRM sweet lure, Simons change-blindness films, and balloon-photo false memories are one research program: memory is a regenerative system optimized for meaning, not for courtroom detail.
Two refinements since 2012 are now mainstream. (1) John Wixted and colleagues showed that a witness's initial, uncontaminated confidence in a fair, double-blind lineup is fairly diagnostic; the danger is confidence that grows after feedback, as Thompson's did. (2) The 2014 National Academy of Sciences report Identifying the Culprit pushed many U.S. jurisdictions toward double-blind, sequential or recorded-confidence lineups. DRM, change blindness and photo-based false memories remain among psychology's most replicated findings.
Jennifer Thompson / Ronald Cotton / Bobby Poole: the rape-ID case that DNA reversed. John Dean / Ulric Neisser: Watergate testimony vs tapes. Hugo Münsterberg / Franz von Liszt: early staged-error demos. Solomon Shereshevsky / Alexander Luria: perfect memory, poor gist. Frederic Bartlett: War of the Ghosts. Elizabeth Loftus tradition: plantable memories (balloon, Bugs Bunny).
GIST → FILL → BELIEVE. Memory stores meaning, rebuilds detail, then awards confidence. Recite after every vivid recollection: gist first, fill next, belief last. Confidence measures vividness, not accuracy.
Do the DRM sweet test properly. Study for thirty seconds: candy, sour, sugar, bitter, good, taste, tooth, nice, honey, soda, chocolate, heart, cake, eat, pie. Close the list. Wait two minutes. Then ask: was "taste" on it? "point"? "sweet"? Most people correctly reject "point," accept "taste," and many also "remember" sweet, which was never shown. That regenerating theme is the chapter's laboratory.
"Memory is reconstructive" can become a license to dismiss every testimony, or a license to rewrite your own past ("I never said that"). Shadow: using reconstructiveness to gaslight a witness or to erase accountability. The chapter's ethics cut both ways: treat memory as evidence with known error rates, not as either tape or fiction.
- Memory stores gist and rebuilds detail, then presents it with confidence.
- Eyewitness error is common and drives wrongful convictions.
- Forgetting detail is adaptive (Shereshevsky shows the cost of keeping all).
- False memories can be planted with photos and suggestion.
4. The Importance of Being Social
| Story or study | What it shows |
|---|---|
| Helper and hinderer | Infants judge character |
| Schachter's shocks | Fear drives affiliation (63% vs 33%) |
| Cyberball, Tylenol | Rejection registers as distress |
| ToM orders, Dunbar's 150 | Brains scaled for social tracking |
| Ewes, voles, oxytocin | Bonding has chemistry |
| Langer's copier | Scripts run small exchanges |
Mlodinow comes home after a frustrating day co-writing with Stephen Hawking and drops in on his mother, who reads his mood in an instant. Hawking, almost entirely paralyzed, could not show expressions, and the contrast makes the chapter's point: reading other minds is something we do constantly, effortlessly and mostly unconsciously, and losing its channels is disabling.
Born social
Six-month-old infants watched a wooden shape with eyes trying to climb a hill, while a "helper" shape pushed it up and a "hinderer" pushed it down. Offered the toys afterward, babies reached for the helper. Before language, we judge character.
In Stanley Schachter's 1950s study, about thirty University of Minnesota women were told to expect shocks that were either painful or "tickly." Asked whether to wait alone or with others, 63 percent of the anxious group chose company versus 33 percent of the calm group. Fear drives us together.
Rejection hurts, literally
Social exclusion activates the dorsal anterior cingulate cortex, a region also involved in the distress of physical pain. In the Cyberball game, players tossed a virtual ball until the other two "players" stopped throwing to them, and the brain lit up like an injury. Then the strange twist: people who took acetaminophen (Tylenol) for three weeks reported less social pain day to day, and in the scanner their exclusion response was reduced.
Loneliness is a health risk
Mlodinow argues that social pain is not only a feeling: lack of connection damages the body. In the Alameda County study (Lisa Berkman and Leonard Syme, published 1979), researchers scored 4,775 adults on a "social network index" built from marriage, contact with friends and family, and group memberships, then tracked them for nine years. After adjusting for health at the start, smoking, drinking, obesity, exercise and income, the most isolated people died at roughly twice the rate of the best connected. Isolation, Mlodinow writes, rivals smoking, high blood pressure, obesity and inactivity as a risk factor.
This is one of the chapter's strongest claims. A 2010 meta-analysis (Holt-Lunstad, Smith and Layton) pooled 148 studies and about 308,000 people and found stronger social ties linked to about 50 percent higher odds of survival over the follow-up periods. In 2023 the US Surgeon General issued an advisory calling loneliness an epidemic, and the WHO set up a Commission on Social Connection (2023 to 2026). The caveat: nearly all the evidence is observational, so some of the link runs the other way (sick people lose contact) or through shared causes such as poverty.
Why we have big brains
The social brain hypothesis says intelligence evolved for managing relationships more than for tool use. The key skill is theory of mind (ToM): modeling what others believe and want. ToM is graded by orders of intentionality: I believe (first), I believe you think (second), and so on. Chimps reach between first and second order: a chimp may think I believe George wants my banana, but not I believe George thinks that I want his banana. Most adults handle third and fourth order, enough to write or follow a novel, and some reach sixth. Mlodinow says climbing those levels feels to him like doing theoretical physics.
ToM develops early: by age four nearly all children can assess other people's mental states. When it fails, as in autism, the cost is social. Oliver Sacks, in An Anthropologist on Mars, described Temple Grandin watching playground children respond to signals she could not perceive: "something swift, subtle, constantly changing: an exchange of meanings, a negotiation, a swiftness of understanding so remarkable that sometimes she wondered if they were all telepathic."
A newly hired games-company executive Mlodinow knew ("call her Alice") suspected an outside programming contractor of financial improprieties but had no proof, and its airtight contract demanded $500,000 for early termination. Alice knew that the contractor's CEO, Bob, knew she was new and afraid of a misstep (third order), and that Bob knew she knew he was not afraid of a fight (fourth order). Would such a cautious person make a grand claim she could not back up? Reasoning that Bob would think not, she bluffed that she had proof. Bob released her employer for a small fraction of the contractual sum. ToM is not only for novelists; it is a negotiating tool.
- Across primates, neocortex ratio predicts typical group size.
- Extrapolated to humans, the prediction is about 150, matching clans, villages, military companies and Christmas card lists (Robin Dunbar).
- In Stanley Milgram's 1960s chain-letter study, of about 300 starters in Nebraska and Boston, 64 chains reached a stockbroker in Sharon, Massachusetts, with a median of about six links: "six degrees of separation."
Culture as a group project
Mlodinow ties big social brains to a burst of behavior in the archaeological record. Before about 50,000 years ago, humans rarely fished or hunted dangerous large game; then, within a few millennia, European sites show fishing, bird catching, mammoth hunting, shelters, symbolic art and complex burials. He reads this as the moment cooperative social intelligence took off. The payoff today is organization far beyond 150: one Large Hadron Collider experiment alone involved more than 2,500 scientists, engineers and technicians in 37 countries. Formal structure and communication technology let us scale past the limits of the social brain.
The idea of a sudden Upper Paleolithic revolution comes from European sites, which were the best studied. Since McBrearty and Brooks's 2000 review "The revolution that wasn't," most archaeologists see gradual accumulation in Africa: shell beads and engraved ochre at Blombos Cave date to about 75,000 to 100,000 years ago, and pigment use is older still. The social-brain argument survives; the "sudden switch" story does not.
The chemistry of bonding
Much social behavior is shared with other mammals and partly chemically programmed. A ewe rejects lambs not her own, yet injecting oxytocin into her brain makes her accept them. Prairie voles are monogamous and have dense oxytocin and vasopressin receptors in a reward region; meadow voles are promiscuous and lack them; adding the receptors with a gene makes meadow voles pair-bond. In humans, oxytocin rises with positive contact and was reported to increase trust in investment games. A variant of the vasopressin receptor gene (AVPR1A) was linked to men's marital bonding: men carrying two copies of one allele reported more marital crisis and were less likely to be married.
Mlodinow adds a broader mammalian pattern from the bonding literature: in most mammals males compete for mates and form hierarchical, aggressive coalitions, while females invest in few offspring and form strong bonds with infants and with each other. Because nonhuman mammals have a small prefrontal cortex relative to their automatic machinery, they "react more and think less," which makes them useful windows on the unconscious parts of our own social behavior. His example: your unconscious might raise an alarm when Uncle Matt stabs his arm with a shish kebab skewer, until your conscious mind recalls that Uncle Matt likes shocking magic tricks. A pet rabbit would simply flee; it "just can't take a joke," yet the brain regions doing its unconscious processing are not that different from ours. He is careful to say this is a tendency across species, not a script for human men and women.
From behaviorism to automatic scripts
The chapter retells psychology's swing: Freud popularized the unconscious, behaviorism banished the mind, and the cognitive revolution brought it back while staying wary of the unconscious. Then studies of scripts showed how much social behavior runs on autopilot. Ellen Langer's copier study: asking to cut in line "because I have to make copies," a reason containing no information, worked almost as well as a real reason, because the word "because" triggered the script. Researchers concluded that following preset scripts "may indeed be the most common mode of social interaction."
Seeing the brain at work
The idea is older than the machine. In The Principles of Psychology (1890), William James cited the Italian physiologist Angelo Mosso, who watched the brain pulse through gaps in the skulls of surgical patients and saw some regions pulse harder during mental activity, which he correctly attributed to neural activity. In 1899 Cambridge scientists linked brain circulation and metabolism by working directly on the exposed brains of dogs, cats and rabbits, not a method for people. Scripts research, meanwhile, drew on Frederic Bartlett's memory work: the distortions in recall he observed suggested unconscious scripts that fill gaps to make information fit how we think the world works.
Finally Mlodinow explains fMRI: it detects changes in blood oxygenation, a proxy for neural activity, letting researchers slice the brain in any plane. He describes the brain's three crude zones (brain stem and "reptilian" core, limbic system, neocortex) and notes the frontal lobe's prefrontal cortex, home of planning and self-control, which grew largest in our lineage. The dreams of Wundt and James, a science that sees the mind work, are finally being met.
| Claim | Original evidence | Status |
|---|---|---|
| Infants prefer helpers | Hamlin, Wynn and Bloom 2007 | Failed in ManyBabies 4 (2024-25: 567 infants, 37 labs, 49.34% chose helper, chance level) |
| Fear drives affiliation | Schachter 1959 | Consistent with later work |
| Isolation shortens life | Berkman and Syme 1979 | Strong (148-study meta-analysis) |
| Social pain = physical pain | Eisenberger 2003 (Cyberball, fMRI) | Overlap reinterpreted as general salience |
| Tylenol eases social pain | DeWall et al. 2010 | Small, inconsistent |
| Oxytocin raises trust | Kosfeld et al. 2005 | Failed to replicate in larger samples |
| AVPR1A and marriage | Walum et al. 2008 | Weak, not confirmed |
| Dunbar's 150 | Primate neocortex regression | Disputed (2021 reanalysis: wide intervals) |
| "Because" script | Langer, Blank and Chanowitz 1978 | Classic, rarely replicated directly |
Failed: infant helper/hinderer preferences did not replicate in the preregistered ManyBabies 4 study (published 2025; senior author Kiley Hamlin, who ran the original): 567 infants, 37 labs, 49.34 percent chose the helper (chance). Robust: affiliation under stress, social brain correlations, Milgram-style small-world structure (Facebook-scale about 3.5 degrees in 2016). Contested: the social pain = physical pain overlap (dACC often tracks general salience); acetaminophen effects are small and inconsistent; intranasal oxytocin and trust (Kosfeld 2005) failed in larger work including Declerck et al. 2020, and pooled human oxytocin-trust reviews through 2026 remain weak; the AVPR1A "monogamy gene" finding did not replicate well. Dunbar's 150 was challenged in 2021 by a reanalysis giving confidence intervals from about 4 to 520. The triune brain is rejected by modern neuroanatomy, as Mlodinow half-concedes.
Why the social machinery runs below awareness
Mlodinow's larger claim in this chapter is that sociality is old, so much of it lives in the older, automatic layers of the brain. Bonding chemistry is shared with sheep and voles; mate guarding, dominance and grooming have mammalian roots; even in humans the copier study shows social exchanges running on scripts that never reach deliberation. In Langer's study the request with a real reason ("because I'm in a rush") got about 94 percent compliance, the empty reason ("because I have to make copies") about 93 percent, and no reason about 60 percent, at least for a small request. When the favor was large (twenty pages), the empty reason lost its power, which shows when the script yields to thought: low stakes run on autopilot, high stakes wake the conscious mind.
That pattern, cheap automatic defaults with expensive conscious override, is the template for the rest of the book. It also explains why the social unconscious is so hard to notice: it works well enough, most of the time, that we never have to check it.
Depth: what survives when the charming studies fall
Chapter 4 is the book's highest-variance science chapter. The broad claim (humans are compulsively social; much of that machinery is automatic) is solid. Several marquee exhibits are not.
ManyBabies 4 (published 2025; 567 infants, 37 labs; senior author Kiley Hamlin of the original helper/hinderer work) found infants chose the helper at 49.34 percent, chance. The 2007 Hamlin, Wynn and Bloom preference does not clear a large preregistered bar. Keep the broader point that infants are socially attentive; drop the specific "babies vote for helpers" party piece.
Oxytocin and trust. Kosfeld et al. 2005 (intranasal oxytocin raises trust in an investment game) failed in larger samples. Declerck et al. 2020 and later pooled analyses (including Cortex 2026-era reviews) treat the human intranasal-trust literature as weak. Vole pair-bonding receptor biology remains interesting; the leap to human nasal-spray trust does not.
Dunbar's 150. The primate neocortex-ratio regression is a useful heuristic for team design. A 2021 reanalysis reported confidence intervals from about 4 to 520. Treat 150 as a planning number, not a constant of nature.
What holds: Schachter affiliation under fear; Berkman-Syme / Holt-Lunstad loneliness-mortality link (with the usual observational caveats); ToM as a graded skill; Langer-style scripts at low stakes; small-world network structure (Facebook-scale ~3.5 degrees).
AFFILIATE → MODEL → BOND. Under threat we affiliate; to scale we model other minds (ToM); chemistry and scripts bond and automate the rest. Recite: fear→company, ToM→150, scripts→because. Keep ManyBabies 4 in the pocket: infancy claims need large preregistered tests.
For one day, count how many times you use a placebic "because" ("because I need to," "because that's how we do it") and how many times others use one on you. Then, for one negotiation or email thread, write the other party's second- and third-order beliefs (they think that I think…). That is ToM as a practical checklist, not a parlor trick.
Social-brain talk becomes essentialism ("women bond, men compete") or oxytocin hype (nasal spray as trust drug). Shadow: selling bonding chemistry as a managerial fix, or treating Dunbar's 150 as a hard org-chart law when the 2021 reanalysis gave intervals from about 4 to 520. Use the machinery; refuse the myth.
- Social judgment runs automatically, though the claim that it starts in early infancy failed a large 2025 replication.
- Exclusion registers as distress, and we seek others when afraid.
- Big brains may exist to run theory of mind and track about 150 relationships.
- Hormones and scripts steer social behavior below awareness, though many human hormone findings are weaker than once claimed.
5. Reading People
Chapters 5 to 10 cover reading others, judging covers, sorting, tribes, feelings and the defended self.
| Story or study | What it shows |
|---|---|
| Clever Hans | Unwitting cues |
| Maze-bright rats | Expectation changes outcomes |
| Pygmalion classroom | Teacher expectations and IQ |
| Darwin, Ekman, Fore | Expressions evolved and are shared |
| Partner approach study | Mate guarding in humans |
| Visual dominance ratio | Gaze tracks status |
Clever Hans
In 1904 the New York Times reported on a German horse, Clever Hans, who answered arithmetic questions by stamping his hoof, under the tutelage of a retired schoolteacher, Wilhelm von Osten. A commission found no trickery. The young psychologist Oskar Pfungst then noticed that Hans failed whenever the questioner did not know the answer or stood out of sight. Hans was reading tiny involuntary cues: a slight lean forward as he began tapping, a minute straightening of the head when he reached the right number. Pfungst proved it by playing Hans himself: of twenty-five people who questioned him, nearly all gave the same unwitting signals. Nobody was cheating. The questioners did not know they were signaling.
Alongside language we run a parallel nonverbal track: posture, gaze, expression, touch, voice tone. It is older than words, mostly automatic in both sending and receiving, and often decisive when the two tracks disagree.
Expectation leaks: Rosenthal's experiments
Robert Rosenthal told students that one group of rats was bred "maze-bright" and another "maze-dull." The rats were identical; the "bright" rats performed significantly better, because handlers' expectations changed how they treated them. In a human version, student experimenters read an identical script asking subjects to rate photos for success. Experimenters told to expect high ratings got high ratings. A follow-up that filmed the sessions found the effect worked even when only the experimenter's voice, or only their face, was available.
Rosenthal and Lenore Jacobson gave children in eighteen classrooms an IQ test and told teachers that certain pupils were "bloomers" about to spurt. The names were random. Eight months later, about half the ordinary children gained IQ points, but about 80 percent of the "bloomers" gained at least 10 points, and about 20 percent gained 30 or more, against 5 percent of other children. Teachers had given the labeled kids warmer attention and more feedback, without knowing it.
Darwin's expressions
Darwin studied expression in The Expression of the Emotions in Man and Animals (1872), arguing that our expressions evolved and are shared with other animals. He noted the difference between a polite smile and the genuine Duchenne smile, which involves the muscles around the eyes and is hard to fake. Paul Ekman later showed photos of expressions in literate countries and to the Fore people of Papua New Guinea, who had little contact with outsiders; they recognized happiness, fear, anger, disgust, sadness and surprise as well as anyone. Infants make nearly all the basic expressions from early on, and children born blind produce them too.
Why nonverbal signals evolved
Language is perhaps 50,000 to 100,000 years old in its modern form, but social life among our ancestors is far older, so a nonverbal system had to carry intention, threat, submission and affection long before words. That is why it is processed so quickly and so automatically, and why, when words and body disagree, listeners tend to trust the body. Darwin argued that many expressions began as useful actions (baring teeth to bite, widening eyes to see more) that became signals because others learned to read them. Some, like the smile, have parallels in other primates, where a bared-teeth grin signals appeasement rather than joy.
Space, dominance and the gaze
In a field study, experimenters approached women sitting with a male partner and asked for directions. Men moved closer and touched their partners more as the stranger approached, a human version of mate guarding, far subtler than the fights of hamadryas baboons.
Humans maintain two kinds of dominance: physical (fear) and social (admiration). Social dominance is signaled largely by gaze:
Mlodinow recalls his father in Buchenwald, under a hundred pounds, who learned that meeting a guard's eye could mean death and that a look could also, once, win him an extra crust. Gaze carries power.
The body speaks in motion
In point-light studies, actors wear small lights at the joints and are filmed in the dark. Standing still they look like random dots; moving, observers instantly recognize gender, mood and even emotions like anger or joy. Our detectors are tuned to human motion.
Nonverbal skill as social capital
Children who read nonverbal cues well are more popular; adults who read them well do better in business and are judged more competent. In the 1950s, researchers tried to build a "kinesics" dictionary of gestures like a grammar, and popular books still promise to decode crossed arms. Mlodinow is skeptical: cues are context dependent, and the real skill is unconscious integration, not lookup tables. The point is we send and receive far more than we intend.
Dogs, Method actors and the limits of kinesics
Mlodinow notes that dogs, not chimps, are often better at reading human social signals: domestication selected for companions who track our cues. Method actors try to feel the emotion so the face follows; politicians who can conjure genuine warmth outperform those who fake it. 1950s kinesics dictionaries (crossed arms means X) mostly failed because cues are context dependent. The real skill is unconscious integration across channels, not a lookup table. That is why Clever Hans fooled a commission: the cue was distributed, tiny and sincere.
Clever Hans remains the classic warning about experimenter cues, and the reason for double-blind designs. Pygmalion is contested: reanalyses (Thorndike, Snow, and later Jussim and Harber's 2005 review) found the IQ effects small, mostly in the youngest grades, and not consistently replicated; teacher expectations do matter, but less dramatically. Ekman's universality has been challenged since 2014 by Lisa Feldman Barrett and others, who find that recognition drops sharply when people label faces freely rather than pick from a list, and that the same expression means different things across cultures. The ROTC gaze data come from Dovidio and Ellyson's 1980s work and have held up broadly.
Depth: expectation as a causal channel
Clever Hans is the chapter's methodological ancestor. Pfungst's finding (Hans fails when the asker does not know the answer or stands out of sight) is why modern labs insist on double-blind designs. The horse was not calculating; the humans were leaking. Rosenthal's maze-bright rats and photo-rating experimenters showed the same leak in labs: expectation changes treatment, and treatment changes outcome, while everyone sincerely denies cueing.
Pygmalion in the classroom is the applied version. The honest 2026 reading (Jussim and Harber; later reviews) is: teacher-expectation effects are real but smaller than the original drama, strongest in early grades, and easy to overclaim. The mechanism still matters for organizations: a label ("high potential," "flight risk," "difficult customer") changes how people are treated, which then "confirms" the label.
Darwin → Ekman → Fore established a research program on shared basic expressions. Lisa Feldman Barrett's critique since about 2014 does not erase the data; it tightens the claim: forced-choice recognition looks universal; free labeling and cross-cultural work show more variance. Use expressions as probabilistic cues, not as a dictionary of truth.
Gaze (visual dominance ratio), mate-guarding tie-signs, and point-light walkers complete the channel map: status, affiliation and emotion ride on motion and eyes as much as on words.
Clever Hans / Wilhelm von Osten / Oskar Pfungst: the horse and the debunker. Robert Rosenthal / Lenore Jacobson: maze-bright rats and classroom bloomers. Darwin / Paul Ekman / Fore people: expression research program. Dovidio and Ellyson: visual dominance ratio.
TWO CHANNELS. Words say one thing; gaze, face, posture, voice say another. Clever Hans proves the second channel can run the show. Recite: Hans → Pygmalion → leak. Expectation leaks into treatment; treatment leaks into outcome.
Film yourself in a short mock interview (or watch a recorded meeting). Mute the audio and score warmth and competence from face and posture alone. Then play audio only and score again. Compare to the full track. Most people discover they are broadcasting a channel they never edited. That is Hans from the inside.
Reading people becomes surveillance theater: managers scoring microexpressions, interviewers over-weighting a handshake, "lie detection" apps sold as science. Shadow: treating the nonverbal channel as a truth oracle. Ekman-style cues are probabilistic; Pygmalion loops cut both ways (low expectations also leak). Use two channels; refuse certainty cosplay.
- Clever Hans read cues his questioners did not know they gave.
- Expectations leak through tone and face and can shape outcomes.
- Basic expressions are widely shared, though how universal is debated.
- Gaze, space and motion signal status and intent below awareness.
6. Judging People by Their Covers
| Story or study | What it shows |
|---|---|
| Cowbird and speaker | Responses to recordings |
| Nass's computers | Gender and politeness applied to machines |
| Deep voices | Attraction and dominance cues |
| Thatcher's voice | Voice can be trained |
| Touch studies | One second of touch shifts compliance |
| Kennedy-Nixon, Todorov | Looks predict votes |
Cowbirds and computers
A female cowbird will go into a mating display for a recording of a male's song from a loudspeaker. Surely humans are not so easily fooled? Clifford Nass at Stanford sat a couple hundred computer-savvy students in front of computers that tutored and evaluated them with recorded voices. Students rated a female-voiced tutor as knowing more about love and relationships and a male-voiced tutor as knowing more about computers, and found praise from a male-voiced computer more convincing. They were also polite to computers: when asked to rate a computer's performance on that same computer, they gave kinder reviews than when rating it on a different machine. Every student denied that a computer's voice or "feelings" mattered.
The deep voice
Women's preferences in faces vary, but they are surprisingly consistent about voices: they prefer lower-pitched male voices, and the preference is strongest when they are most fertile in their cycle. In one study men competed for a woman over a video link while hearing a male rival; men who believed themselves dominant lowered their pitch, and men who felt less dominant raised it. A deep voice is linked to testosterone, and in a study of the Hadza hunter-gatherers of Tanzania, men with deeper voices had more children.
Voice beyond content
Using filters that keep tone and remove words, researchers had speakers answer two questions (one about affirmative action in college admissions, one about happiness). Altering only pitch, pitch variation and speed changed how listeners rated speakers: faster speech, lower pitch and more modulation were judged more energetic, intelligent and knowledgeable. Margaret Thatcher famously took voice lessons to lower her pitch and sound more authoritative, coached by advisers including people from the National Theatre.
| Study | Without touch | With 1-second touch |
|---|---|---|
| French men asking women to dance or for a phone number | 10% success | 20% |
| Tips in restaurants (various studies) | lower | higher (e.g. 14.5% vs 17.5% in one) |
| Diners taking server's suggested special | ~40% | ~60% |
Touch also raised petition signing and helpfulness. A 2010 Berkeley study found NBA teams with more early-season fist bumps, high fives and hugs performed better later in the season.
Touch matters enough that we have a dedicated nerve route: slow C-tactile fibers respond to gentle stroking and feed emotional rather than discriminative brain regions. In primates, grooming maintains alliances; human touch is grooming's descendant.
Kennedy versus Nixon
On 26 September 1960, Kennedy arrived tanned and rested; Nixon, recently out of hospital, was pale, underweight and sweating, refused professional makeup, and wore a gray suit that blended into the backdrop. About seventy million people watched. The legend, supported by a Sindlinger poll, is that radio listeners thought Nixon won while TV viewers favored Kennedy. Mlodinow notes the poll was never published and had dubious methods, but the election was so close (fewer than 1 vote in 500) that appearance needed to sway only a sliver.
Faces decide elections
In a California study, volunteers read fictional campaign flyers with real policy information and photos. Swapping in a photo rated more competent-looking gave a candidate about 59 percent of the vote, a landslide by U.S. standards (only Lyndon Johnson's 61 percent in 1964 beat it since the Depression). In 2005, Alexander Todorov at Princeton showed people pairs of black-and-white headshots of winners and runners-up in Senate and House races. Judgments of competence, made in about a second, picked the winner in 72 percent of Senate races and 67 percent of House races; in a later forecast of 2006 races, 69 percent of governors and 72 percent of senators. Later work found exposures as short as a tenth of a second were enough.
Darwin himself was nearly kept off the Beagle because Captain FitzRoy, a believer in physiognomy, doubted that a man with his nose could have the energy for the voyage.
Behind the curtain
The chapter ends with the Wizard of Oz: a big face and booming voice, a small man behind the curtain. We meet most people only as personas, and our unconscious forms impressions of competence, warmth and trustworthiness from looks, voice and touch, impressions that often outweigh what people say. Knowing this, Mlodinow argues, we can at least ask whether we are judging the speaker or the speech.
Voice, touch and the cowbird parallel
The chapter opens with the cowbird: males do little parenting yet females choose on song and display. Humans, Mlodinow argues, still overweight displays that once tracked fitness. Lower male voice pitch correlates with testosterone, not with wisdom; Thatcher flattened her "dangerous to passing sparrows" high notes because politics is a cover contest. Light touch on the forearm moved specials from about 40 to 60 percent uptake in restaurant studies; bar tips and petition signatures move similarly. Nass's students insisted they would never apply social norms to a computer, then did: politeness, reciprocity, gender stereotypes toward a voice in a box.
The adult response is not to pretend covers are irrelevant. It is to route important decisions through structures that dilute covers: blind resume review, standardized scorecards, policy text before photos on ballots where feasible.
Robust: Todorov's face-competence effect replicated in many countries (though it predicts better for low-information races); voice-pitch preferences are well documented; the media equation (people treat computers socially) has held up and became central to human-computer interaction. Weaker: ovulation-shift effects on mate preferences largely failed to replicate in large preregistered studies (e.g. Jones et al. 2018); the touch-tipping effects are real but small; the NBA touch study is correlational. The Nixon radio story is now considered largely a myth; modern reanalyses find little evidence radio listeners judged differently.
Nass's findings now govern voice assistants and AI companions. Since 2024, real-time voice models (from OpenAI, Google and others) can modulate pitch, warmth and pace on demand, which means the cues Mlodinow describes can be engineered at scale. Regulators in the EU (AI Act Article 50 transparency duties, applying from 2 August 2026) require that people be told when they are speaking with an AI.
Depth: thin slices that still predict
Chapter 6 is where the social unconscious meets markets and ballots. Clifford Nass's computer-voice studies (the "media equation") showed people applying politeness and gender stereotypes to machines, then denying they would ever do so. Voice pitch tracks testosterone more than competence, yet voters and clients still treat pitch as a competence cue; Margaret Thatcher's voice coaching is the political case study. Light forearm touch raises compliance and tips in small, replicated field effects. Todorov's face-competence judgments (about a second, sometimes a tenth) predicted U.S. election winners at roughly two-thirds to three-quarters in low-information races and have replicated across countries.
The Kennedy-Nixon radio vs TV legend is now treated as mostly myth in modern reanalyses; keep the broader point that visual presentation swamps content when information is thin, and drop the tidy "radio listeners preferred Nixon" morality play.
Todorov's competence-from-faces effect remains one of the better-replicated findings in political psychology, strongest when voters know little else. Voice-pitch preferences are well documented. Ovulation-shift mate-preference effects largely failed large preregistered tests (e.g. Jones et al. 2018). Touch-tipping effects are real but small. Design implication: structured interviews, blind auditions and policy-first ballots are debiasing tools, not insults to intuition.
COVERS FIRST. Voice pitch, touch, face shape and polish hit before policy. Todorov: competence looks predict votes in about a second. Recite: cover → thin slice → correct for it.
Print two versions of a one-page bio for a fictional candidate (or product), identical text, different stock headshots rated high vs low on "competence" by a friend who has not read the text. Have five people who have not seen either pick a winner. Note the swing. Then reverse the photos for five more. That is Todorov's lesson at kitchen-table scale.
"Looks matter" becomes lookism as strategy: hire the jawline, elect the chin, reject the accent. Shadow: optimizing covers while starving substance, or using voice and face research to launder discrimination. The chapter's adult use is debiasing (structured interviews, blind auditions, policy-first ballots), not weaponising the cover.
- People respond socially to recorded voices and computers, then deny it.
- Voice pitch, pace and touch shift judgments of competence and warmth.
- Faces judged for competence predict elections within a second.
- Ask whether you are judging the person or the persona.
7. Sorting People and Things
| Story or study | What it shows |
|---|---|
| Supermarket lists | Categories aid memory |
| Eight lines | Categorizing polarizes |
| Didot, Lippmann | Origin of stereotype |
| Shoplifter study | Clothes stand in for character |
| Tajfel's survival | Prejudice as cognition |
| IAT; ninety-year-old doctor | Implicit bias and contact |
Why we categorize
Read someone a list of twenty supermarket items and they recall a few; group the same items into categories (fruit, cleaning products, dairy) and they recall far more. Every object and person is unique, but we could not function if we treated them that way. Categories let us respond to "bear," "chair" or "erratic driver" at speed, inheriting everything we know about the class. The cost is distortion. Categorizing works by maximizing differences between groups and minimizing differences within them.
Subjects estimated the lengths of eight lines, each 5 percent longer than the next. For half the subjects, the four longer lines were labeled "Group A" and the four shorter "Group B." Once labeled, lines within a group were judged closer in length than they were, and the gap between groups larger. The same effect appears with judgments of color, of attitudes, even of temperatures in different cities grouped by month. When we categorize, we polarize.
Stereotypes
The word "stereotype" was coined in 1794 by French printer Firmin Didot for a metal plate that printed identical copies. Walter Lippmann borrowed it in 1922 for the "pictures in our heads" we carry of groups, absorbed from culture and mass media. Mlodinow shows a nineteenth-century chart classifying people by the animal they resemble, physiognomy as pseudo-science.
He tells of a young Salvadoran man, the only nonwhite worker at a rural box plant, fired after his employer read ordinary behavior through a stereotype. We cannot know the employer's mind, which is exactly the point: categorical judgment usually feels like judging the individual.
In a discount store, a researcher's accomplice shoplifted in full view of a customer, then walked out of earshot but stayed visible. Sometimes the thief was well dressed, sometimes unkempt. Customers were far more likely to report the scruffy thief, and did so with relish ("that son of a bitch just stuffed..."), while hesitating over the well-dressed one. Clothing became a proxy for character.
Tajfel and the root of prejudice
The insight that ordinary perceptual categorization lies at the root of prejudice came from Henri Tajfel, a Polish Jew who served in the French army, was captured, and survived the war in German POW camps by hiding that he was Jewish and posing as French. He saw that how others categorized him decided whether he lived. After the war he argued that prejudice is not only hatred but a cognitive by-product of sorting. The field took decades to catch up.
The Implicit Association Test
Mlodinow walks the reader through the IAT logic with a list of names and relatives:
| Phase | Task | Typical speed |
|---|---|---|
| 1 | Say "hello" to male names and male relatives | fast (associations aligned) |
| 2 | Say "hello" to male names and female relatives | slower, about three-quarters of a second per word |
When the pairing follows your mental associations you speed up; when it cuts across them you slow down. The same method applied to science vs arts and male vs female, or to white vs Black faces and good vs bad words, shows that about 70 percent of test takers display a pro-white association, including many who are consciously appalled by the result.
Mlodinow quotes three heroes to show that stereotyping is not limited to villains: Gandhi in South Africa complaining that Indians were being degraded "to the level of the raw Kaffir"; Che Guevara writing disparagingly of Black people; and Abraham Lincoln in an 1858 debate saying he did not favor social and political equality of the races. Each was progressive by some measure, and each carried his culture's categories.
Fighting bias
U.S. discrimination law largely requires proof of conscious intent, which, Mlodinow argues, misses most real bias. But categorization can be shifted by conscious goals and motivation: studies show people can reduce automatic stereotyping when motivated, and exposure to individuals gives the mind counterexamples.
At his mother's assisted-living home, Mlodinow saw the residents as interchangeable: white hair, walkers, "building the pyramids." Then his mother mentioned dizziness when tilting her head back at the hairdresser, and a friend said it was a bad sign, the classic symptom of an occluded carotid artery. The friend was a retired physician. Individual contact dissolved the category.
The chapter closes with a seventy-seven-year-old London shopkeeper whose stroke in the lower occipital lobe left him unable to see that two different jugs, brushes or trains were the same kind of thing, or that "a" and "A" were the same letter. Setting a table became a struggle. Without categories, one could hardly survive even as an individual.
From socks to statute
The coordinated shoplifting field study (category of the "thief" varied; bystander intervention shifted) makes accentuation concrete. Tajfel's career, which the field was slow to absorb into the 1980s, reframed prejudice as often arising from normal categorization, not only from conscious hate. U.S. anti-discrimination law still largely requires proof of purposeful disparate treatment, which sits awkwardly beside a science of automatic association. That gap is why structural remedies (blinds, audits, diverse panels) matter more than sincerity pledges.
The IAT reliably measures group-level associations, but its use as an individual diagnostic has been heavily criticized. Its test-retest reliability is modest (around 0.5), and meta-analyses (Oswald et al. 2013; Forscher et al. 2019) found that IAT scores predict discriminatory behavior only weakly, and that changing IAT scores does not change behavior. Its creators now agree it should not be used to label individuals. The broader claim, that categorization distorts perception and fuels bias, remains well supported, and contact effects (Allport's contact hypothesis) are robust in meta-analysis.
Depth: categories as speed hacks with side effects
Tajfel's insight, which Mlodinow channels, is that categorization is not optional. Without categories the world is unusable; with them, within-group differences shrink and between-group differences grow (accentuation). Shoplifter studies that vary the offender's social category, the IAT's reaction-time asymmetries, and everyday stereotyping are one family: the mind sorts first and individuates later, if at all.
The IAT (Greenwald, McNee and Banaji, 1998) reliably detects group-level associations. As an individual diagnostic it is weak: test-retest around 0.5, and meta-analyses (Oswald et al. 2013; Forscher et al. 2019) find only modest prediction of discriminatory acts; changing IAT scores does not reliably change behavior. Axt's later work on SC-IAT variants improves some prediction contexts, but the creators themselves caution against labeling individuals. Contact (Allport) remains the more actionable solvent in meta-analysis.
Use the IAT to audit systems and corpora (word embeddings famously reproduce IAT-style associations: Caliskan et al. 2017), not to brand employees. Implicit-bias training that only shifts IAT scores tends to fade within days (Lai et al. 2016) and does not reliably change behavior (Forscher 2019, 492 studies). Prefer structural fixes: blind review, structured interviews, diverse panels, category audits in credit and hiring models.
SORT → POLARIZE → STICK. Categories speed thought and distort the items inside them. In-group items look more alike; out-group items look more extreme. Recite: category → accentuation → IAT association. Contact and individuation are the solvents.
List ten people you work with. Without overthinking, sort them into two piles by any quick rule (role, seniority, "easy/hard"). Then write one surprising, individuating fact about each person in the "other" pile. Notice how the first sort flattened them. That is accentuation catching itself.
IAT workshops that label individuals as biased, or that treat a reaction-time score as a moral verdict, invert the chapter. The IAT is a group-level association measure with modest test-retest reliability; meta-analyses find weak links to individual discriminatory acts. Shadow: weaponising the test, or denying that categories distort at all.
- Categories give speed and memory but exaggerate group differences.
- Stereotypes are categories applied to people, learned from culture.
- Implicit associations are common; the IAT measures them at group level.
- Motivation and individual contact reduce automatic bias.
8. In-Groups and Out-Groups
| Story or study | What it shows |
|---|---|
| Robbers Cave | Competition breeds hostility |
| Autokinetic norms | Groups converge on shared beliefs |
| Profession ratings | We like our own |
| Out-group homogeneity | They all look alike |
| Asian American women test | Identity cues shift performance |
| Klee or Kandinsky | Minimal groups and spite |
Robbers Cave
In summer 1954, at a camp in southeastern Oklahoma, psychologist Muzafer Sherif ran a field experiment that would be unethical today. Twenty-two eleven-year-old boys, all white, Protestant and middle class, were split into two groups of eleven, balanced for size and skill, and kept apart. Within days each group had a name (Rattlers and Eagles), a flag, and norms. Then the researchers introduced competitions with prizes: baseball, tug-of-war, treasure hunts. Hostility exploded: flag burning, cabin raids, stolen jeans, name calling, food fights. Ordinary boys had become warring tribes in a week.
Why groups matter to us
Humans evolved in bands, where belonging meant survival and rival bands meant danger. Researchers use in-group for any group a person feels part of and out-group for groups that exclude them. We belong to many (family, firm, profession, nation, fan base), and our identity shifts with context: the same person feels like a lawyer at a bar meeting, a woman in a male-dominated meeting, an American abroad.
Belonging has a price
Switching which in-group we adopt is, Mlodinow says, a trick we all use to keep a cheery outlook, because our groups are part of our self-image. Experimental and field studies show people will make large financial sacrifices to feel they belong to a group they aspire to, which is why some pay heavily for exclusive country clubs whose facilities they never use.
A games-company executive told Mlodinow about a senior producer who stormed in after a junior colleague got a promotion and raise. He had no money for another raise in a fiercely competitive market. Talking it through, he saw what really bothered her: the junior producer now had the same title. The compromise: a new title now, the raise later. Like the country club, he had sold her a high-status in-group membership in exchange for money. Advertisers know the dynamic: Apple spent hundreds of millions to tie the Mac in-group to smarts, elegance and hipness and the PC in-group to the opposite.
Shakespeare's Mark Antony knew it too. "Friends, Romans, countrymen, lend me your ears," Mlodinow translates, was really "In-group members, in-group members, in-group members . . ." A wise appeal.
Norms change what we see
Once inside a group, we absorb its norms, and they can reach down into perception. Muzafer Sherif, a Turk who came to America for graduate school, earned his Columbia PhD in 1935 with work on how group norms affect vision. In a dark room a fixed dot of light seems to move: tiny eye movements jiggle the image, and with no other objects for reference the brain cannot correct for them, so estimates of how far it moved vary widely. Sherif had people call out estimates in groups of three. Their numbers started far apart and, without discussion, converged on a narrow group norm; the norm differed widely from group to group. Brought back a week later and tested alone, people reproduced their group's estimate. The perception of the in-group had become their perception.
How we treat us and them
- Liking. People rate their own profession more likable: doctors, lawyers, waiters and hairdressers each rated their own group highest.
- Credit. We judge products made by our group as better, even when groups are arbitrary and we had no hand in the work.
- Out-group homogeneity. We see our in-group as varied and out-groups as all alike. Professionals estimating variability saw their own trade as more diverse. Newspapers run headlines such as "Blacks Seriously Split on Middle East" as if disagreement were news.
The numbers make the point. Doctors, waiters and hairdressers rated the other professions around 50 on a 100-point scale but their own around 70. Only lawyers rated fellow lawyers no higher than anyone else, which Mlodinow puts down to the one profession whose members routinely oppose each other. In a second study, people in groups of three made art from a toy set, designed senior housing and wrote fables; a "nonparticipant" from each group who did none of the work then judged the products. When they had a preference, they chose their own group's work two times out of three, though the groups had been formed at random.
Out-group homogeneity does not come from knowing the in-group better. Mlodinow jokes that physicists seem varied to him while investment bankers all read the Wall Street Journal and drink expensive wine, but the experiments show mere categorization is enough to produce the effect, even with random groups of strangers.
Harvard researchers gave Asian American women a hard math test. Before it, one group answered questions highlighting their Asian identity, another their female identity, a third neither. Those primed as Asian scored best and those primed as women scored worst. The identity in focus brought its stereotype with it.
Norms also backfire. A sign at Petrified Forest National Park lamenting that "many past visitors" stole wood nearly tripled theft, from about 3 percent of pieces in ten hours with no sign to about 8 percent with it, because it signaled that stealing was what people like us do (Robert Cialdini's work on descriptive norms).
The same trap runs through public messaging. The famous anti-litter ad with a tearful Native American (the 1971 "Keep America Beautiful" spot) preached against litter while showing that people like us litter. In a controlled study, an ad that simply condemned littering reduced it, but one that announced "Americans will produce more litter than ever!" did not. A campus warning that binge drinking "is prevalent" risks being heard as a norm to follow. Condemning a behavior while advertising its popularity can increase it.
Mlodinow also turns the idea into a personal tool. When he smokes an occasional cigar, he feels kinship with cigar-loving physicists, Einstein among them. When the habit grows, he switches focus to a different in-group of smokers: his father, who had lung problems, and a cousin with mouth cancer. Choosing which identity is salient changes which norms pull on you.
Social identity
Tajfel and his student John Turner later explained these effects through social identity theory: part of our self-esteem comes from the groups we belong to, so we are motivated to see our groups as positively distinct. That is why in-group favoritism appears even when there is nothing to gain materially: the gain is to the self-image. It also explains why the identity that is salient matters so much. When a context makes you feel like an engineer, the engineers' norms and rivalries govern your judgments; when it makes you feel like a parent or a citizen, other loyalties take over. Mlodinow notes that once he became aware of his many in-groups, he could catch himself shifting allegiance as situations changed, a small but useful form of self-knowledge.
Groups inside organizations
Because group identity needs so little to form, it matters at work. Companies gain by fostering employees' in-group identification through a distinctive, salient corporate culture, as Mlodinow says Disney, Apple and Google did very successfully. But strong identities in internal departments or divisions are dicey: they breed favoritism and discrimination, and research suggests hostility erupts more readily between groups than between individuals. Many companies also cultivate group identity among customers: Mac versus PC, Mercedes versus BMW versus Cadillac. We then treat such labels as meaningful far beyond their right, like dog person versus cat person or rare meat versus medium.
Minimal groups
Subjects were split by trivial criteria, such as preferring paintings by Klee or Kandinsky, or overestimating versus underestimating the number of dots flashed on a screen (often actually assigned at random), and then asked to distribute points or money among anonymous others identified only by group. They favored their own group. More striking, given a choice between maximizing their own group's reward and maximizing the difference between groups, they often chose the difference, accepting less so that "they" got even less. Group identity needs almost nothing to start, and it values relative standing over absolute gain.
| Option for the awarder | Own group gets | Other group gets |
|---|---|---|
| Maximum joint profit | 19 | 25 |
| Maximum in-group profit | 19 | 25 |
| Maximum difference (often chosen) | 7 | 1 |
Illustrative matrix in the style of Tajfel's; the pattern, not the exact numbers, is the finding.
Ending the war
At Robbers Cave, contact alone (shared meals, movies) made things worse. What worked were superordinate goals: Sherif secretly cut off the camp water supply and both groups had to work together to find the blockage; a truck carrying food "broke down" and both teams pulled it with the same rope once used for tug-of-war; they pooled money to rent a film. Hostility drained away; on the last day the boys chose to ride home on one bus.
Mlodinow lived near the World Trade Center on 11 September 2001 and describes how New Yorkers of every background drew together afterward. A common threat creates a larger "us."
Norms that backfire
Descriptive-norm messages can increase the bad behavior they name. "Americans will produce more litter than ever" and "many visitors steal petrified wood" both raised the rate by painting the vice as common. Injunctive norms ("please do not; it is wrong") work better. Minimal groups show the start condition; bad norm design shows how institutions accidentally instruct people in the wrong "we."
Minimal group favoritism is among the most replicated findings in social psychology. Robbers Cave is now viewed more critically: Gina Perry's 2018 archival work showed Sherif's team actively provoked conflict, and an earlier 1953 attempt (the Middle Grove study) failed when the boys refused to turn on each other. The stereotype-threat math study belongs to a literature whose effects shrank sharply in large replications (Flore and Wicherts 2015 meta-analysis; a 2018 multi-site replication). The Petrified Forest result is from Cialdini's group (published 2006) and fits a robust body of descriptive norms research, though the effect sizes in field studies are often modest.
Identity priming specifically. The Asian American women study (Shih, Pittinsky and Ambady 1999) was one of the targets of a 2014 replication special issue of Social Psychology: one team (Gibson, Losee and Vitiello) found a partial replication, another (Moon and Roeder) did not. That fits the broader priming crisis: behavioral priming classics such as the "elderly" walking-speed study (Bargh 1996) failed in direct replication (Doyen et al. 2012), and money priming failed in Many Labs. Treat any single "prime an identity, change performance" result as unproven. Minimal group effects survive, but some researchers (Yamagishi and colleagues) argue they partly reflect expected reciprocity within one's group rather than pure identity, since favoritism drops when people know their own group cannot reward them back.
What reduces group bias, then and now
| Approach | Robbers Cave | Evidence, today |
|---|---|---|
| Mere contact (shared meals, films) | Made things worse | Contact helps on average (Pettigrew and Tropp 2006: 515 studies, average correlation about -0.21), but a 2019 review of randomized trials (Paluck, Green and Green) found effects weaker for race and ethnicity and few long-term tests |
| Superordinate goals | Water, truck, film | Supported by the common in-group identity model (Gaertner and Dovidio); effects depend on groups keeping some distinct identity |
| Highlighting a shared threat | Not tested | Works short-term (post-9/11 cohesion), fades, and can create a new out-group |
| Implicit-bias training | Not applicable | Weak: Forscher et al. 2019 (492 studies) found changes in implicit measures do not reliably change behavior; Lai et al. 2016 found IAT shifts gone within days |
Design for shared outcomes rather than hoping attitudes change first. Interdependent goals, mixed teams with real stakes, and descriptive norms that show the good behavior as common have better evidence than awareness workshops alone.
Depth: from Robbers Cave to minimal groups
Sherif's Robbers Cave (1954) remains the narrative spine: twenty-two eleven-year-old boys, two cabins (Eagles / Rattlers), staged competition, hostility, failed mere-contact peace talks, then healing through superordinate goals (fixing a water supply, moving a stalled truck). Tajfel's minimal group paradigm stripped the story to the bone: a coin-flip or a Klee/Kandinsky preference is enough to bias point allocations toward "us." Norm formation (autokinetic effect groups converging on a shared estimate) shows how fast a we-feeling acquires content.
The practical lesson Mlodinow wants is organizational: mergers, party politics and fandoms rerun Cave unless leaders invent shared tasks that require both sides. Descriptive norms can backfire ("many visitors steal wood" increased theft); injunctive framing works better.
US / THEM → SHARED GOAL. Minimal cues create tribes; superordinate goals dissolve them. Recite Robbers Cave: conflict → contact fails → shared task heals. Tajfel: a coin flip is enough to start the split.
In your next cross-team meeting, invent a shared enemy that is a problem (outage, deadline, regulator), not a person or rival group. Assign joint ownership of one measurable fix. Watch whether language shifts from "your side / our side" to "the problem." That is Sherif's third stage without a summer camp.
"Shared goals heal tribes" becomes forced fun (trust falls, offsites) that leaves power imbalances untouched, or a rationale for dissolving legitimate dissent into "one team." Shadow: using superordinate goals as a smokescreen while keeping the spoils unequal. Robbers Cave worked when the boys needed each other for water and a stalled truck, not for slogans.
- Groups form fast, and competition turns them hostile.
- We like, credit and diversify our in-group, and flatten out-groups.
- Even meaningless groups produce favoritism and spite.
- Superordinate goals, not mere contact, repair group conflict.
9. Feelings
| Story or study | What it shows |
|---|---|
| Three Faces of Eve | Many selves |
| Sham angina surgery | Expectation heals symptoms |
| William James | We feel afraid because we tremble |
| Schachter-Singer, Capilano | Arousal gets labeled by context |
| Face swap, split brain | We explain choices we never made |
| Hiring study, rug | Introspection is theory |
Many selves
In the early 1950s, Chris Costner Sizemore came to psychiatrist Corbett Thigpen with headaches and blackouts. During treatment a second personality emerged with a different name and manner, the case later told as The Three Faces of Eve. Hers was extreme, but Mlodinow's point is general: we all have many identities, shifting with age, role and situation. Psychology long assumed fixed traits drive behavior, yet the book has already shown that names, fonts and weather sway us. So where do feelings come from, and do we know?
Pain and placebo
Even pain is constructed: sensory and emotional components come from distinct signals, and expectation modulates both. Placebos work beyond sugar pills.
In the 1950s, surgeons treated angina by tying off the internal mammary arteries, and patients improved. In 1958 a team ran a test: some patients got the real operation, others only skin incisions. Among those who got the real operation, 76 percent improved, and the sham group did just as well. The surgery relieved symptoms through expectation, not anatomy, and was abandoned.
William James and the body-first view
The modern view of emotion traces not to Freud but to William James. A sickly young man who joined Louis Agassiz's Amazon expedition, then fell into depression, James recovered partly by deciding to believe in free will. In his 1884 article "What Is an Emotion?" he turned common sense around: we do not tremble because we are afraid; we feel afraid because we tremble. Emotion is the perception of bodily change. Today's neo-Jamesian view says emotions are like perceptions and memories: reconstructed from available data, including body signals, context and expectation. His Principles of Psychology (1890) took twelve years; Wundt disliked it and James himself called it "a loathsome, distended, tumefied, bloated, dropsical mass."
Arousal plus explanation
Stanley Schachter and Jerome Singer (1962) injected subjects with adrenaline, telling some its true effects and others nothing (or that it was a vitamin). Each then waited with a confederate acting either euphoric (paper airplanes, hula hoop) or angry (over an insulting questionnaire). Uninformed subjects, feeling a racing heart with no explanation, caught the confederate's emotion; informed subjects attributed their arousal to the drug and stayed calm. Emotion = arousal + the label the context supplies.
Donald Dutton and Arthur Aron had an attractive woman interview men either on a solid, low bridge or on the swaying Capilano Suspension Bridge in Vancouver (five feet wide, 450 feet long, a 230-foot drop). She gave each man her phone number. Of the low-bridge men, 2 of 16 called her; of the high-bridge men, 9 of 18. Fear-driven arousal was misattributed to attraction. In a related study, men aroused by exercise or erotic film rated women as more attractive.
Yoga teachers' "calm your body, calm your mind" now has support: changing body states (slowing breathing, relaxing posture) changes the raw material the brain uses to build feelings.
We don't know why we like what we like
A researcher showed people two photos of women's faces and asked which was more attractive. Then, by sleight of hand, he handed over the one they had rejected and asked why they chose it. About 75 percent of the time people did not notice the swap, and they explained their "choice" with reasons, sometimes citing features ("I like her earrings") that only the rejected photo had. The same trick worked in a supermarket jam and tea tasting. Ask people why they prefer a product and you may get a fluent fiction.
Split-brain patients, whose hemispheres are surgically disconnected, show the same thing at the extreme. Instructed via the right hemisphere to wave, a patient's verbal left hemisphere, unaware of the instruction, explained that he waved because he thought he saw a friend. Michael Gazzaniga called this the left-brain "interpreter." Oliver Sacks's Korsakoff patient "Mr. Thompson," unable to form new memories, constantly confabulated new identities for people around him, throwing "bridges of meaning over abysses of meaninglessness."
How introspection actually works
When asked why we feel something, the brain searches its theories of what should cause such a feeling, then reports a plausible match. That predicts a test: if introspection is just theory, outsiders with the same cultural theories should explain our choices as well as we can.
Subjects evaluated a job applicant whose file varied in details: high grades, a car accident, meeting the applicant beforehand, spilling coffee in the interview. They then rated how much each fact influenced them. Outside observers who had never seen the file guessed how much each fact would matter. Subjects and outsiders agreed closely, and both were wrong about what actually moved the ratings (for instance, subjects denied that spilling coffee affected their liking, when it did). Introspection was no more accurate than a stranger's guess.
The $2,500 rug
Sham surgery, bridges and the supermarket narrator
1950s angina "internal mammary ligation" looked successful until pathologists found no new vessels: expectation and care had carried the relief. Dutton and Aron's shaky bridge, and exercise-then-film misattribution studies, show arousal from one source getting spent on another story. Choice blindness in the lab (defending the wrong face) and in the supermarket (defending the wrong jam) is the narrator at industrial scale. Sizemore's split self-images under two identities are the extreme clinical edge of a mundane fact: the story of why we feel something is often written after the feeling arrives.
Robust: the construction view of emotion, now championed by Lisa Feldman Barrett's theory of constructed emotion; choice blindness (Johansson and Hall), replicated in many domains, including political attitudes; split-brain confabulation; the general unreliability of introspection about causes. Weaker: Schachter-Singer has a poor replication record, and the adrenaline manipulation often failed to produce the predicted misattribution; the Capilano bridge result is famous but small (34 men) and later replications are mixed. The sham angina surgery trials (Cobb 1959, Dimond 1960) are solid and still cited in debates on surgical placebo; a 2018 trial (ORBITA) found similar placebo-level benefit for coronary stents in stable angina.
Depth: the narrator after the pulse
Chris Costner Sizemore (the "Eve" case) opens the chapter as a vivid reminder that self-reports of feeling can split. The scientific core is Schachter and Singer 1962: adrenaline arousal plus an angry or euphoric confederate produced emotion congruent with the context in uninformed subjects. James's older claim (we are sad because we cry) returns as physiology-plus-appraisal. Dutton and Aron's bridge study and misattribution-after-exercise studies show arousal from one source getting relabeled as attraction.
Choice blindness (Johansson et al.): people defend a face they did not pick, inventing sincere reasons. Supermarket taste-test variants show the same narrator at work on jam jars. The chapter's thesis for the rest of the book: introspection often reports a press release, not a transcript.
Constructionist accounts (Barrett) and appraisal theories both remain active; the shared ground is that raw affect is underdetermined and concepts/labels shape the experience. Choice blindness replicates. Sham-surgery and placebo literatures continue to show expectation folded into bodily report. Practical rule unchanged: treat "why I feel this" as a hypothesis, not as a download.
AROUSAL + LABEL → FEELING. Schachter and Singer: the body supplies intensity; the story supplies the name. Choice blindness and confabulation show the narrator inventing reasons after the fact. Recite: pulse first, plot second.
Next time you feel a spike (anger in traffic, nerves before a talk), pause and write two labels that could fit the same arousal: e.g. "threat" and "excitement," or "insult" and "surprise." Act for sixty seconds as if the second label were true. Note whether the feeling shifts. That is appraisal as a lever, not as denial.
"Feelings are constructed" becomes "your anxiety is optional" gaslighting, or a blank cheque for every impulse ("my feeling is my reason"). Shadow: using constructionism to dismiss real harm, or using "go with your gut" to skip evidence. Constructed does not mean fake; confabulated does not mean free.
- Emotions are built from body states plus context, then labeled.
- Arousal can be misattributed (bridge, adrenaline), though some classic studies are shaky.
- We often don't notice our own choices changing and explain them anyway.
- Introspection about causes is no better than an outsider's theory.
10. Self
| Story or study | What it shows |
|---|---|
| Michael Brown and Katrina | Self-serving blame |
| One million seniors | Above-average effect |
| Three Christs | Defending identity |
| Horse or seal, Princeton-Dartmouth | Seeing what we want |
| TAA, police chief, lawsuit | Shifting standards and stakes |
| Nicolai, parents' sewing night | Positive illusions enable action |
Nobody admits fault
After Hurricane Katrina in 2005, FEMA head Michael Brown was accused of mismanagement. Before Congress he blamed state and local officials and the media. Was he lying or did he believe it? Mlodinow argues that the stronger the threat to self-esteem, the more the mind views reality through a distorting lens, and that we usually believe our own spin.
The above-average effect
- Of nearly one million high school seniors, 100 percent rated their ability to get along with others as at least average; 60 percent put themselves in the top 10 percent, and 25 percent in the top 1 percent. Only 2 percent rated their leadership below average.
- 94 percent of college professors say they do above-average work.
- 30 to 40 percent of engineers put themselves in the top 5 percent.
- Physicians diagnosing pneumonia averaged 88 percent confidence and were right 20 percent of the time.
- Acquiring firms pay about 41 percent above market price, confident they can run the target better, and the combined firms usually lose value.
- The U.S. GAO estimated new-technology military equipment arrived on time and on budget 1 percent of the time.
We readily see overconfidence in others, but not in ourselves. Mlodinow also traces where the self comes from. By about age two, children recognize themselves in a mirror and begin to see themselves as social agents, and from then on the self is something to be protected. Freudian therapists and experimental psychologists agree on at least one point: the mind defends its self-image. The difference is mechanism. Freud imagined repression of dangerous wishes; the modern view sees a steady, low-key editing of evidence so that the story of the self stays coherent and mostly flattering. The executive whose department misses its numbers blames the market; the officer whose unit fails blames the orders. Each believes it, because the editing happened before the conclusion reached awareness. Mlodinow recalls Milton Rokeach's 1959 study at Ypsilanti State Hospital, where three patients who each believed he was Jesus Christ were housed together. Faced with contradiction, each found ways to preserve his identity. The self is defended at nearly any cost.
The scientist and the lawyer
Jonathan Haidt's metaphor: there are two ways to reach a belief, the scientist's (gather evidence, then conclude) and the lawyer's (start with the conclusion, then gather evidence). The brain, Mlodinow says, is a decent scientist but an outstanding lawyer. The name for this is motivated reasoning.
Our unconscious does not ignore unwelcome facts outright, which our conscious mind would catch. It uses subtler tools: 1. Interpretation of ambiguity in our favor. 2. Asymmetric standards: "Can I believe this?" for welcome evidence, "Must I believe this?" for unwelcome. 3. Shifting criteria after seeing which side we prefer. 4. Selective memory of our past.
Seeing what we want
David Dunning showed subjects an ambiguous figure that could be a horse or a seal. Half were told a farm animal meant orange juice and a sea creature meant a foul-tasting drink; the other half the reverse. Of those hoping for a farm animal, 67 percent saw a horse; of those hoping for a sea creature, 73 percent saw a seal.
In the 1950s "They Saw a Game" study, Princeton and Dartmouth students watched film of a brutal 1951 football game. Princeton students saw Dartmouth commit more than twice as many infractions as Princeton; Dartmouth students saw roughly equal numbers. Same film, different games. Scientists do it too: researchers judged studies more favorably when results supported their prior views, while denying any influence.
Mlodinow tells a joke in which God admits a Catholic and a Jew after they spell "God," then asks a Black man to spell "Czechoslovakia": the bar quietly rises for unwelcome cases. In a real study, volunteers tested themselves for a fictitious enzyme deficiency, "TAA," with a strip of paper. Those told that a color change meant good news waited calmly; those told no color change meant a health risk kept testing, dipping the strip again and again and waiting longer, seeking the result they wanted. The same dynamic, he argues, kept doubt about climate change alive despite scientific consensus.
Vested interest and the shifting yardstick
- Planning fallacy. Schedules are rarely ambiguous in hindsight, yet projects run late because the mind finds reasons to believe the hoped-for date.
- Hiring. Participants evaluating male and female candidates for police chief decided that "street smarts" or "formal education" mattered most depending on which credential the male candidate had: the criteria shifted to fit the preferred choice.
- The lawsuit experiment. Volunteers assigned to the plaintiff or defendant in a real injury case, with a bonus for guessing the judge's award within $5,000, predicted about $40,000 and $20,000 respectively. When they read the case before learning their role, the gap fell to about $7,000, and failures to settle dropped sharply.
- Memory. Of ninety-nine students recalling high school grades, accuracy fell from 89 percent for A's to 64 percent for B's, 51 percent for C's, and lower still for D's.
The case for positive illusions
Here the book turns. Mlodinow's son Nicolai, a short ninth-grader, wanted to play in the NBA; a letter from his younger self anticipated it. He practiced daily, and a new junior varsity coach noticed his work and put him on the team. Steve Jobs's "reality distortion field" at Apple and the optimism of early humans facing northern winters make the same point: unrealistic optimism is often what makes effort possible. Research links positive illusions to persistence, health and happiness.
His parents, broke on Chicago's North Side, faced eviction. His father, a sweatshop tailor, came home saying he had found his wife a seamstress job starting tomorrow. She could not sew. She thought herself clumsy and timid; he insisted she was capable and brave, a survivor like him. "And so my parents did not sleep that night, while my father taught my mother how to sew."
His closing thought: in physics, events obey one theory; in life, what happens often depends on which theory of ourselves we choose to believe.
Ambiguity as the lawyer's friend
Mobsters who called themselves public benefactors, professors who are all above average, and fans who saw two different Princeton-Dartmouth games are running the same subroutine: when the category is fuzzy, the self-serving fill wins. Climate-science "not yet settled" beliefs in the face of institutional consensus are the civic-scale version. The wetware Mlodinow cites (amygdala, ACC, posterior cingulate, precuneus) is the cost of defending identity. The adaptive upside (persistence, health, recovery) is real; so is the planning fallacy. Outside view, pre-mortem and accountability to people with no stake in your story are how adults pay the bill without declaring bankruptcy on the self.
Robust: motivated reasoning, biased assimilation (Lord, Ross and Lepper 1979), the planning fallacy, the lawsuit-role studies (Babcock and Loewenstein), and self-serving memory. The above-average effect is real, though part of it is statistical (skewed distributions and vague criteria). The Dunning-Kruger effect, a sibling claim, has been argued since 2020 to be partly a regression artifact. The benefits of positive illusions (Taylor and Brown 1988) are debated: later work suggests moderate optimism helps while large self-enhancement predicts poorer relationships. The TAA test is a real study (Ditto and Lopez 1992), and its finding has held up.
Depth: useful illusions and their bill
Katrina and the self-serving stories around it open the chapter; Dale Carnegie's gangsters who saw themselves as kindly businessmen set the tone. The above-average effect (94 percent of professors rating themselves above average, and the many twin findings) is the statistical joke that reveals the mechanism. Ambiguity is the ally: when "good driver" or "good leader" is fuzzy, the lawyer-mind fills the brief in its client's favor.
Hastorf and Cantril's Princeton-Dartmouth football study showed two audiences watching one film and seeing two games. Motivated reasoning about climate science is Mlodinow's civic example: the same evidence, different priors, different "settledness." Brain imaging (amygdala, ACC, posterior cingulate, precuneus) is offered as the wetware of defending the self and the tribe.
The adaptive upside: optimism and self-enhancement correlate with persistence, health behaviors and recovery. The bill: planning fallacy, ethical blind spots, and resistance to disconfirming data. Pre-mortems, reference-class forecasting and accountability to outsiders are the chapter's practical counterweights.
Dale Carnegie / Dutch Schultz: self-image of the "kindly" gangster. Hastorf and Cantril: Princeton-Dartmouth game, two films in one. Daniel Kahneman tradition: planning fallacy and outside view as counterweights (Mlodinow's neighboring theme from The Drunkard's Walk).
LAWYER MIND. The self is a defense attorney, not a judge: above-average illusions, motivated reasoning, self-serving memory. Recite: defend → distort → persist. The upside is resilience; the downside is blindness.
Write down your last three failures at work or home. For each, force a 50/50 split: half the inventory must be factors you controlled, half external. Then swap: rewrite so the controlled share grows by one item. The discomfort is the chapter. Pair with a pre-mortem on your next plan: "It is six months later and this failed; list three reasons."
Self-serving bias training becomes humiliation culture, or "own everything" advice that ignores structural barriers. Shadow: using motivated-reasoning talk to shame people who advocate for themselves, or celebrating delusional optimism as "growth mindset" while skipping the feedback loop that makes optimism useful.
- Most people rate themselves above average on nearly everything.
- The brain argues like a lawyer, with subtle double standards.
- Roles and stakes change what evidence we see.
- Positive illusions can fuel effort; the trick is to know when to check them.
Synthesis
The unconscious does most of the work; consciousness writes the press release.
The argument, reassembled
Read end to end, Subliminal is one argument made in two moves. Part I establishes the machinery: a narrow conscious channel sits on top of a vast automatic system that constructs perception (Chapter 2), reconstructs memory (Chapter 3), and evolved chiefly to run a social life (Chapter 4). Part II shows what that machinery does to our judgments of people: we read them through nonverbal leakage (Chapter 5) and surface cues like voice and face (Chapter 6), sort them into categories (Chapter 7) and teams (Chapter 8), and then turn the same apparatus on ourselves, building feelings we cannot explain (Chapter 9) and defending a self we overrate (Chapter 10).
Three threads run through every chapter. First, construction: whether the output is a scene, a memory, a feeling or a self-image, the unconscious assembles it from partial data and expectation, and hands it to consciousness as fact. Second, confident ignorance: people consistently fail to report the causes of their own behavior and supply plausible reasons instead. Third, adaptiveness: the shortcuts exist because they work most of the time; Shereshevsky, the London shopkeeper and TN show what happens when parts of the system fail.
Twenty-five takeaways
- Conscious bandwidth is tiny next to sensory input; automatic processing is the default, not the exception.
- Above-chance performance can feel like guessing (Peirce and Jastrow, TN, forced-choice designs).
- The new unconscious is adaptive automatic processing, not Freud's repressed cauldron.
- Context shapes experience itself, not just choices: price changes how wine tastes in the brain.
- Fluency is misread as truth, quality and safety.
- Perception is filled in: blind spots, saccades and masked sounds are patched invisibly.
- Hunches can be data; treat them as hypotheses to test.
- Memory keeps gist and invents detail; vividness and confidence are poor guides to accuracy.
- Post-event information contaminates memory; protect first accounts.
- False memories can be implanted in a sizeable minority of people.
- We are built for connection; exclusion registers as distress and fear drives affiliation.
- Theory of mind may be the reason for our large brains.
- Expectations leak through tone and face and can change others' performance.
- Gaze and space signal status; we calibrate them unconsciously.
- People respond socially to machines with voices.
- Faces judged in a second predict elections; looks carry competence signals that may be false.
- Categorization polarizes: groups look more different and members more alike than they are.
- Implicit associations are widespread at group level; measuring individuals is harder.
- Contact with individuals and conscious goals can weaken stereotypes.
- Minimal groups produce favoritism and even spite.
- Superordinate goals repair intergroup conflict better than contact alone.
- Feelings are interpretations of body and context.
- Introspection about causes is theory, no more accurate than an outsider's guess.
- Motivated reasoning works through double standards, not denial.
- Positive illusions fuel effort; pair them with external checks.
Frameworks table
| Framework | Source in book | What it explains | Use it when |
|---|---|---|---|
| Two-tier mind (two railway systems) | Ch1 to 2 | Most processing is automatic | Designing any human-facing system |
| Choice = preference + context + noise | Ch1 | Why irrelevant cues move behavior | Product, pricing, onboarding design |
| Construction of perception | Ch2 | Filling-in, illusions | Interpreting dashboards, sensor data |
| Gist, fill, believe | Ch3 | Reconstructive memory | Interviews, incident reviews, testimony |
| Social brain and ToM orders | Ch4 | Group size limits, mind reading | Team design, stakeholder mapping |
| Clever Hans and Pygmalion | Ch5 | Expectation leakage | Evaluations, A/B tests, model audits |
| Visual dominance ratio | Ch5 | Status signals in gaze | Negotiation, leadership presence |
| Media equation | Ch6 | Social responses to machines | Voice and AI assistant design |
| Categorize to polarize | Ch7 | Stereotypes from sorting | Segmentation, taxonomy, fairness reviews |
| Minimal groups, superordinate goals | Ch8 | Tribalism and its cure | Mergers, cross-team programs |
| Arousal plus attribution | Ch9 | How feelings form | Stress management, customer experience |
| Interpreter / confabulation | Ch9 | Invented reasons | Surveys, user research, AI explanations |
| Scientist vs lawyer | Ch10 | Motivated reasoning | Forecasts, investment cases, reviews |
Replication scorecard (as of October 2026)
| Claim | Chapter | Status |
|---|---|---|
| Blindsight | 2 | Robust (nature of residual awareness debated) |
| Phonemic restoration, filling-in | 2 | Robust |
| Price changes wine pleasantness | 1 | Robust in several labs |
| Portion size increases intake | 1 | Robust |
| Name-letter marriage effects | 1 | Weakened by confound analyses |
| Weather and stock returns | 1 | Mixed |
| Reconstructive memory, DRM | 3 | Robust |
| Planted false memories | 3 | Robust, rates vary by method |
| Change blindness | 3 | Robust |
| Infant helper preference | 4 | Failed in ManyBabies 4 (2025) |
| Social pain overlaps physical pain | 4 | Contested |
| Acetaminophen reduces social pain | 4 | Weak |
| Oxytocin increases trust | 4 | Failed large replications |
| Dunbar's 150 | 4 | Contested |
| Pygmalion IQ gains | 5 | Small and inconsistent |
| Universal facial expressions | 5 | Contested (constructionist challenge) |
| Media equation | 6 | Robust |
| Face competence predicts elections | 6 | Robust |
| Ovulation shifts preferences | 6 | Largely failed |
| Touch increases tips and compliance | 6 | Small effects |
| Category accentuation | 7 | Robust |
| IAT predicts behavior | 7 | Weak at individual level |
| Minimal group favoritism | 8 | Robust |
| Stereotype threat | 8 | Much smaller than first reported |
| Schachter-Singer arousal misattribution | 9 | Weak |
| Choice blindness | 9 | Robust |
| Motivated reasoning, biased assimilation | 10 | Robust |
| Above-average effect | 10 | Robust, partly statistical |
Mlodinow's thesis has aged well; the dual-process picture is now mainstream, and the evidence for construction in perception and memory has only grown. What has aged badly is the genre's habit of resting big claims on single, small, surprising studies. The 2015 Open Science Collaboration reproduced only about 36 to 40 percent of social psychology results, and behavioral priming in particular took heavy losses. Mlodinow largely avoided the most notorious priming claims, which is to his credit, but a 2026 reader should treat any sentence of the form "a single study showed people did X after Y" as provisional. A second limit: the samples are mostly WEIRD (Western, educated, industrialized, rich, democratic), and several findings, including expression recognition and some status cues, vary across cultures.
- Predictive processing (the brain as a prediction engine minimizing surprise, associated with Karl Friston and Andy Clark) has become a leading framework that unifies Mlodinow's chapters on perception, memory and feeling.
- Constructed emotion (Barrett, 2017) has displaced strong basic-emotion theories in much of affective science.
- Large preregistered replications and registered reports are now standard in social psychology.
- Neural decoding with AI models has advanced fast (see the Prologue update), raising neural-privacy questions; Colorado (2024) and California (2024) passed laws protecting neural data.
A practice guide
TWO TIERS, ONE NARRATOR. Unconscious processes do most of the work; consciousness press-releases a story. Recite the spine: cue → construct → remember gist → affiliate → read covers → sort tribes → label feelings → defend self.
Pick one live decision this week (hire, buy, reply, escalate). Write three columns before you act: (1) cues in the room, (2) cover impressions, (3) self-serving story you already prefer. Then force one disconfirming check (data, outside view, or a colleague with no stake). That is the book as a checklist.
A little subliminal knowledge becomes cynical fatalism ("everyone is biased, so anything goes") or manipulator's handbook. Shadow: using the science only to excuse yourself or to steer others. The adult use is design and self-catching: defaults, blinds, pre-mortems, recorded first judgments.
- Write down first impressions and first accounts before discussion contaminates them.
- Separate the speaker from the speech: read the transcript, not just the delivery.
- Blind the evaluator whenever you can: names off CVs, labels off samples.
- Ask "can I or must I believe this?" and apply the stricter standard both ways.
- Decide criteria before seeing options (the police chief study).
- Seek individual contact with groups you stereotype.
- Build superordinate goals across teams.
- Name your body state before naming your feeling.
- Instrument behavior rather than trusting stated reasons.
- Keep the optimism, schedule the audit: believe in the project, then let someone else check it.
Test yourself
- Why does Mlodinow say the conscious mind is more a narrator than a pilot? Give two experiments that support it.
- What did TN's walk down the corridor show, and which brain pathway explains it?
- Explain "gist, fill, believe" using the Jennifer Thompson case and John Dean's testimony.
- Why would a memory like Shereshevsky's be a handicap?
- What does the copier study reveal about when the unconscious hands control to conscious thought?
- How does the Clever Hans effect show up in machine learning?
- Compute your own visual dominance ratio in a meeting with a senior colleague. What do you predict, and why?
- Why did the competent-face effect matter more in low-information races?
- Describe the eight-lines study and explain what it says about segmentation in business.
- Why did Tajfel's subjects often prefer the maximum difference option?
- What made Robbers Cave hostility fade, and what did not?
- In the Capilano study, what was misattributed to what? Why is the study now treated cautiously?
- Why were outsiders as accurate as subjects in the Nisbett and Wilson hiring study?
- Give three tools of motivated reasoning and a professional example of each.
- When is a positive illusion helpful, and what guardrail keeps it from becoming dangerous?
Where to go next
- Daniel Kahneman, Thinking, Fast and Slow (2011): the System 1 and System 2 framing of the same two-tier mind, with more on judgment under uncertainty.
- Lisa Feldman Barrett, How Emotions Are Made (2017): the constructionist account that extends Chapter 9.
- Daniel Schacter, The Seven Sins of Memory (2001, updated 2021): the authoritative tour of Chapter 3's territory.
- Andy Clark, The Experience Machine (2023): predictive processing for general readers, the framework that now ties perception, memory and feeling together.
- Stuart Ritchie, Science Fictions (2020): the replication crisis, for calibrating which of the studies above to trust.
- Mlodinow's own Elastic (2018) and Emotional (2022) continue this line of work on flexible thinking and on feelings.
Glossary
| Term | Meaning |
|---|---|
| Blindsight | Responding to visual stimuli without conscious sight |
| Binocular rivalry / CFS | Technique that keeps one eye's image out of awareness |
| Change blindness | Failure to notice changes in unattended details |
| Confabulation | Producing false explanations or memories one believes |
| DRM paradigm | Word-list method that induces false recall of a theme word |
| Fluency | Ease of processing, misread as truth or quality |
| IAT | Reaction-time test of implicit associations |
| Implicit egotism | Unconscious preference for things linked to the self |
| Minimal group | Group formed on trivial criteria |
| Motivated reasoning | Reasoning steered by the conclusion one wants |
| Phonemic restoration | Hearing a masked speech sound that is not there |
| Superordinate goal | A goal that requires rival groups to cooperate |
| Theory of mind | The capacity to model others' beliefs and intentions |
| Visual dominance ratio | Gaze while speaking divided by gaze while listening |
About this book
Subliminal: How Your Unconscious Mind Rules Your Behavior by Leonard Mlodinow, 2012. This is a chapter-by-chapter study of the book, with summaries, maps, tables and diagrams; short quotations keep their page references. It is shared free for personal study, is not for sale, and is not endorsed by the author or publisher. Please read and buy the original.
Insight Editions · rajeshrmahapatra.com/insight-editions