IQ at Eleven Predicted Survival at Eighty. Richard Haier on the Finding Psychology Tried to Bury

Lex Fridman opens by asking how human intelligence is measured. Richard Haier, a psychologist who wrote *The Neuroscience of Intelligence* and taught *The Intelligent Brain* for The Teaching Company, starts with the vernacular: being smart, reasoning well, figuring things out, knowing what to do when you do not know what to do. For research, almost everyone in the field uses the *g* factor — general intelligence — the mental ability common to virtually all tests of mental ability. The term goes back more than a century to Charles Spearman, who noticed that students' exam scores across subjects were all positively correlated and essentially invented the correlation coefficient to describe it.

Positive correlation means that doing well on one test predicts doing well on another, even across disciplines. People push back with anecdotes: someone excellent at X and mediocre at Y. Haier says exceptions do not overturn a general tendency, and everyday experience is not the same as the data. The person who is merely "not so good" at the second thing is not the opposite of good; that is not a negative correlation. "I know a guy" is not a research design.

Asked how well *g* has held up, Haier says it is the most replicated finding in all of psychology. Cynics mention psychology's replication crisis; he replies that there is no such crisis around the existence of this general factor. Give any battery of mental tests to a large, diverse sample and extract what they share with factor analysis — not a statistical artifact, he insists. You will find verbal, numerical, and visuospatial clusters, and those clusters still share variance. That shared part is *g*. It usually accounts for around half the remaining variance. It is the single biggest factor, the most reliable, the most stable, and the one most influenced by genetics. Training and drugs have not been shown to raise it.

The other half includes test-taking skill, specific strengths such as arithmetic, and short-term memory, which is among the cognitive processes most highly correlated with *g* without being *g* itself. It does not matter which battery you use. Take any 12 tests at random, extract a *g*, take another 12, extract another *g*, and those two *g*s correlate above 0.9. The same structure shows up across cultures. Haier likens the mind to a computer designed to be general, and Fridman quotes Miyamoto Musashi: if you know the way broadly, you will see it in everything.

IQ and *g* get used interchangeably in casual speech. Haier pulls them apart. *g* cannot be measured directly; it is a statistical extract. An IQ score is a good estimate of it, because a standard IQ test is itself a battery of verbal, numerical, spatial, and memory subtests compressed into one number. The score works precisely because it is not specialized. You still cannot put a ruler on someone's head and read off *g*.

IQ scores are interpreted as percentiles, not as a ratio scale. A person with IQ 140 is in the top 1 percent; IQ 70 is two standard deviations below the mean. You cannot say the first person is twice as smart as the second, any more than a doubled chess rating means twice the player. There is no absolute zero on the scale, even if, Haier jokes, you think you have met people with zero intelligence.

Stability is not a slogan. In the 1930s, researchers in Scotland gave an IQ test to every eleven-year-old in the country. The paper sat in a storeroom at the University of Edinburgh until Haier's colleague Ian Deary found it, digitized it, and followed the survivors, including with brain imaging. IQ at 11 and IQ at 70 were highly correlated. Split the age-11 scores into quartiles and look at who was still alive 70 years later: roughly twice as many people from the highest quartile as from the lowest, in both men and women, in a country with universal healthcare. Why is not obvious. The correlation stands anyway.

What makes a good IQ item is not how it looks; it is its empirical relationship to a total score that is already valid by other means. Haier illustrates with the Minnesota Multiphasic Personality Inventory, whose true-false items — preferring a shower to a bath, ranking Lincoln against Washington — have no obvious connection to personality disorders and still distinguish groups. That method is "dust-bowl empiricism": content does not matter if the items separate the criterion groups. Most intelligence items, by contrast, have face validity. They look like they measure a mental ability.

A standard battery mixes general information (roughly how far from Washington to Miami), vocabulary (the word *regatta* was dropped after complaints that disadvantaged test-takers would not know a sailing race), digit span forward, and digit span backward. Vocabulary and backward digit span, surprisingly, are highly correlated with *g*. Another example is the Hick paradigm: simple reaction time to a light barely correlates with IQ, but reaction time to a slightly more complex rule — press the button by the light farthest from the other two — correlates strongly. Time limits matter for the same reason. Unlimited time lets people do better without distinguishing them as well. Speed of information processing is part of *g*.

Fridman describes severe test anxiety: sweating through timed verbal SAT items, seeing typography instead of meaning, always preferring homework in a cave to being watched with a clock. Haier does not deny it. Anxiety can make a score a bad estimate of *g*; it does not mean *g* is absent. Retake the test without a 102-degree fever and the score may rise. That is a better estimate, not evidence that *g* increased. The SAT, he adds, is highly *g*-loaded. The College Board does not want it called an intelligence test, but a reasoning test of verbal and mathematical reasoning has to be related to *g*.

College admissions almost never rest on one number, and should not. High-school GPA misses the bored-smart kid; a single SAT sitting misses the student with a fever. Many people show "tilt," a disparity between math and verbal scores. Non-native English speakers who crush math and lag verbally are not mysterious to a math department. Fridman worries about missing a genius the way a justice system worries about convicting the innocent. Haier says it is statistically unlikely, and the real question is which mix of information maximizes decision quality. Personality, he notes, is also stable, and it is one reason Fridman's SAT anxiety still shows up in the rest of his life.

The reason to study intelligence, Haier says, is not the number on the paper. It is what the number reflects about navigating a complicated world. "Life is one long intelligence test," and academics who criticize the research are, as a group, pretty smart people who have enormous difficulty imagining daily life at IQ 80 or 75. Even 85 or 90 is a different everyday experience. He points to *Judge Judy*, where ordinary disputes display the full range of problem-solving, sometimes for laughs that he does not find funny. The U.S. military, which he calls perhaps the world's best organization at training diverse people, has historically used a cutoff around IQ 83 because people below it could not be trained for military jobs. Reasoning skill, he says, cannot simply be taught.

Nobody in the field, Haier insists, actually reduces a person to a number. The fear is legitimate; the practice is not. A graduate-school professor of his was the only person he ever knew who judged students solely by test scores, and even that man backed off later. Fridman presses the moral wound: if some people are more intelligent in a stable, partly genetic way that education cannot compensate, does that mean some people are better than others? Haier has not yet fully unpacked the genetics, but he grants the interpretation that genes strongly influence *g*. Twenty-first-century genetics is often probabilistic, not deterministic; expression can be influenced. Whether those influences are only biological, or also environmental and cultural, is the live controversy.

On "better," he is blunt. There is zero evidence that smart people are better at honesty, kindness, or likability. Fridman, rereading *The Rise and Fall of the Third Reich*, notes that some of history's worst atrocities were committed by extremely intelligent people in the intellectual class. If there is a *g* factor for goodness, it is independent of intelligence and harder to define. Haier still asserts that more intelligence is better than less — not that more intelligent people are better people. All else equal, would you take a pill that made you smarter or dumber? Intelligence does not lead to happiness. N-back training does not raise *g*. The Mozart effect was never believed by intelligence researchers and meta-analyses killed it. Shifting the whole distribution even half a standard deviation, he thinks, would be world-shaking because so many social problems are worsened by weak everyday reasoning. Fridman wonders whether the gain is obvious only at the low end, and whether some human bugs are features. Haier would like science to predict whether an IQ pill would make a given life better or worse. *Flowers for Algernon* is the literary version of that question.

Charles Murray and the late psychologist Richard Herrnstein published *The Bell Curve* in 1994. Haier knows Murray and has talked with him at length. The book, he says, is an empirical study of intelligence in everyday life, using large databases to show that IQ scores correlate with social problems and social benefits. That is not why it exploded. The fire was one chapter on the average difference in scores between Black and white Americans, about one standard deviation, roughly 15 points — an observation recorded for decades on academic tests, mental-ability tests, and IQ tests. Herrnstein and Murray wrote clearly, Haier says, that they were agnostic about whether the gap was genetic, that they did not care, and that a group average never licenses an inference about an individual. Critics, he suspects, often did not read the book. The take-home they carried was that the authors were calling Black people genetically inferior, or at least implying it by discussing the gap at all.

To understand the blast radius, Haier goes back 25 years. In 1969 the *Harvard Educational Review* asked UC Berkeley educational psychologist Arthur Jensen — whom Haier also knew — to review early-childhood programs meant to raise minority children's IQs, before federally funded Head Start was fully underway. The programs were literally called compensatory education. The governing assumption was that the average gap came from unequal opportunity and racism, and that lunches, cognitive stimulation, and missing inputs would close it. Jensen, an empirical psychometrician, wrote a hundred-plus-page article cataloguing design flaws. Some programs claimed five-point gains; a few claimed 10, 20, even 30. One, the "miracle in Milwaukee," later sent its investigator to jail for fabricating data. Jensen's opening line, as Haier remembers it: we have tried compensatory education and it has failed.

The fateful extra step was suggesting that, given the lack of measurable effect, researchers should consider a genetic influence on the group difference. Haier calls it the most infamous paper in all of psychology. Genetic data in 1969 were skimpy. Jensen was vilified to a degree Haier says is hard to overstate: bomb threats emptied APA lecture halls, campus security watched him, mail was opened, he changed addresses. This was before internet mobs. Overnight, Haier has written, all intelligence research became radioactive. Nobody wanted to talk about it; almost nobody wanted to do it.

When *The Bell Curve* arrived 25 years later, Murray has said they debated including the chapter and wrote it carefully, which did them no good. A *Nightline* producer asked Haier on to discuss the book. He said it was a good book about intelligence in society. She wanted the race chapter. He said he would tell television that it was not clear there was a genetic component to group differences — but that if there were a strong one, that would be a good thing, because the more genetic a difference is, the more biological, and the more biological, the more science might fix it. Silence on the line. Would he say that on television? He said yes. That was the end of the booking.

Fridman asks whether Bell Curve-style research can increase racism and hate. Haier has thought about it as a person, not only as a scientist. His sense is that enormous reservoirs of hate and racism have nothing to do with scientific knowledge of the data. He does not want to give racist groups veto power over what scientists study. Virtually no one, he notes, denies that score gaps exist. The fight is over causes and fixes. If you think the problem is cultural, which part of the culture do you want to change, and are you willing to accept academic disadvantages if you will not change it? If you think it is environmental, name the parameters. Lead in gasoline, paint, and water is an environmental toxin society can eliminate, and should.

A journalist once asked Jensen straight out whether he was a racist. Jensen said he had thought about it a lot and concluded it did not matter. Haier thinks Jensen had a touch of Asperger's, and that the sentence would get him cancelled on Twitter immediately. What Jensen meant, Haier says, was his "default hypothesis": the factors that affect individual intelligence are likely the same factors that affect group differences. It was a hypothesis to test, and he was willing to drop it if the data failed. He was committed to the ideal that this is an empirical question. From that perspective the scientific method cannot be racist; biased scientists get replicated, or they do not.

Fridman pushes hard. Psychology is full of gray areas; media and virality misinterpret findings. He has had emails accusing him of blood on his hands over semi-autonomous-vehicle papers. Scientists, he argues, must anticipate misreading when they decide whether to run a study, publish it, and which words to use. Haier does podcasts and public books to say what he thinks the data mean and do not mean. He barely explains papers on Twitter because they are complicated. Choosing not to publish from fear of harm is not an unreasonable thought, and people will decide differently. As editor of the journal *Intelligence*, he sometimes publishes group-difference papers written for scientists, not for Twitter, and does not promote them much. He has refused a BBC invitation that would have cast him as justifying "race science." After *The Bell Curve*, he says, there has been little new work; what exists still finds group differences, and there is still no demonstrated way to fix school underperformance for many disadvantaged and minority students.

The first public-policy step, Haier says, is for people who write education policy to look at intelligence data. Search a 20,000-word policy paper for the word *intelligence* and, in most discussions of achievement gaps, it does not appear. California's recent fight over high-school math curriculum featured a Stanford education professor who asserted there is no such thing as mathematical talent and wanted advanced classes gone because not everyone could do them. The review retreated somewhat. Haier finds it shocking that a university professor in charge of the process believed talent does not exist, let alone that intelligence data were absent from the discussion.

Leave race out of it, he says, and look at the Coleman Report. It measured school variables and teacher variables against academic achievement and found the strongest predictors were what the student brought into the school — essentially ability. Combine school quality, funding, teacher quality, and teacher education, and they barely accounted for 10 percent of the variance. That result has been replicated. If you want a policy on raising achievement, he asks, how much money do you want to put into teacher education and administration? Ask the Gates Foundation: they spent a tremendous amount, measured the outcome because they are measurement people, found no impact, and pulled back from that kind of program.

Fridman worries, via a *Clerks* riff about inconvenient truths, that laying the data out will cause society to give up on people, misallocate attention, and abandon equal opportunity. Haier calls the concern realistic and misplaced. If you want to fix the problem, you have to know what it is. About 16 percent of the population has an IQ under 85. In the United States that is about 51 or 52 million people, including 14 million children. Below that line, finding gainful employment that sustains even a minimal modern life is very difficult — not impossible, very difficult. The overlap with school failure and life struggle is large. That is why an IQ pill that shifted the curve, if it existed, would matter. Ignoring the number does not make it smaller.

Stephen Jay Gould argued in *The Mismeasure of Man* that *The Bell Curve* rested on four false assumptions: intelligence reduces to a single number, people can be ranked linearly, intelligence is primarily genetic, and it is essentially immutable. Haier says Gould's views on this were overtly political, not scientific. Listeners should Google Gould's claims and then Google the scientific reviews; those reviews take the statements apart. By the second edition, MRI studies already showed brain volume correlated with IQ scores, a biological finding Gould had denied and then declined to add. The book remains a staple on college assigned-reading shelves. People who understood the data took it apart point by point. Gould, Haier says, did not care.

Technical criticisms of *The Bell Curve*'s statistics exist; Murray answered them in articles that are online. Haier also recommends Murray's later *Human Diversity*, which marshals data against the claim that race is only a social construct with no biological meaning, and *Facing Reality*. You need not agree, he says, but Murray presents data and his critics' data without personal attack. People who think they know what is in *The Bell Curve* are likely incorrect and should read it. Fridman finds even the existence of *g* difficult in the way mortality is difficult: it collides with the American wish that everything is possible.

What Haier wants done with the information is more neuroscience on the molecular biology of learning and memory, because one definition of intelligence is how much you can learn and remember. Nobel-level work already exists there; the missing step is taking individual differences seriously. Cognitive psychologists who became molecular biologists have historically hated that concept. Haier once sat next to a Nobel laureate in memory research who told him not to go there, that it would set the field back 50 years.

On life success, there is a tremendous amount of validity data. Success is a broad word, but measurable pieces — including life expectancy and income — correlate with IQ. Personality and so-called emotional intelligence matter; empirically, general reasoning ability matters most, and it matters more as jobs get more complex. Ceiling effects are real inside occupations, but not always. The Study of Mathematically Precocious Youth, which Haier worked on at Johns Hopkins as a graduate student, gave 11- and 12-year-olds the SAT math exam and found children scoring like incoming Hopkins freshmen. A 50-year follow-up of that top 1 percent, split into quartiles, found the top quarter of the top 1 percent earned more patents, publications, and university tenure than the bottom quarter of the same 1 percent — based on a test at age 12. Fridman finds that painful. Haier asks whether untrainable charisma is also painful. The slogan that hard work lets you be anything you want is, Haier says, a mutation of the older claim that hard work can make you successful. He can work as hard as he likes and will not become a successful theoretical physicist. That is not a judgment on anyone's basic humanity.

James Flynn, who had died about a year before the recording, documented that IQ scores have been drifting up over the years. Test makers already knew, because they have to renorm periodically: ten items correct meant a different percentile 50 years ago than it does today. People are getting more items right. The rise is about three IQ points per decade, a cohort effect rather than a personal one. Two questions follow. Does the gain sit on the *g* half of the variance or on the other half? There is evidence it is a *g* effect. And what on earth causes it?

The second question has been used to attack a genetic influence: natural selection does not move on a scale of decades, so how can intelligence be genetic? Haier replies that the Flynn effect can act on the non-genetic portion. Genes may account for about half the variance, perhaps as high as 80 percent in adults, which still leaves room. The cause remains a mystery, though as editor of a Flynn memorial issue he was seeing papers converge on nutrition and health care, with evidence the effect is slowing and possibly reversing. Nutrition would still connect to *g* because it acts on the biology that produces intelligence; infant data make that link less unreasonable, and it does not logically cancel a genetic influence.

Score a modern test on 1940 norms and you get a much higher IQ number. Are we really smarter than a couple of generations ago? Haier says no, but we might solve some problems a little better because of *Sesame Street*, more schooling, and a pile of factors still waiting to be disentangled. The population looks, from a zoomed-out view, as if it is getting smarter. Clear answers are not in yet.

How much of intelligence is nature, how much nurture? All of it, Haier says: all nature and all nurture, because the two are not independent. Many behavioral geneticists think the interactions are too complex to disentangle now or ever, whatever happens with DNA technology. Haier only half agrees. If the complexity were truly that intractable, a stable *g* factor should not exist. It does, from many data sources over decades. So he is more willing to believe gene-environment interactions can be explicated and used as a basis for a molecular biology of intelligence.

IQ correlates with cortical thickness and with the brain's glucose metabolic rate. Identical twins reared apart are highly similar in IQ. Does anyone believe intelligence has nothing to do with the brain? The live dispute is how deterministic the hardware is. Development from conception through childhood is probabilistic; uterine environment and chance affect how neurons connect. Haier's reading is still that a genetic blueprint has a profound influence on later intellectual development, including *g*. Potential can be damaged or, within limits, supported. That is why new parents buy educational toys and fuss over organic baby food. Asked to steel-man the view that genetics is a tiny component, he says the data do not support it.

What would change his mind? If environment were very important, he thinks it would have been obvious by now. School interventions, compensatory education, and early-childhood programs have been earnestly tried in well-funded, well-designed studies. They do not show the effect. That is what people told Jensen: start earlier, pre-pre-kindergarten. After 50 or 60 years, something should have produced a result you can see without a p-value of 0.05. Adopted children, as adults, have IQs that do not correlate with adoptive parents and do correlate with biological parents they never met. Some adoption data show a small shift toward the adoptive family; Haier uses a "weight of evidence" standard and does not let any one study dominate. Mozart, memory training, and computer games have shown no impact on intelligence test scores.

Books in the house correlate with IQ, even if the child never reads them, because more intelligent people own more books and, if intelligence is partly genetic, pass along genes as well as shelves. The claim that SAT scores merely measure parental wealth invites the next question: why are the parents wealthy? Studies of socioeconomic status and IQ usually lack genetically informative designs, so causal stories about SES are a stretch. Where such designs exist, most outcome variance traces to the genetic component; SES sometimes adds a little, not much. Birth-order and fertility findings, Haier says, are messy and not clearly holding up. The closer you look, the more questions you get. That is normal in science.

Haier has long believed consciousness and intelligence are the two fundamental problems of the human brain, and therefore related. Early in his career, with anesthesiologist Mike Alkire, he used PET imaging — injecting radioactive sugar so harder-working regions took up more of it — on volunteers anesthetized as if for surgery. Each person came three times: no drug, drowsy, fully under. Different anesthetic drugs hit different regions; the common thread seemed to involve the thalamus in the sequence of losing and regaining consciousness. They later wanted fMRI for real-time resolution PET could not give. A question Haier still cannot answer: does an IQ-140 person need more anesthetic to lose consciousness than an IQ-70 person, or less? Textbooks dose by weight. Chart reviews failed because exact doses were not recorded. His early PET intelligence studies had found the opposite of "brighter is more activity": glucose metabolic rate was negatively correlated with test performance, the brain-efficiency hypothesis, still being studied. Fridman jokes that a consciousness factor and a *g* factor might sum to one — depth of experience traded against depth of reasoning. Haier calls it a great hypothesis for someone listening to run.

On machines, Haier quickly hits the edge of his knowledge. When Alexa appeared, he suggested feeding it IQ items; it searched the internet and could answer Washington-to-Miami, but that is not the same as repeating a digit string backward. Failure there might be a language-channel problem, like giving an English IQ test to someone who does not speak English. Fridman notes that the Jeopardy-winning system was heavily hard-coded. He describes François Chollet's ARC challenge: low-bar visual puzzles about symmetry, counting, and object-ness that are trivial for a three-year-old and still brutal for machines. Novelty is part of the hardness. Test authors will not share items because fluid intelligence — using what you know on a problem you have never seen — dies once the item is public. Crystallized intelligence is the fact stock; fluid intelligence is what you do with it.

Will we one day modify biology or genetics to change intelligence? Haier is supremely confident a metaphorical IQ pill is possible, he says, because he is supremely ignorant of neurobiology's complexities. The nightmares of synaptic cascades are what fuel some people. Cancer is still being figured out; it was only recently understood why aspirin works. The history of science is the history of solving the impossible-seeming. An Alzheimer's drug is, in practice, a learning-and-memory drug, and learning and memory sit at the core of intelligence. The so-called IQ pill may well come out of that pipeline. What it would do in a college student without Alzheimer's is an open question. Fifty-one million Americans with IQs under 85, struggling with everyday life, is the number he thinks should motivate people to look.

Fridman asks for advice to young people. Haier will not dictate anyone's passion. If you care about psychology, consciousness, intelligence, or psychiatric illness, pursue the science of the big questions. An Alzheimer's program is already an intelligence program in disguise. The most important rule in a research life: follow the data where the data take you. You cannot decide in advance where you want them to go. If they lead into a field where you lack technique, recruit people who have it. If they lead somewhere controversial, still push forward, and communicate with skill, compassion, and a wider sense of the human condition. A famous psychologist wrote that a balance must be found between pursuing the science and respecting people's legitimate sensitivities. He did not say how. The "somehow" is left to the reader.

Asked the meaning of life, Haier will not tell him. Does he fear finiteness? More and more as he gets older. It is a human condition: finite, widely known, mostly denied. Sometimes it becomes estate planning; sometimes it makes you work harder because time is shorter. It sits in the back of the mind. Fridman compares it to *g*: a hard truth you walk past and still cannot escape. Everybody already knows the everyday version. You knew who the smart kids were in school. You have been furious at a customer-service script that does not answer the question you asked. Everybody has dealt with people they judged as not at their level. That is how humans are. It is also why internet pile-ons that declare someone hopeless from one interaction are dangerous. Our judgments of other people are worse than a battery of tests, soaked in bias, history, and emotion.

Whatever the science says, Fridman argues, it does not mean we should withhold compassion. Haier says that is exactly what the science says, not the opposite. Everything he knows about psychology and intelligence points to treating people as individuals, respectfully and with compassion, because through no fault of their own some people are not as capable as others. You can turn a blind eye, or invent theories about why the difference is not there. He would like to fix some of it as best he can. And everybody is deserving of love. Fridman is still unconvinced that more intelligence always creates more beauty; poems, music, and tenderness are loosely correlated with IQ, if at all. Haier agrees he never claimed more intelligence is always better for beauty — only that, all else equal, more is better than less. Einstein's line closes the episode: it is not that he was so smart, but that he stayed with the questions much longer.