Haier, Richard

intelligence researcher, editor of the journal Intelligence, author of The Neuroscience of Intelligence

Richard Haier: IQ Tests, Human Intelligence, and Group Differences

Lex Fridman Podcast · Jul 14, 2022

• Richard Haier
• g factor and general intelligence
• IQ testing and psychometrics
• The Bell Curve
• Arthur Jensen controversy
• genetics and heritability of IQ
• race and intelligence controversy
• Flynn effect
• brain efficiency hypothesis
• neuroscience of intelligence
• AI intelligence testing

Core ideas

Haier explains that intelligence research centers on the "g factor" (general intelligence), a statistical construct dating to Charles Spearman's early-1900s discovery that scores on virtually any battery of mental tests correlate positively with one another, implying a common underlying factor extractable via factor analysis. He states that the existence of g is, in his assessment, "the most replicated finding in all of psychology," found across cultures and stable regardless of which specific tests are used (independently drawn 12-item batteries yield g factors correlated above 0.9 with each other), typically accounting for about half the variance in test performance, with the rest attributable to specific abilities, test-taking skill, and memory. He describes IQ scores as a practical estimate of g (not a direct measurement), explains that good test items are selected empirically for their statistical relationship to overall score regardless of surface content (illustrated via the unrelated example of the MMPI personality inventory), and notes that verbal, backward-digit-span, and simple-reaction-time-to-complex-choice tasks are particularly g-loaded, while emphasizing that time limits matter (untimed tests fail to discriminate well) and that state factors like anxiety or illness can degrade an individual test score without changing the underlying trait — a point Fridman raises from personal experience with test anxiety.

Haier discusses a Scottish cohort study (children tested at age 11 in the 1930s, later followed into old age by researcher Ian Deary) finding that IQ quartile at age 11 predicted mortality roughly 70 years later (about twice as many survivors in the top quartile as the bottom), even under universal healthcare, though he says the explanation (genetic, environmental, or some combination) remains unresolved. He states his own interpretation that genetics has "a very important influence" on g — citing twin studies (identical twins reared apart show highly similar IQ), correlations between IQ and cortical thickness or brain glucose metabolism, and adoption studies (adult adoptees' IQ correlates with biological, not adoptive, parents) — while stressing that gene expression is probabilistic rather than strictly deterministic, and that decades of attempted interventions (early childhood compensatory education, memory/n-back training, the "Mozart effect") have failed to move g in controlled studies, which he cites as the evidentiary basis for his view rather than dismissing environmental influence outright. He states that if evidence turned up (e.g., environmental interventions clearly and reproducibly closing gaps at a population level) he says would change his view, but that it has not appeared despite 50-60 years of well-funded attempts, including Gates Foundation-funded education initiatives he says showed no measurable impact.

The conversation's most sensitive segment concerns The Bell Curve (Herrnstein and Murray, 1994) and the earlier 1969 Arthur Jensen article. Haier, who knew both authors personally, describes the book's core empirical content (IQ correlates with many social/economic outcomes) as uncontroversial, and states that its authors were explicit and agnostic about whether the documented ~15-point average test-score gap between Black and white Americans has a genetic component, and explicitly stated group-average differences say nothing about how to treat any individual — but that this nuance was, in his account, widely lost in public reaction, which he says framed the book as asserting genetic inferiority. He recounts Jensen facing bomb threats and being "vilified" after his 1969 article argued that early-childhood compensatory education programs had failed to close the gap and that a genetic contribution should therefore be considered, and states his own view (offered explicitly as personal opinion, not scientific consensus) that publishing accurate data does not hand "racist groups... veto power over what scientists study," while adding that if anything, a genetic explanation would be less socially fraught than people assume, since it might in principle be more tractable to biological intervention than diffuse cultural/environmental causes. He recounts declining, on a taped call with a Nightline producer, to frame this position on-air, and criticizes Stephen Jay Gould's The Mismeasure of Man (a popular critique of psychometrics) as scientifically flawed and "overtly political," noting Gould omitted contrary brain-imaging evidence from a later edition. Haier states plainly that current evidence still shows the average group differences persist and that there is very little active research on their causes, calling it professionally risky territory that most researchers avoid; he distinguishes this from his own editorial practice of publishing group-difference research in the journal Intelligence for a scientific audience while declining broader media appearances that seek to stage the topic as a debate over "race science."

Fridman pushes back at several points, raising the ethical burden on scientists to anticipate misinterpretation (drawing an analogy to his own autonomous-vehicle research) and citing the historical example of highly intelligent people having propped up atrocities (citing his own rereading of The Rise and Fall of the Third Reich) as evidence that intelligence and moral goodness are unrelated — a point Haier agrees with explicitly, stating there is "zero evidence that smart people are better" people. Both agree the core policy takeaway should not be to "give up" on any individual or group, and Haier repeatedly emphasizes that population-level statistical findings say nothing about how to treat or evaluate a given person, invoking "all men are created equal" as a value independent of empirical psychometric findings.

On practical implications, Haier notes that about 16% of the U.S. population (an estimated 51-52 million people, including 14 million children) has an IQ below 85, which he says correlates strongly with difficulty finding stable employment (citing the U.S. military's cutoff around IQ 83 for trainability), and argues this population-level reality is largely invisible in education policy discussions, citing the Coleman Report's finding that school and teacher variables together explain only about 10% of variance in student achievement outcomes. He discusses the Study of Mathematically Precocious Youth (a 50-year longitudinal study of profoundly gifted children first identified via early SAT-taking) finding that even within the top 1% of scorers, the top quartile substantially outperforms the bottom quartile in later-life measures like patents and tenure. He also discusses the "Flynn effect" (population-average IQ scores rising roughly three points per decade across the 20th century, now possibly slowing or reversing), attributing this — per emerging research he is now seeing as a journal editor — primarily to improved nutrition and healthcare rather than genetic change, while noting this doesn't logically undermine a genetic component to individual differences.

On his own past neuroimaging research, Haier describes early PET-scan studies (with anesthesiologist Mike Alkire) examining which brain regions turn off during loss of consciousness (implicating the thalamus), and separately describes his own surprising early finding of an inverse correlation between brain glucose metabolic rate and intelligence test performance — leading to the "brain efficiency hypothesis," still studied today, that efficient rather than more intense neural processing underlies higher g. He discusses attempts (with mixed success) to build intelligence tests for AI systems, citing François Chollet's abstraction-and-reasoning test designed around tasks trivial for humans (symmetry, counting, pattern completion) but difficult for machines, as a more meaningful AI intelligence benchmark than voice assistants that merely search the internet. He closes on a note of moral caution consistent with his framing throughout: whatever the science says about the stability and heritability of individual differences in intelligence, he argues it implies not resignation but an obligation to treat all people "as individuals, respectfully and with compassion," precisely because some people are, through no fault of their own, less cognitively capable than others.

Excerpts