On intellect

Intellect is among the most consequential human differences and one of the least comfortably discussed. We accept variation in height, speed, musical ear, visual-spatial sense, patience, and coordination without much unease. Cognitive differences provoke a different reaction, as though describing variation in the capacity to learn, reason, and handle complexity were the same as ranking human worth.

It is not, and it has made the subject nearly impossible to discuss usefully. This is not an essay about whether intelligence differences are permissible to notice — they exist, they are large, and they have shaped education, work, and institutions for as long as records exist. The more interesting question is practical: what is this trait, how do we actually measure it, how does it relate to personality and other capacities, and how should a society organize itself around a form of variation it cannot legislate away.

Not whether IQ is real. What to do with the fact that it is.

What the score actually measures

What people call "IQ" is not a substance or an organ. It is a score derived from performance on a battery of standardized cognitive tasks — typically some mix of verbal comprehension, fluid reasoning on unfamiliar problems, working memory, processing speed, visual-spatial reasoning, and quantitative reasoning. The instruments that produce the number, such as the Wechsler Adult Intelligence Scale, break performance into indices along these lines and then combine them into a composite¹.

The reason a single number is defensible at all is a finding, not an assumption: people who perform well on one demanding cognitive task tend, on average, to perform well on structurally unrelated ones. Charles Spearman documented this "positive manifold" in 1904 and called the shared factor g, for general intelligence². John Carroll's later reanalysis of more than 460 datasets found that this general factor accounts for something like 40-50% of the variance across diverse cognitive batteries³. That is a large, replicated effect by the standards of the social sciences, and it is the reason psychometricians treat g as the most useful single index of cognitive capability, rather than a philosophical convenience.

g is not omniscience, wisdom, creativity, or virtue. It is a statistical regularity, not a mechanism, and it says nothing about whether a person uses their capacity well, kindly, or at all.

Where the trait comes from, and why that matters for policy

The uncomfortable part of the picture is heritability. Twin and adoption studies consistently find that genetic influence on general cognitive ability is substantial and, unusually for a psychological trait, increases across the lifespan: from roughly 41 percent in childhood to about 55 percent in adolescence and 66 percent by young adulthood in one large longitudinal sample⁴, with adult estimates in other twin cohorts running as high as 75 to 85 percent⁵. Shared family environment, by contrast, tends to fade in influence as people age and increasingly select their own environments.

This matters practically, not morally. It means a meaningful share of adult cognitive variation is not attributable to effort, upbringing, or schooling in any simple sense, and it is one reason "just work harder" is an incomplete answer to differences in learning speed and complexity tolerance. It also means environment is not irrelevant — the same research tradition that produced the Flynn effect (a rise in raw test scores of roughly two to three IQ points per decade across most of the twentieth century, observed from the United States and the United Kingdom to Japan, Kenya, and China) shows that population-level cognitive performance is sensitive to nutrition, health, schooling, and other environmental inputs, and that the rise has stalled or partially reversed in several developed countries since the 1990s, including Norway, Denmark, and the UK⁶. A Norwegian study using within-family data confirmed the reversal was itself environmentally caused, not a change in the underlying gene pool⁷. Genes set a range; environment and circumstance decide where in that range a person lands, and where a population's average drifts over time.

Intellect is not personality, and that independence is the whole point

A recurring confusion is treating IQ and personality as versions of the same thing — as if a "smart" person should also be disciplined, or a conscientious person should also be sharp. The data says otherwise. Across the Big Five, only openness to experience shows a reliably positive association with measured intelligence, and even that correlation is modest, typically in the range of r = 0.17 to 0.30, concentrated in the "ideas" and intellectual-engagement facets rather than openness broadly⁸. Conscientiousness — the trait most consistently linked to real-world achievement — correlates with g at essentially zero. The largest current meta-analysis, spanning 369 studies and over 345,000 people, put the correlation at r = 0.01⁸.

This is not a footnote. It is the reason that the claim that IQ and conscientiousness are society's two best predictors of career outcomes carries real statistical weight: they are not two measurements of one underlying quality, they are two nearly orthogonal axes of variation, each explaining outcome variance the other cannot touch. There's a distinction between simple and complex jobs. In simple, repetitive work — line work, stocking, checkout — IQ predicts how fast someone learns the job, but once learned, it is conscientiousness, not IQ, that predicts how well and how consistently they do it⁹. In complex work, where the demands change often enough that the job can never be fully "learned" — most managerial and administrative roles — IQ becomes the dominant predictor again, and is roughly three times as powerful as conscientiousness for that category⁹. Entrepreneurial and creative work adds a third variable: after IQ, the next-best predictor is trait openness, the same dimension that clusters entrepreneurs with artists.

The academic literature behind this popular framing is Frank Schmidt and John Hunter's meta-analytic work on personnel selection, which found that general mental ability alone predicts overall job performance at a validity of about r = 0.51 — the single best predictor psychology has for that outcome — and that adding a conscientiousness measure lifts the combined validity to roughly 0.60, an 18 percent gain, precisely because the two traits are measuring different things¹⁰. It is worth noting, in the interest of not overstating the case, that more recent re-analyses using twenty-first-century data have found meaningfully smaller validity coefficients for general cognitive ability alone — closer to 0.22 to 0.31 once modern corrections for range restriction are applied¹¹. The direction of the effect has held up for over a century of research; its exact magnitude has not stayed as large as the original estimate.

How complexity, not job title, sets the requirement

Jobs differ enormously in how much reasoning under novel, changing conditions they demand, and the validity of cognitive tests for predicting performance rises in step with that complexity — from about 0.23 in the simplest, most repetitive jobs to roughly 0.58 in professional and managerial work^{10, 12}. Her research also maps the middle 50 percent of applicant IQ ranges onto specific occupations:

Source: Gottfredson (1997, 2004)^{12, 13}.

The pattern holds at the population level, too. Gottfredson describes IQ of 91–110 as the "middle 50 percent" who are readily trainable for the bulk of jobs in the economy; 111–125 as "out ahead," with most occupations cognitively within reach and a real shot at graduate or professional work; above 125 as "yours to lose," where the constraint on career outcome shifts almost entirely away from raw ability; and 76–90 as an "uphill battle," where the jobs realistically within reach are concentrated in production, food service, and custodial work¹². No civilian occupation in the modern American economy routinely recruits below IQ 80, and federal law bars military enlistment below roughly the same threshold — for years the Army's practical floor sat closer to 85¹². Roughly 10 percent of the population falls below an IQ of 83, a level below which formalized training in a standard job setting has historically produced little benefit, and which a modern, increasingly automated economy has fewer and fewer structural niches for¹⁴. The occupational sketch for the 116–130 range — defense attorney, chemist, engineer, executive manager, research analyst, roughly the top 8 to 20 people out of 100 — sits close enough to Gottfredson's empirical bands to function as a reasonable popularization of it, not a distortion of it¹⁵.

What this is worth, in dollars, and why the premium compounds

The income effects are real but smaller than the discourse around them often implies. A widely cited meta-analysis by Tarmo Strenze puts the raw correlation between IQ and income at about r = 0.20 across 31 studies and nearly 59,000 participants — meaningful, but explaining only a modest share of the variance in earnings on its own¹⁶. Jay Zagorsky's analysis of the National Longitudinal Survey of Youth found each additional IQ point associated with roughly $234 to $616 in additional annual income after controlling for age, gender, race, and education — a range wide enough to reflect genuine uncertainty rather than a precise formula¹⁶. Notably, the premium is not flat across a career: it is nearly absent for new graduates in their early twenties, when entry-level pay clusters tightly regardless of ability, and roughly doubles by the mid-thirties to mid-forties as complex, self-directed work increasingly rewards the ability to handle novelty¹⁶. One re-analysis of the same data found conscientiousness carrying a comparable, in some specifications even slightly larger, standardized effect on earnings than IQ itself (betas of roughly 0.44 versus 0.41), suggesting conscientiousness can partially compensate for several points of cognitive ability over a career¹⁷. The two traits behave less like rivals and more like a portfolio: one determines a person's ceiling in complex, changing work, the other determines how reliably they perform toward it.

Building a society around a difference you cannot vote away

This is where the discourse usually collapses into either denial or fatalism, and both are practical failures. The honest, non-moralized version of the argument runs like this: variation in the capacity to learn quickly, hold abstractions, and adapt to novel complexity is real, substantially heritable, largely independent of virtue or effort, and strongly predictive of which kinds of work a person will find tractable versus exhausting. A society that pretends this variation does not exist will keep routing people into educational and career paths mismatched to their actual cognitive profile — the four-year college default for people whose aptitude and interest point toward skilled trades, apprenticeships, or technical certification being the clearest current example. A society that treats the variation as a caste system, using a single test score as a hard gate for opportunity, ignores the wide confidence intervals around any individual score, the modest and complexity-dependent validity coefficients above, and the fact that conscientiousness, openness, and plain circumstance still do enormous work in determining outcomes.

The more useful design goal is matching, not sorting: build enough distinct, respected paths — technical, managerial, creative, entrepreneurial, skilled-trade — that people land in roles whose complexity matches their capacity, rather than funneling everyone toward one credential and calling the rest failure. Gottfredson's own data point toward the sharpest practical problem: as an economy automates away simple, repetitive tasks, it also erodes the very tier of jobs that historically gave people below the middle of the distribution a stable, dignified place to work. The jobs that have proven hardest to automate so far tend to combine fine motor skill with unscripted human interaction — food service, care work, skilled trades — rather than sheer manual repetition, which suggests where policy and training investment should concentrate next¹⁸.

What artificial intelligence changes, and what it doesn't

This is the live variable. For most of the twentieth century, complex work rewarded exactly the cognitive machinery g measures: holding many variables in working memory, reasoning quickly through unfamiliar problems, synthesizing information under time pressure. AI systems are now automating a meaningful share of that specific labor — drafting, summarizing, coding, first-pass analysis — which raises an open question the psychometric literature has not settled: does routine use of these tools amplify a person's effective intelligence, or quietly erode the underlying capacity by removing the practice that builds it.

Early evidence points in an uncomfortable direction. An MIT Media Lab study tracked 54 participants writing essays with either an LLM, a search engine, or no tool at all, using EEG to measure neural connectivity across four months. Brain connectivity scaled down systematically with the amount of external support: the unassisted group showed the strongest and widest neural engagement, the LLM group the weakest, and participants who had grown reliant on the LLM struggled to quote from essays they had written minutes earlier and reported markedly lower ownership of their own work¹⁹. Separately, a 2025 survey study found heavier AI tool use associated with weaker critical-thinking performance, with cognitive offloading identified as the likely mechanism²⁰. Neither result proves that AI use permanently lowers general intelligence — both are early, and the MIT study is explicit that its findings are preliminary — but they are consistent with a plausible and testable hypothesis: outsourcing the "hard part" of thinking to a tool, the way outsourcing arithmetic to a calculator changed which mental skills stayed sharp, changes which cognitive muscles get exercised.

The competing view, grounded in the philosophical "extended mind" thesis, treats this the wrong way. On that account, cognition was never confined to the skull — a notebook, a calculator, and now a language model are all just external components of a person's problem-solving system, and the right question is not whether people are "getting dumber" but whether they are learning to direct and evaluate a more powerful extended system²¹. Both things are probably true at once, sorted by how a person uses the tool. Offloading a task you never intend to learn — routine formatting, boilerplate, first drafts you will rewrite anyway — is straightforward amplification. Offloading a task you needed to struggle through to build the underlying skill is the atrophy the MIT data captured.

If that split holds, it reshapes the practical value of the two traits this essay opened with. Raw processing speed and working-memory capacity — the parts of g most directly substitutable by AI — become relatively less scarce as differentiators, because a tool can supply them to anyone who asks. What AI cannot yet supply is judgment about which of its outputs to trust, taste about which problems are worth solving, and the conscientiousness to actually verify, refine, and ship the work rather than accept the first plausible answer. In other words, the future may not eliminate the value of the two traits so much as rebalance them: less reward for raw horsepower alone, more reward for the discipline to use a vastly more powerful tool well. A new form of stratification is already visible along exactly that seam — not simply high-IQ versus low-IQ, but people who use AI to extend their standards versus people who use it to lower them.

The demographic frontier: how many high-IQ minds remain to be found

Before we turn to conclusions, there is a live and uncomfortable question about the global supply of intellect itself: how many very-high-IQ people will there actually be in the future, and where will they come from? On the surface, it seems the answer should scale simply with population — more people means more bright people. The truth turns out to be more complex, because cognitive ability is unevenly distributed across populations, and populations are changing at very different rates.

There is a robust, well-documented negative correlation between a country's average IQ and its fertility rate, on the order of r ≈ −0.72^{22, 23}. Sub-replacement fertility has concentrated in the world's richest, highest-scoring nations — East Asia, Europe, and North America — while faster-growing populations in Sub-Saharan Africa and South Asia sit at the lower end of the measured IQ distribution. There is also a smaller but real "dysgenic" effect within many countries: lower-IQ individuals tend to have more children than higher-IQ ones, estimated to shave roughly 0.35 to 0.4 IQ points per decade off the average within a given country²².

The practical consequence of these combined trends is what gives the "demographic frontier" its bite: the global count of very-high-IQ people is projected to decline, not because any individual is becoming less intelligent, but because the populations that historically produced the most IQ-130+ individuals are shrinking relative to those producing fewer. A recent dataset by Parra and Kirkegaard puts national IQs at roughly 100 for East Asia, 95 for Europe, 75 for South Asia, and 70 for Sub-Saharan Africa. Because IQ-130 is an extreme tail probability, small differences in a country's mean translate into enormous differences in how many of its citizens reach that level: at a mean of 100 and standard deviation of 15, about 2.3% score above 130; at a mean of 70, essentially no one does. This means the global 130+ count is dominated almost entirely by a handful of high-IQ, low-fertility countries. As their share of world population falls, the absolute number of people in that cognitive bracket falls with it, even though no individual's ability has changed²⁴.

This is not a fringe or speculative concern. It is a defensible extrapolation of a genuine, robustly-documented fertility–IQ relationship: the core sentiment — that the global count of very-high-IQ people is declining because high-IQ populations are shrinking faster than low-IQ populations are growing — follows from simple arithmetic once you accept the empirical premises. A world where sub-replacement fertility concentrates in the richest, highest-scoring countries will, by that logic, produce fewer 130+ individuals over time.

The practical implication for how we organize society is significant nonetheless. If the absolute number of people capable of the most complex scientific, technical, and institutional work is shrinking or leveling off — rather than growing with population — then the talent pool for solving hard civilizational problems is more constrained than a simple "more people, more progress" model would suggest. That makes it more important, not less, to identify and develop that capacity wherever it exists, rather than assuming high intelligence will continue to scale automatically with raw population growth.

The practical conclusion

None of this licenses treating a test score as a verdict on anyone's worth, and none of it licenses pretending the score measures nothing real. The honest position sits between those two failure modes: intellect is a genuine, substantially heritable, and unevenly distributed capacity that predicts — and more strongly as job complexity rises — how readily a person handles novel and changing demands; conscientiousness is a nearly independent trait that predicts how reliably they execute once the demands are familiar; and a society serious about both dignity and function needs institutions, career ladders, and now AI tools designed around that reality rather than around the comfortable fiction that everyone starts from the same cognitive baseline and only effort divides them. The task is not to argue the difference away. It is to build enough distinct, well-matched paths — and, increasingly, well-designed tools — that the difference translates into varied contribution rather than exclusion.

References

1. Verywell Mind — The Wechsler Adult Intelligence Scale: https://www.verywellmind.com/the-wechsler-adult-intelligence-scale-2795283

2. cogn-iq.org — Spearman's Two-Factor Theory (g and s): https://www.cogn-iq.org/blog/spearmans-two-factor-theory-g-and-s/

3. Wikipedia — G factor (psychometrics): https://en.wikipedia.org/wiki/G_factor_(psychometrics)

4. PMC — Heritability of cognitive ability, childhood to adulthood: https://pmc.ncbi.nlm.nih.gov/articles/PMC2889158/

5. PMC — Adult twin studies, heritability 75–85 percent: https://pmc.ncbi.nlm.nih.gov/articles/PMC3276760/

6. Wikipedia — Flynn effect: https://en.wikipedia.org/wiki/Flynn_effect

7. PNAS — Flynn effect reversal in Norway, within-family study: https://www.pnas.org/doi/10.1073/pnas.1718793115

8. Wikipedia — Intelligence and personality: https://en.wikipedia.org/wiki/Intelligence_and_personality

9. Planned Man — Jordan Peterson on IQ, conscientiousness, and career success: https://www.plannedman.com/the-means/work/jordan-peterson-looks-into-your-soul-predicts-your-career-success/

10. Omnia/Brainpower — The Validity of Cognitive Ability (Schmidt & Hunter meta-analysis): https://cdn2.hubspot.net/hubfs/4094759/omnia-brainpower-the-validity-of-cognitive-ability.pdf

11. PubMed — Contemporary re-analysis of general mental ability validity coefficients: https://pubmed.ncbi.nlm.nih.gov/38059952/

12. Gottfredson (1997) — Why g Matters: The Complexity of Everyday Life: https://gwern.net/doc/iq/ses/1997-gottfredson.pdf

13. Gottfredson (2004) — Life, Death, and Intelligence: https://www1.udel.edu/educ/gottfredson/reprints/2004LifeDeathIntelligence.pdf

14. YouTube — Jordan Peterson on IQ 83 and automation: https://www.youtube.com/watch?v=okOj61U-tLs

15. YouTube — Jordan Peterson on IQ ranges and career fit: https://www.youtube.com/watch?v=fjs2gPa5sD0

16. IQ Career Lab — Income and wealth data, IQ point salary value: https://www.iqcareerlab.com/resources/income-and-wealth/iq-point-salary-value-data

17. Human Varieties — How well do personality traits predict social outcomes: https://humanvarieties.org/2023/02/28/how-well-personality-traits-predict-social-outcomes-well-its-complicated/

18. greyenlightenment — Jordan Peterson and Stefan Molyneux discuss IQ: http://greyenlightenment.com/2017/09/03/jordan-peterson-and-stefan-molyneux-discuss-iq/

19. MIT Media Lab — Your Brain on ChatGPT: https://www.media.mit.edu/projects/your-brain-on-chatgpt/overview/

20. Phys.org — AI use linked to eroding critical-thinking skills: https://phys.org/news/2025-01-ai-linked-eroding-critical-skills.html

21. SAGE Journals — The extended mind thesis: https://journals.sagepub.com/doi/pdf/10.1177/18344909241309376

22. Jensen, Pesta & Kirkegaard (2025) — International meta-analysis of differential fertility for intelligence: https://openpsych.net/paper/82/

23. Shatz (2008) — IQ and Fertility: A Cross-National Study, Intelligence: https://www.sciencedirect.com/science/article/abs/pii/S0160289607000244

24. Parra & Kirkegaard (2025) — National IQs: Measurement and Defense: https://openpsych.net/paper/85/

25. United Nations — World Population Prospects 2024: Summary of Results: https://population.un.org/wpp/assets/Files/WPP2024_Summary-of-Results.pdf

26. Gibson & Light (1967) Cambridge-academics IQ study, cited in "IQs of Academics in Different Disciplines": https://ijrr.web.baylor.edu/sites/g/files/ecbvkj2171/files/2026-03/ijrr10001.pdf

27. Simonton (2015) — Genius, Creativity, and Talent: https://simonton.faculty.ucdavis.edu/wp-content/uploads/sites/243/2015/08/GeniusCreativityTalent.pdf