Appearance
1.9 — What Intelligence Is, and What Actually Moves It
In 1904 a British army officer turned psychologist named Charles Spearman collected the school marks of a group of village children — classics, French, English, mathematics, pitch discrimination, and the teachers' ratings of how bright each child seemed.
He expected the subjects to be unrelated. A child good at music need not be good at Latin.
Every pair of scores was positively correlated. Not perfectly, but consistently: children who did well in one thing tended to do well in the others, across subjects with nothing obvious in common. Spearman argued that some single underlying factor was contributing to all of them, called it g for general intelligence, and started an argument that has run for a hundred and twenty years without stopping.
The argument is worth understanding properly, because both popular positions on it are wrong. "IQ is meaningless" is wrong. "IQ is your worth as a person" is grotesquely wrong. The truth sits in an awkward middle that nobody enjoys quoting.
What the finding actually is
Spearman's observation, called the positive manifold, has been reproduced thousands of times in every population where anyone has looked. Give people a set of mentally demanding tasks that seem unrelated — vocabulary, spotting the next shape in a sequence, repeating digits backwards, arithmetic, assembling blocks to match a pattern — and performance on all of them correlates positively.
Statistically, a single common factor accounts for something in the region of 40 to 50 per cent of the variation between people across such a battery. The rest is specific abilities and noise.
Two competing readings of that fact have coexisted for a century.
Something real and singular is being measured — perhaps the general efficiency of the brain, its processing speed, or the integrity of its wiring.
Or lots of separate abilities happen to develop together, because they share nutrition, health, schooling, home environment and practice, which would produce exactly the same correlations with no single underlying quantity.
Nobody has settled this, and it is worth saying plainly rather than picking a side. What is settled is that the measurement behaves consistently, whatever it is measuring.
The useful division: fluid and crystallised
Raymond Cattell's split from the 1940s is the one that matters practically, because the two halves behave completely differently over a life.
Fluid intelligence is reasoning with new material — solving a problem you have never seen, spotting a pattern in unfamiliar symbols, holding several conditions in mind while working something out. It needs no prior knowledge.
Crystallised intelligence is everything you have accumulated — vocabulary, facts, procedures, the judgement built from having seen this situation eleven times before.
The lifetime curves are the important part and are not widely known.
The practical reading of that graph: the thing that declines with age is the thing that matters least in real work, provided you keep accumulating. Expertise is crystallised intelligence, and it is still being built at sixty.
What an IQ score is
An IQ score is not a quantity of anything. It is a position in a distribution. The tests are constructed so that the average for the age group is 100 and about two-thirds of people fall between 85 and 115. A score of 130 means roughly the top two per cent for your age. That is the entire meaning of the number.
Three consequences people routinely miss.
It is relative, so it can move without anything changing inside you. If everyone else improves and you do not, your score falls.
The tests are re-normed periodically, because raw performance has been rising. Across the twentieth century, raw scores rose by roughly three points per decade in many countries — the Flynn effect — which means a person scoring 100 in 1930 would score somewhere in the seventies against a modern sample. Nobody thinks our grandparents were intellectually disabled. What rose was familiarity with abstract classification, schooling, nutrition and health. In several rich countries the rise has stalled or slightly reversed since the 1990s. The Flynn effect on its own destroys any simple reading of the score as a fixed biological quantity.
Individual scores are noisier than people assume. Retest the same person and a movement of five points either way is unremarkable. Treating 118 and 124 as meaningfully different is over-reading the instrument.
What the score predicts, honestly
This is where both camps get uncomfortable, so here it is in order.
It predicts academic performance reasonably well — correlations commonly reported around 0.5 with school achievement. That is a real relationship and it is far from determinism.
It predicts job performance modestly, and the size of the correlation has been revised downward in recent re-analyses of the classic studies, which had methodological problems. The relationship is larger in complex jobs than simple ones.
It correlates with income, health and longevity — partly through education, partly through the kinds of jobs it opens, partly through health behaviour.
And here is the part that matters for your life: within the normal range, it is a weak predictor of individual outcomes. Group averages differ; individuals scatter enormously. Take a hundred people at 115 and a hundred at 100, and the two groups will overlap in almost every outcome you can measure. Conscientiousness — showing up, finishing, doing what you said — predicts life outcomes about as well as intelligence does, and in some studies better. So does the sheer number of hours of deliberate practice (3.5). So do circumstances nobody chose.
The honest summary: it measures something real, it matters most at the extremes and least in the middle, and it is one modest input among several.
The popular alternatives, assessed
Multiple intelligences. Howard Gardner proposed in 1983 that there are separate intelligences — linguistic, logical-mathematical, spatial, musical, bodily-kinaesthetic, interpersonal, intrapersonal, naturalistic. The idea is enormously popular in education and has essentially no supporting evidence. The specific problem is that the proposed intelligences are not independent when measured — they correlate, which is the positive manifold again. What Gardner actually identified is that people have different talents and interests, which is obviously true and does not require a new theory of intelligence.
Learning styles — that you are a visual, auditory or kinaesthetic learner and should be taught accordingly. This is the most thoroughly refuted popular idea in education. The test is straightforward: teach matched and mismatched groups and see if the matched ones learn more. Study after study finds no benefit. People do have preferences; the preferences do not predict what works. 3.4 explains what actually determines whether a format helps, which is whether it fits the material rather than the person.
Emotional intelligence. Real as a set of skills, oversold as an "intelligence". 4.4 gives it a full and fair treatment, including where the popular claims outran the evidence by a wide margin.
Growth mindset — that believing ability can improve leads to improvement. Attractive, partly true, and much smaller than the popular version. Large pre-registered trials find effects that are real but modest, concentrated among lower-achieving students. Worth knowing; not transformative.
The heritability trap
You will read that intelligence is "about 50 to 80 per cent heritable" and you will almost certainly misunderstand it, because the technical meaning of the word is nearly the opposite of the everyday one.
Heritability is a statement about the causes of differences between people in one population at one time. It is not the proportion of your intelligence that came from your genes. That question is not even coherent — you cannot separate the contribution of the seed from the contribution of the soil to the existence of one plant.
Two facts make the point concrete.
Heritability of intelligence rises with age, from around 30 per cent in childhood to considerably higher in adulthood. That looks backwards until you see the mechanism: as people gain freedom to choose their own environments, they select ones that fit their tendencies, which amplifies small initial differences. Some of what looks genetic is genes causing an environment.
Heritability is lower in poor environments. In samples with significant deprivation, the heritability estimate drops sharply, because when the environment varies a lot, environmental differences do more of the work. In good conditions, most of the remaining variation is genetic — precisely because the environmental obstacles have been removed for everyone. High heritability is therefore, perversely, a sign of an equal environment rather than a fixed destiny.
And the decisive point: heritable does not mean unchangeable. Height is highly heritable and rose by many centimetres in a few generations on better nutrition. Eyesight is heritable and is corrected with glasses.
What actually raises it
Now the useful part. Sorted by how good the evidence is.
Schooling. This is the strongest finding on the page. Studies exploiting natural experiments — changes in school leaving age, birth-date cutoffs, lottery-allocated places — converge on an increase of roughly one to five IQ points per additional year of education, and the effect persists. Whatever the tests measure, education raises it, and it is the only intervention with this quality of evidence behind it.
Removing what is holding it down. Iodine deficiency in pregnancy causes large, permanent losses, and supplementation in deficient regions produces some of the biggest gains ever recorded. Lead exposure in childhood lowers scores measurably; removing lead from petrol raised them at population scale. Untreated hearing and vision problems in a child look exactly like slowness. Chronic sleep deprivation costs a substantial fraction of a standard deviation. These are not small margins, and they are the most tractable things on this list.
Deep, extended practice in a domain. It builds crystallised ability, which is what actual competence is made of. It does not transfer much to unrelated domains, and that limitation is the theme of the next item.
Brain training games. Almost certainly not. The consistent finding across dozens of trials is that you get better at the trained task, somewhat better at very similar tasks, and no better at anything else. In 2014 a large group of cognitive scientists issued a public statement saying the commercial claims outran the evidence. Working memory training in particular has been tested extensively and does not produce general gains.
Music lessons, chess, learning a language. Genuinely valuable for their own sake. The evidence for general transfer to unrelated abilities is weak.
What to do with this
Stop asking how intelligent you are. It is the wrong question and it has no useful answer. The score, if you got one, is a noisy position in a distribution measured on one afternoon, and within the normal range it will not tell you what you can accomplish.
Ask instead what you are removing and what you are accumulating. Those two are where the leverage is.
Removing means the boring physical floor: sleep in the right range, treated eyes and ears, no chronic alcohol, movement, iron and iodine and B12 at adequate levels if your diet is restricted. None of this makes you a genius. All of it prevents you from operating several points below your own ceiling, which is where a surprising number of people spend their lives.
Accumulating means the whole of Part 3. Crystallised ability is the part that keeps growing until you are sixty, it is what expertise actually consists of, and it is entirely under your control. The person who reads seriously for an hour a day for ten years has built something no test score substitutes for.
And treat the concept with care when you point it at other people. A hundred years of this measurement has been used to justify some genuinely appalling things — immigration restriction, forced sterilisation, colonial hierarchy — always by treating a group average as a fact about an individual and a snapshot as a destiny. The correct response is not to pretend the measurement does not work. It is to remember what it is: one modest, noisy, movable input, which tells you almost nothing about the person in front of you and nothing whatsoever about what they are worth.
Next: Part 2 opens the machine. 2.1 gives you a working map of the brain — what the main regions do, what the popular claims about them get wrong, and how much a coloured blob on a scan is actually telling you.