Being intelligent is about being good at many things. Intelligence researchers assign a single number to a diverse constellation of aptitudes, and this single number is the best across-the-board predictor of performance that differential psychology has ever yielded. Intuitively, there should be an ability behind this constellation: the more of this ability people have, the better they are at all types of cognitive tasks. For all practical purposes, this explanation is sufficient. The construct of ‘ability’ simply means that certain skilled behaviours go together, and we can encompass this ability under a single concept, intelligence.
Science however goes beyond the immediately obvious, and the nature of intelligence has been a matter of an undying debate in psychology for over a century. The fact that it can, in principle, be subsumed under a single number in behaviour, does not necessarily imply that it has a unitary basis. There is hope that neuroscience will be able to resolve this by picking out the neural substrate of intelligence, that single underlying feature that will correspond to an individual’s IQ.
A (cognitive) neuroscientist will immediately see that this is not straightforward. Sometimes, dissociations can be demonstrated between phenomena that routinely go together, such as speech production and language comprehension, short term memory and long term memory, or perception and awareness. At the same time, solving a mental task will activate most of the brain, though most of this activation would not be directly attributable to IQ.
A (differential) psychologist will also see that this is not straightforward. Intelligence has to do with a little bit of everything, and pooling together the correlates normally requires an intricate process of creating a variable that we don’t have access to (i.e. intelligence) in order to explain commonalities among other intangible variables (verbal skill, numerical skill, etc.). Moving that to the space of neural activity, where we directly measure our variables of interest, would involve some careful methods adjustment.
What neither the psychologist nor the neuroscientist can easily see, is that they come from different methodological traditions. One looks at differences between individuals, the other at differences between groups of events or groups of people. They use the same terminology, but their statistics, even when identical, are imbued with different meanings. Here are a few of these meanings from the intelligence literature in psychology that are (based on my impression of the neuroscience literature) not immediately obvious even though they are fundamental.
IQ is a rank order measure
A basic difficulty in looking for any IQ correlates, is that there is no objective benchmark for measuring intelligence. The score therefore denotes the position a person has relative to peers. What this means is that an IQ score is a ranking, even when expressed as the number of points on a familiar scale. And that means that the distance in real intelligence between any two points on an IQ scale might be different from the same distance between any other two points.
To illustrate how this could complicate looking for a neural correlate, let’s imagine the following scenario. I share an office with four other people, and there is a clear preference for certain desks. Being nerds, we decide to allocate desks based on an intelligence test – the highest ranking person gets to choose first, the second one second, etc. For purposes of fairness, we decide to re-do this exercise every year. I am very happy with the outcome, because out of the five of us, I come second.
But then, my life circumstances change. My grant applications get rejected, my experiments fall through, and my husband leaves me for a social constructivist. I begin to do drugs. A year later, my colleagues get roughly the same number of questions correctly, but I am so cognitively impaired that I come last. For me, the change in ranking is a logical consequence of my declining mental ability. But the three colleagues who used to be below me now get a free ride: their IQ improved (because their rankings went up), but their intelligence didn’t. If we were to try to connect these changes in our IQ (as a group) to changes in our neural activity from one year to the next, what could we conclude? Not much.
IQ is a measure of variance
Behaviorally, psychologists have mostly been investigating what makes people differ among themselves in intelligence. This property, of capturing individual differences rather than the expression of the trait itself, is trickier than it seems at first sight.
Take the finding that the IQ of adopted children correlates with the IQ of their biological, but not adoptive parents. Does this mean that adoptive children’s IQ is closer to their biological parents’? It’s easy to make this assumption, but it doesn’t. It means that when we rank children by their IQ, this ranking can be used to predict the ranking of their biological parents (among the group of biological parents), but not of their adoptive parents (among the group of adoptive parents). Adoption, which usually implies relocating to a better environment for development, enhances IQ. But it does so in a systematic fashion: we can imagine that we’ve added an equal bonus to the IQ points of every adopted child. Thus, even though the IQ of adopted children as a group might come closer to the IQ of their adoptive parents, their rankings will stay unperturbed, so the correlations won’t change. Correlation here implies variation, and leaves the commonalities in intelligence unexplored.
Accordingly, looking at which brain feature varies between individuals in relation to their test performance answers a different question (what neural feature is related to IQ?) from looking at the neural difference between a mentally challenging and a baseline dull task (which areas are most active during mental tasks?). It might make sense to correlate a neural measure of commonalities with a behavioural measure of differences, but then again, it might not be the right place to look.
Intelligence develops over time, IQ does not
We can imagine a personal, developmental trajectory of intelligence to begin at some real zero at conception, then grow towards and past other species’ intelligence (while undergoing qualitative changes), and finally find its place in the cloud of human interpersonal variance. Given this developmental trajectory from no intelligence to a particular amount of intelligence, the line between the (shared) species-specific and individual-specific parts of intelligence can be fuzzy at times. We are in fact not good at assessing intelligence that is too far below the line, as nearly no correlates of future IQ are found in healthy children below the age of three.
Environmental influences can systematically enhance or diminish intelligence. This means that the dividing line between shared and personal intelligence might be shifted for entire populations. If we want to search for neural correlates of intelligence, then, we would want to know whether these shifts always entail only a quantitative change, or, alternatively, a different quality of intelligence in certain populations. The IQ scale is only quantitative, test scores don’t tell us if people in certain parts of the IQ range simply have more or less intelligence, or if they have a different kind of intelligence. These two scenarios would lead to different patterns of IQ-correlation with brain features.
So what are we looking for?
For me to call something the neural substrate of intelligence (as measured by IQ), it would have to fulfil three (difficult!) prerequisites.
First, this neural activity or structural feature should be implicated in all tasks present in intelligence tests (digit span, picture completion, verbal comprehension, mental rotation, matrix reasoning, searching for similarities, all of it)
Second, it should be systematically more (or less) pronounced when comparing healthy people of different levels within the normal range of intelligence.
And third, it should be systematically more (or less) pronounced within one person, when comparing types of tasks of varying difficulty for that person.
Or maybe we are looking for something different when we look for intelligence in the brain? Maybe we’re not really looking for IQ, but rather for some universal, shared information processing capacity? That’s perfectly fine, as long as we define what it is, and how it differs from the more standard notions of intelligence as expressed by IQ. The psychologists’ definition of intelligence as a single (Spearman), dichotic (Cattell & Horn), triarchic (Sternberg) or multiplex trait (Gardner) that involves the capacity to adapt to the environment (Binet; Wechsler), or to solve problems (Gardner; Anderson; Bingham), vary multiple items in working memory (Kyllonen & Christal), grasp new knowledge (Humphreys; Dearborn), think abstractly (Thurston; Terman), perceive order (Young) and so on, certainly leaves something to be desired.
Neuroscience here faces the same problem as psychology. We need to quantify an intuitive concept, intelligence. In this process something is gained – the possibility of controlled experimentation and statistical testing – but some essence of the thing we measured is inevitably lost as well. Then we have to turn this quantified and statistically manipulated set of variables back into our question of interest, and match it up with a reasonable answer. So on the one hand it’s important to unpack the consequences of quantification, but on the other hand it’s important to match our inferences to the data rather than to our intuitions. As much as the psychology literature can be annoying with its endless theorising, I often miss this kind of thinking when we discuss the meaning of our neural data.
I still think that it doesn’t matter much in this case, because I think the search for the neural substrate of intelligence is guided by unrealistic expectations. Stay tuned for the next post, where I’ll tell you why.