Should we expect to find intelligence in the brain?

I previously discussed why our search for intelligence in the brain bypasses some critical features of how IQ is measured, thus leading to data which add up to a theoretically confusing picture. It is possible that a clear-cut neural correlate of intelligence will be found some day. It is also possible that intelligence is a concept that has meaning in interpersonal space, but that the neural activity in its basis does not fall into a coherent pattern – at least not in a way that will make us go aha!

The biggest hint that this second alternative might be the case, comes from genetics.

Intelligence was decisively linked to an inborn capacity back when we had only vague ideas about DNA. Let me pick out the most striking example: the difference in IQ between two identical twins is the same as the difference between two IQ tests that the same person does at two points in time. So when behavioural genetics came to life, it was natural to start a furious search for the genes underlying intelligence.

And yet, so far, we have found so little that it could just as easily be summarized as nothing. Here and there, a gene crops up in more than one study. It ends up explaining something like 0.1% of the variance in IQ test results. Most genes show up only once, never to be found again. The exception is a group of (currently) about 700 genes that are linked to retardation. These might have the power to trigger a developmental cascade of deterioration, leading to substantial changes in intelligence. Within the normal range, that most of these impressive twin studies are based on, we find virtually nothing. Nothing!

Why would this be? My favourite speculation is that this is an accurate picture: many genes contribute a tiny bit, but the big picture emerges only when we surface to the level of the person doing a test. In fact, the argument in this paper appeals to the brain, saying that many thousands of genes code for proteins that in one way or another contribute to brain activity. Put together the added detriment of a number of point mutations within those genes, and you’ll end up with a nice distribution of interpersonal differences in brain efficiency. Within the normal range of intelligence, the effect of each individual gene would therefore be minuscule, while the cumulative effect can still be profound.

Does this genetic heterogeneity have to do with the elusiveness of intelligence? Not necessarily. On a smaller scale, the same is true of height, a tangible feature. Height is a highly heritable trait, many genes have been linked to it, very few in a way that really matters on its own – except the handful linked to conditions such as dwarfism.

I believe that we should expect a similar finding in terms of neural correlates of intelligence – that we will not find The Correlate, but that instead, with large enough samples, everything we look at should be connected to some aspect of intelligence a little bit. Small contributions of grey matter density, white matter integrity, functional connectivity, localized responsivity, all of it.

Well, maybe not all. A famous neuroscientific study on the location of intelligence in the brain, which brought us the parieto-frontal integration model, took a look at neural activity across 37 different studies involving tasks that either represent IQ subtests or are assumed to correlate with general intelligence. The authors translated the outcomes of these diverse PET and fMRI studies into a comparable set of coordinates, the space of Brodmann maps. The idea was to find the areas that were repeatedly involved in intelligence tasks, to say where in the brain intelligence is. While the paper represents a colossal effort to pool together heterogeneous research, the step from the data to the conclusion (that intelligence is mainly a function of parietal and frontal areas, and the white matter pathways connecting them) left me uneasy. How do you decide where to draw the line – how many studies need to show activation for a brain area to be declared the neural substrate of intelligence? (The answer is eleven.) In addition to that, nearly a third of the studies showed no task-related activation in these areas, and the fMRI studies showed – to my mind – troublingly little overlap in anatomical space with the PET studies.

But my inner neuroscientist said ooooh, data! I still thought the cutoff was too hypothesis-driven, and I wanted to see which brain areas were involved in performing these tasks at all. So I printed a Brodmann map, started going through the 37 individual experiments, and crossing out brain areas whenever they would be involved in task performance.

And by the time I was done, I was all sad for the primary olfactory cortex, because that’s the only little guy that didn’t make the cut.

Does this mean that the entire brain is involved in intelligence? It doesn’t. I don’t think that these tasks, taken individually, represent intelligence well enough – and I certainly don’t think they can be tallied up like that. (And Brodmann maps largely ignore the existence of sulci, but never mind that.) But still, I would not be surprised if one day we end up concluding that some complex index of overall brain efficiency is the closest we will come to the neural underpinnings of intelligence.

Looking for general intelligence in the brain is like looking for general health in the body. Yes, it’s there. And yes, there are a few focal points which, if things go wrong, cause a lot of predictable damage. But still, the concept of general health encompasses the combined working of all the parts, and if we wanted to compare two mostly healthy people, some structured questions about how they feel are probably more useful than poring over a full-body MRI scan. Just like giving them an IQ test is a more useful measure of intelligence than looking at their brains. 

The real question is – what do we want to find out? The explanatory value of describing intelligence on the neural level differs from describing it on a cognitive one, but we often implicitly hope to unlock a cognitive truth with neural data. Should we hope for it here?

If intelligence, something so salient at the interpersonal level, can indeed be reduced to a complex but dispersed measure of brain efficiency, that’s a fascinating finding as far as I’m concerned. And finding any of these neural correlates is – of course – a big deal. They are likely to be more structured than the genetic correlates, because they too, just like intelligence, will reflect a downstream effect of many genes. But if we hope to describe the cognitive notion of intelligence better by looking at brains, or to say what neural feature differs among people of different IQ, or to perhaps stall the reduction in intelligence near the end of life, I have my doubts that this search is going to lead us there. For that, we would have to find the neural substrate, the feature, the connection that will contain it all. I’m not convinced that it’s there to be found.

Note: I can’t remember whether I saw the direct comparison with general health somewhere, or whether it was a thought inspired by the comparison to physical fitness I found here. I hope I’m not blatantly stealing ideas.

3 Comments

  1. Ben

    Great post. Perhaps IQ is best thought of as a measure of global neurocognitive function? It integrates various cognitive abilities and their neural networks to enable an organism to solve problems in different situations.

  2. Need to update to include results from Okbay et al (2016) and related papers since using its Education Attainment (phase 2) polygenic score. Rather than zero loci, there are 74 (and many more in the original full dataset), replicating in other datasets.

    Okbay, A., Beauchamp, J. P., Fontana, M. A., Lee, J. J., Pers, T. H., Rietveld, C. A., . . . Benjamin, D. J. (2016). Genome-wide association study identifies 74 loci associated with educational attainment. Nature, 533(7604), 539-542. doi:10.1038/nature17671

    • Thanks for the input! I have to confess, this text is based on a talk adapted to a group of neuroscientists, that I gave back in 2011. I’m ignorant of most of the literature since then.

      If I understand you correctly, there are now numerous genes identified. I think that fits well with the watershed model (I guess it would fit better if it were thousands of genes, but supposedly better methods will identify more and more?), and less well with the idea of a unitary g factor. I still think it wouldn’t point to a unitary neural basis. Or am I misunderstanding your point?

      I’ll take a look at the paper in any case, I’m intrigued now.

Leave a Reply

Your email address will not be published. Required fields are marked *