Can brain activity during hard tasks predict general intelligence?
Brain activation patterns recorded while people did a demanding working-memory task predicted their general cognitive ability about twice as well as resting brain scans did.
Source
Toward a "treadmill test" for cognition: Improved prediction of general cognitive ability from the task activated brain
Study at a glance
- Design
- Cross-sectional — Cross-validated predictive modelling of a general-ability factor from 15 task-fMRI contrast maps (and resting-state connectomes) in one large dataset.
- N
- N=967 · 967 adults in the task-fMRI prediction analysis; 1,192 for building the ability factor; 903 for the resting-state comparison.
- Population
- Healthy young adults from the Human Connectome Project 1200 release
- Outcome
- Cross-validated correlation between predicted and actual general cognitive ability
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
Using Human Connectome Project data, the researchers built a general cognitive ability score from 10 cognitive tests, then trained models to predict that score from whole-brain activation maps for 15 contrasts across seven scanner tasks (such as working memory, relational reasoning and language). They used 10-fold cross-validation, keeping family members in the same fold, and compared the task-based results with the same model applied to resting-state connectivity. Finally they asked whether a task's ability to predict depended on how strongly it activated the frontoparietal network and deactivated the default mode network.
What they found
The 2-back versus 0-back working-memory contrast gave the best prediction, correlating 0.50 with actual ability, whereas resting-state connectomes reached only r = .26; 13 of the 15 task contrasts outperformed rest. Tasks with more executive demand (working memory, relational reasoning, math versus story) did best. Across contrasts, mean frontoparietal activation strongly predicted how useful a task was (r = .68), and frontoparietal activation plus default-mode deactivation together tracked prediction accuracy closely (r = .82).
The limits
What it doesn't show
Even the best model explained only 28% of the variance in ability, so most individual differences remain unexplained. The tasks were those HCP happened to include, not ones designed to predict ability, and the sample is healthy young adults, so it may not generalise to children, older adults or patients. Prediction is correlational: a pattern that predicts ability does not show which regions cause it, and the differences between the top executive tasks were modest.
Key terms
- General cognitive ability (GCA)
- A single factor capturing what performance on many different cognitive tests has in common, similar to 'g'.
- Cross-validation
- Testing a model on people it was not trained on, by repeatedly splitting the data into training and test portions, to guard against overfitting.
- Frontoparietal network
- Lateral prefrontal and parietal regions that activate across many demanding tasks and support control and working memory.
- Default mode network
- Regions such as posterior cingulate and medial prefrontal cortex that are typically more active at rest and deactivate during demanding external tasks.
- N-back task
- A working-memory task in which people say whether the current item matches the one shown n items earlier.
Flashcards
0 of 10 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
Which contrast best predicted general cognitive ability?
Common questions
Why call it a 'treadmill test'?
Like a cardiac stress test, the idea is that putting the brain under cognitive load reveals individual differences that are hidden when it is at rest.
Does a 0.50 correlation mean the scan can measure someone's IQ?
No. It explains roughly a quarter of the variance, which is strong for brain-behaviour research but far too imprecise to replace a cognitive test for an individual.
Why does deactivating the default mode network matter?
Default-mode suppression tends to accompany demanding tasks, and tasks that produced more of it together with frontoparietal activation were the ones whose brain maps best predicted ability.
More on Default mode