AI Unlocks the Brain’s Intelligence Pathways

Scientists are leveraging artificial intelligence to unravel the brain’s intricate intelligence pathways.

Glenn Burgess

A new study in PNAS Nexus shows that artificial intelligence can guess different types of human intelligence by studying brain connections.

Scientists used brain scans from hundreds of healthy adults and found that AI was best at predicting general intelligence, followed by crystallised intelligence, and then fluid intelligence.

This study reveals that intelligence comes from the way brain networks work together, not just from specific parts of the brain.

Previous research has shown that intelligence is not tied to a single part of the brain but depends on networks spread across the brain.

However, many studies have used older methods that focus on individual brain features, providing only limited understanding of how brain structure and function work together to create intelligence.

In this study, scientists used machine learning to study brain connectivity and address these gaps. They focused on three main types of intelligence: general, fluid, and crystallised.

General intelligence, or “g,” is a broad measure of thinking ability, including reasoning, problem-solving, and learning in different situations. It acts as a common factor that links various cognitive skills.

Understanding Fluid and Crystallised Intelligence: How the Brain Connects

Fluid intelligence, a part of general intelligence, is the ability to reason and solve new problems without using past knowledge or experience. It involves abstract thinking, recognising patterns, and adapting to new situations.

On the other hand, crystallised intelligence is the ability to use what you’ve learned through education, culture, and life experiences, such as vocabulary, reading, and factual knowledge.

“Our goal was to explore how differences in general cognitive ability are reflected in the brain,” explained Kirsten Hilger, the study’s lead author and head of the Networks of Behavior and Cognition group at Julius-Maximilians-Universität Würzburg. “We believe the connections between brain regions, which represent communication pathways, play a key role in intelligence.”

Recent studies have used brain connectivity, or communication pathways between brain regions, to predict differences in intelligence.

However, these studies have often focused on achieving the best prediction accuracy, offering little insight into how intelligence actually develops from these connections.

“Our study aims to fill this gap by using methods that not only predict intelligence but also help us understand how intelligence arises from the brain,” said Kirsten Hilger.

The researchers used data from the Human Connectome Project, analysing 806 participants aged 22 to 37, all without cognitive impairments.

They studied brain connectivity through functional MRI (fMRI) scans during rest & tasks that involved working memory, language & emotion recognition.

Fluid intelligence was tested with tasks unrelated to prior knowledge, while crystallised intelligence was assessed through vocabulary & reading. General intelligence was calculated by combining both types into a single score.

The researchers used machine learning to analyse connections among 100 brain regions across various cognitive states. They compared models based on established intelligence theories with those using random connections.

A technique called relevance propagation identified which connections most influenced the predictions. General intelligence was predicted most accurately, suggesting it is linked to consistent brain connectivity patterns.

Crystallised intelligence predictions were also strong, while fluid intelligence was less precise. Brain activity during tasks, such as working memory or language processing, provided better predictions for general and fluid intelligence than resting-state activity.

Crystallised intelligence relied more on stable, task-independent networks, enhanced by integrating data from multiple brain states.

Models based on established theories, like the parieto-frontal integration theory, outperformed random models, highlighting the importance of prefrontal and parietal networks.

However, whole-brain models consistently achieved better results, suggesting intelligence arises from distributed brain networks rather than isolated regions.

The study identified about 1,000 key brain connections distributed across major networks, including the default mode, frontoparietal control, and attention networks.

It also found the brain’s ability to compensate for missing connections remarkably high, with minimal impact on predictions when entire networks were excluded.

“Our findings show that intelligence emerges from global brain communication rather than specific regions,” said Kirsten Hilger.

However, study’s focus on adults aged 22–37 limits its applicability across age groups. Further research into how specific brain connections support processes like memory & attention deepen understanding of intelligence.

The researchers concluded that their work provides a foundation for future studies, emphasising meaningful insights into human traits over purely predictive accuracy.

Conclusion

The study concludes that intelligence is a global property of the brain, arising from distributed networks rather than isolated regions.

By using advanced machine learning techniques, the researchers demonstrated that general, fluid, and crystallised intelligence are linked to patterns of brain connectivity, with general intelligence being the most consistently predicted.

Their findings highlight the importance of communication across the whole brain, involving major networks like the default mode, frontoparietal control, and attention networks.

The research also revealed the brain’s remarkable redundancy, showing minimal disruption to intelligence predictions even when large-scale networks were excluded. This underscores the flexibility and adaptability of neural systems supporting intelligence.

While the study offers valuable insights, it acknowledges limitations, such as the narrow age range of participants and the unclear functional roles of key brain connections.

Future research could explore how these connections support specific cognitive processes and individual differences in problem-solving and knowledge application.

Overall, the findings provide a comprehensive framework for understanding intelligence as a dynamic and distributed brain function, offering a pathway for more meaningful predictive models in neuroscience.