Choosing explanation over performance: Insights from machine learning-based prediction of human intelligence from...

Area of Science:

• Cognitive neuroscience
• Neuroimaging
• Psychometrics

Background:

• Research increasingly links cognitive abilities to brain characteristics, particularly functional brain connectivity.
• Current predictive models often achieve statistical significance but lack neurobiological interpretability.
• Identifying specific predictive brain features is crucial for understanding the neural basis of cognition.

Purpose of the Study:

• To investigate which functional brain connections predict general, crystallized, and fluid intelligence.
• To emphasize interpretability in predictive modeling for enhanced conceptual understanding of human cognition.
• To explore the role of specific connectivity patterns in intelligence prediction.

Main Methods:

• A preregistered study design was employed.
• Functional brain connectivity data were analyzed from 806 healthy adults, with a replication sample of 322 adults.
• Predictive modeling focused on identifying interpretable functional brain connections related to different facets of intelligence.

Main Results:

• The specific component of intelligence (general, crystallized, fluid) and the task during connectivity measurement significantly impacted prediction.
• Intelligence was predictable from various combinations of system-wide functional brain connections, not a single set.
• Identified predictive connectivity patterns complemented existing theories on intelligence and brain regions.

Conclusions:

• Prioritizing the systematic evaluation of predictive brain characteristics enhances the explanatory value of cognitive prediction studies.
• Interpretability in predictive modeling is key to advancing our neurobiological understanding of human intelligence.
• Future research should balance prediction performance with the identification of meaningful neural correlates of cognition.