AI tests the map in our heads: No clear structural signal for navigation in young adults
Published: Monday, Mar 02, 2026
Steven Weisberg, a psychology researcher at The University of Texas at Arlington, reports that advanced AI models could not detect a reliable link between brain structure and navigation ability in healthy young adults. The finding challenges a common assumption that people who are great with routes have larger or differently shaped brain areas.
Why this matters
Navigation supports daily independence, memory, and can reflect early changes relevant to dementia risk. Past work, including famous studies of London taxi drivers, suggested that heavy navigation experience might reshape parts of the brain, especially the hippocampus. See one classic example from PNAS for context: Hippocampal differences in taxi drivers.
How the study worked
The study, published in the peer-reviewed journal Neuropsychologia, analyzed 90 participants (average age 23.1). Participants learned two routes in a virtual environment while researchers measured navigation performance.
On the imaging side, the team applied deep convolutional neural networks and other machine-learning models to MRI data, going beyond simple volume measurements. They compared patterns linked to two regions: the hippocampus (traditionally tied to navigation and memory) and the thalamus (control region).
What the models found
No detectable relationship emerged between brain structure and navigation performance in this young, healthy cohort. Performance differences were minimal when comparing models centered on the hippocampus versus the thalamus.
As Weisberg put it: "With the quality of data we have from MRI scans and this healthy young adult population, there does not appear to be a detectable signal using these advanced metrics." The takeaway isn't that AI is ineffective, but that structural MRI, as used here, may have limited sensitivity for individual differences in everyday navigation skill.
Read the paper
The work appears in Neuropsychologia. For journal context: Neuropsychologia on ScienceDirect.
Practical takeaways for scientists and research teams
- Expect weak structure-behavior signals for healthy young adults using structural MRI alone; effect sizes may be small or absent.
- Prioritize larger, more diverse samples and cross-site datasets to improve sensitivity and generalization.
- Consider multimodal approaches (e.g., functional MRI, diffusion MRI, behavioral logs, longitudinal data) to increase signal-to-noise.
- Treat deep models as hypothesis generators; pair them with clear preregistration and rigorous validation to avoid overfitting.
- Benchmark against biologically informed baselines (region-of-interest, classical morphometry) and report negative results transparently.
- Focus on populations where structural differences are more likely (older adults, clinical risk groups) for stronger tests of brain-behavior links.
- Invest in task design quality: align behavioral metrics with specific cognitive mechanisms you expect anatomy to support.
What's next
Weisberg's team plans to scale samples and study older populations, where anatomical differences may be more pronounced. They also note that future models or modalities could surface meaningful patterns that current tools miss.
"Our ability to get around enables basically everything we do," Weisberg said. "Studying how the brain supports navigation helps us understand what is needed when it goes well and what is lacking when it doesn't."
Related resource
Building AI skills for lab and field work? Explore AI for Science & Research for workflows, models, and evaluation practices relevant to academic studies.
Your membership also unlocks:
