Summary: For years, the “London Taxi Driver” studies suggested that elite navigators have physically larger brain regions. However, new research suggests this might not be the case for the general population.
Using advanced Deep Learning and Convolutional Neural Networks, researchers analyzed MRI scans of healthy young adults to see if AI could find subtle structural patterns that predict navigation ability. Surprisingly, even the most sophisticated AI tools found no measurable connection between the shape or size of brain regions—including the hippocampus—and how well a person finds their way.
Key Facts
- AI vs. Tradition: While simple size measurements once suggested a link, deep learning models capable of detecting complex structural patterns failed to find a “navigation signal” in healthy young adults.
- Hippocampus & Thalamus: The study specifically compared the hippocampus (the brain’s traditional “GPS”) and the thalamus (a control region). Navigation performance was indistinguishable between the two based on structure alone.
- Healthy Young Adults: The data was gathered from 90 participants (average age 23) who learned routes in a virtual environment.
- Predicting Disease vs. Behavior: Researchers noted that while AI is excellent at predicting disease states (like Alzheimer’s), it struggles to map everyday behavioral functions like spatial navigation.
- Redefining Research: The findings challenge the idea that “more real estate” in the brain equals better cognitive performance, suggesting that function and connectivity may be more important than macroscopic structure.
Source: UT Arlington
Steven Weisberg, a researcher at The University of Texas at Arlington, found that advanced artificial intelligence tools could not uncover a clear link between brain structure and navigation ability in healthy young adults—challenging long-standing ideas about how the brain helps us find our way.
For decades, many in the scientific community believed that people with elite navigation skills—such as quickly learning and recalling complex routes—might have larger or differently shaped brain regions than others.
Famous studies of London taxi drivers, for example, suggested that intense navigation training could lead to more “real estate” in certain parts of the brain.
In the new study, Dr. Weisberg and his team, including University of Florida Ph.D. candidate Ashish Sahoo, tested those assumptions using newer analytic techniques, including deep convolutional neural networks and other machine-learning models capable of detecting subtle patterns in brain scans beyond simple size measurements.
Despite these advanced methods, the researchers found no measurable connection between brain structure and navigation performance in healthy young adults.
Understanding navigation is important given its real-world implications for daily life, including independence, memory, and dementia risk.
“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,” said Weisberg, who conducted the study at the University of Florida before joining UT Arlington last fall as part of the RISE 100 initiative.
The study, published in the peer-reviewed journal Neuropsychologia, analyzed data from 90 participants with an average age of 23.1 years. Participants learned two routes using a virtual environment.
Results showed little difference in navigation performance when comparing two brain regions: the thalamus, which served as the control region, and the hippocampus, a region traditionally linked to navigation and memory.
While the findings point to limits in what AI can currently reveal about everyday cognitive skills, the technology remains a powerful research tool. Weisberg said more robust models could detect differences in future studies.
“Our study should be one data point in a larger landscape of what AI can tell us about how brain structure and function map onto behavior,” Weisberg said.
“Machine learning and AI have been pretty successful at predicting disease states. What we’re interested in is whether these models have utility for behavioral function—things like cognitive training or education.”
Future research will focus on larger samples and older populations, Weisberg said.
“Our ability to get around enables basically everything we do. Studying how the brain supports navigation helps us understand what is needed when it goes well and what is lacking when it doesn’t.”
Key Questions Answered:
A: Not necessarily. Intensive, years-long training (like that of London taxi drivers) might still cause structural changes. However, for the average healthy person, being a “good” or “bad” navigator isn’t written in the physical shape of your brain. It’s likely more about how your neurons fire together, not how much space they take up.
A: Physically, at a macroscopic level, yes. This study suggests that “elite” navigation skills in young adults don’t require a bigger hippocampus. The difference is likely “under the hood” in the microscopic wiring or the chemical signals that AI-processed MRI scans can’t yet detect.
A: Navigation is often the first skill to decline in Alzheimer’s. If we can’t find a structural “baseline” for good navigation in healthy people, it suggests that early diagnosis should focus on behavioral changes and functional connectivity rather than just looking at the size of the hippocampus on a scan.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this AI and neuroscience research news
Author: Drew Davison
Source: UT Arlington
Contact: Drew Davison – UT Arlington
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Deep learning approaches to map individual differences in macroscopic neural structure with variations in spatial navigation behavior” by Ashish K. Sahoo, Hajymyrat Geldimuradov, Kaleb E. Smith, Aaron Zygala, Yiming Cui, Mahsa Lotfollahi, Kuang Gong, Alina Zare, and Steven M. Weisberg. Neuropsychologia
DOI:10.1016/j.neuropsychologia.2025.109352
Abstract
Deep learning approaches to map individual differences in macroscopic neural structure with variations in spatial navigation behavior
Understanding the association between structural properties of the human brain and individual differences in behavior is an ongoing endeavor, challenged by the brain’s complexity.
Past approaches, limited by simplistic neural structure measures like brain volume or cortical thickness, have given way to more advanced modeling approaches.
Empirical evidence using these simpler metrics occasionally shows that hippocampal structure relates to individual variation in spatial navigation ability, particularly for older individuals or for expert navigators (like London taxi drivers). Yet high-powered, pre-registered studies in typical younger adults revealed no association between hippocampal volume and navigation ability.
Here, we follow a data-driven approach developing and comparing deep learning methods (graph convolution neural networks, GCNN; 3DCNN) to analyze whether complex aspects of brain structure predict spatial navigation ability in young populations.
To that end, we trained GCNNs and 3DCNNs on a T1 MRI dataset (N = 90) to predict navigational ability as measured by an objective virtual reality test of spatial memory in which participants created as accurate a map as they could of a highly realistic virtual environment.
Across all approaches, we found weak predictive value in held-out test data, despite good fits to training data.
These results could indicate the need for much larger datasets, including more comprehensive behavioral measures (as this study was limited to one measure) to improve predictability but may also support the notion that hippocampal structural features may not be a primary factor associated with navigation ability in healthy younger adults.
