What happens inside your brain when you go from knowing the rules of chess to playing at a master level? The question extends far beyond the 64 squares, it gets at the fundamental neuroscience of expertise itself.
A team at KU Leuven led by Andrea I. Costantino and Hans Op de Beeck has taken a rare, direct look at this transformation. They scanned the brains of 40 adults, 20 expert chess players (mean Elo rating 2036) and 20 novices, using 3T fMRI while the participants performed a 1-back task with 40 chess positions (20 checkmate, 20 non-checkmate). By combining multivariate pattern analysis, representational similarity analysis (RSA), and dimensionality estimates, they identified three distinct principles that characterize the expert brain.
The study, published in Nature Communications on 26 June 2026, shows that expertise is not merely a matter of knowing more, it is a fundamental reorganization of how information is encoded in the brain.
From visual features to relational content
The first principle is a shift in what information the brain represents. Novices, when looking at a chess position, primarily encode visual surface features, the shapes of pieces, their positions on the board, the pattern of light and dark squares. Experts, by contrast, encode high-level relational information: which pieces are attacking which, what tactical themes are present, how the position connects to known patterns.
This is not simply a matter of “seeing more.” The neural representations themselves change. The visual features that dominate the novice brain become background noise for the expert, whose neural activity preferentially encodes the abstract structure of the position.
Compressed but not coarsened
The second principle is a structural optimization. Expert representations are lower-dimensional, more compact, better organized, more efficient, yet they retain the precision needed for accurate evaluation.
“Packing more into less” is how the authors describe it. The brain achieves this by selectively pruning redundant information while preserving decision-critical detail. Using participation-ratio estimates, a manifold-based measure of how many dimensions a neural population’s activity occupies, the researchers found reduced dimensionality across 14 of 22 cortical regions of interest. Despite this compression, the information needed for precise evaluation was preserved: decoding and linear separability analyses showed that the compact codes still carried the fine-grained distinctions needed for accurate chess judgment.
This challenges the intuitive notion that expertise simply means having more neural resources dedicated to the task. The expert brain does not necessarily activate more neurons or more regions. It organizes what it has more efficiently, creating low-dimensional codes that capture the essential relational structure of the problem domain.
The migration to frontoparietal networks
The third principle is anatomical. In novices, the representational load is carried by sensory-specific cortices, visual regions that process what the board looks like. In experts, this load shifts to domain-general frontoparietal networks, the same regions involved in high-level reasoning, planning, and cognitive control across many domains.
This is consistent with a broader picture emerging from expertise research: that the brain does not develop a specialized “chess region” but rather recruits general-purpose cognitive machinery and adapts it to the specific demands of the domain. The frontoparietal network, already implicated in working memory, attention, and abstract reasoning, proves to be the substrate into which expertise is built.
“The expert brain packs more into less,” the authors write, “concentrating richer knowledge into fewer, better-organized representations that support the rapid, flexible decisions of mastery.”
Why it matters
The findings extend beyond chess. If the same principles hold across domains, music, surgery, programming, sports, they suggest a universal neural grammar of expertise: content shifts from surface to structure, codes compress without losing precision, and processing migrates from sensory regions to domain-general networks.
This has implications for training and education. If expertise is about restructuring representations rather than simply accumulating facts, then effective training should focus on helping learners build the right representational structure, not just exposing them to more examples.
Costantino shared the paper on Bluesky upon publication, calling attention to the work from the Department of Brain and Cognition at KU Leuven, with co-authors including Artem Platonov, Felipe Fontana Vieira, Emily Van Hove, Merim Bilalić (Northumbria University), and Hans Op de Beeck. The research was supported by the Research Foundation Flanders and the Flemish Government.
Sources
1. Costantino, A.I., Platonov, A., Fontana Vieira, F. et al., “Low-dimensional and optimized representations of high-level information in the expert brain”, Nature Communications (2026). DOI: 10.1038/s41467-026-74566-z
2. Andrea Costantino on Bluesky: https://bsky.app/profile/costantinoai.bsky.social/post/3mq3itmvxts2l