BCIwiki (bciwiki.com) — People learn to operate a noninvasive brain-computer interface (BCI) only when its control mapping follows the intrinsic geometry of their brain activity, researchers from Yale University, Universite de Montreal and Mila, Quebec Artificial Intelligence Institute reported in Nature Neuroscience on June 10, 2026 (PMID 42265352, DOI 10.1038/s41593-026-02311-2).
The authors note that BCIs could restore and enhance human capabilities, but adoption has been limited by slow and inconsistent learning across users. The team used data diffusion to extract the naturally occurring structure of brain activity, called the intrinsic manifold, which describes the directions along which activity tends to vary.
Participants were trained with real-time functional magnetic resonance imaging (fMRI) to control an avatar in a video game by self-modulating activity in brain regions that support spatial navigation. The researchers then perturbed the mapping between brain activity and avatar movement to test how neural manifolds constrain human BCI learning.
When new mappings relied on directions of significant variance on the intrinsic manifold, participants regained control by realigning their brain activity along those directions. When mappings did not follow the manifold, participants could not learn to control the avatar. The authors say the findings show how manifold geometry in higher-order brain regions guides learning of complex cognitive tasks and point to a principle for improving future neurotechnologies.