Brain2Qwerty v2 decodes thoughts to text without surgery, hits 61% accuracy

What happened

A research team released Brain2Qwerty v2, an AI system that decodes brain activity into text using non-invasive brain recordings (magnetoencephalography), achieving a word accuracy rate of 61%, and for the best participant, 78% word accuracy. The team trained the system on approximately 22,000 sentences from nine volunteer participants, each recorded for 10 hours while wearing an MEG device and actively typing, and is releasing the full training code and datasets to accelerate neuroscience research.

Why it matters

Current brain-computer interfaces that restore communication for people with brain lesions require invasive surgery, which limits how widely they can be deployed. This non-invasive approach could help bridge that gap for millions of people who suffer from brain lesions preventing them from communicating. The findings suggest that decoding accuracy improves log-linearly with data volume, indicating the performance gap with surgical approaches could be narrowed further through data scaling alone.

What to watch

The team is releasing code for both v1 and v2, and their partner, the Basque Center on Cognition, Brain, and Language (BCBL), is releasing the v1 dataset. The research is part of broader efforts including a $5 million（約8億円） fund to stimulate open datasets in the Digital Brain Project, aimed at building foundational models of the brain for identifying, diagnosing, and treating neurological disorders.