Meta Unveils Non-Invasive Brain-Reading Device Translating Thoughts to Text with 78% Accuracy

Meta has introduced Brain2Qwerty v2, an advanced brain-computer interface (BCI) system that translates brain activity into written sentences using only external sensors, without surgical intervention. The system employs magnetoencephalography (MEG) technology to detect tiny magnetic fields generated by neural activity and combines deep learning with advanced language models to decode whole sentences from raw brain signals in real time. On average, Brain2Qwerty v2 achieves 61% word recognition accuracy, with the best participant reaching 78%. Researchers highlight that over half of the tested sentences contained at most one word error, marking a significant breakthrough in non-invasive brain decoding.

Unlike its predecessor, which focused on identifying individual characters, the new version deciphers words and meaning directly. The system was trained on approximately 22,000 sentences collected from nine participants, each undergoing about 10 hours of brain activity recording while typing. Meta credits the success to integrating noisy brain signals with large language models that fill in gaps to produce coherent text. Performance continues to improve as more data is gathered, indicating further potential advancements.

This technology offers a safer alternative to invasive brain implants, which have previously achieved high accuracy but carry medical risks. Brain2Qwerty v2 could transform communication for individuals with brain injuries, paralysis, or neurological diseases by enabling them to "speak" through thought alone. Meta emphasizes this is an initial step toward developing practical tools for such patients.

In a move to promote transparency, Meta released the system's training code, while its research partner, the Basque Center on Cognition, Brain and Language (BCBL), published the dataset from the first version. The study was published in Nature Neuroscience. Public reaction has been mixed, with excitement about the scientific and medical potential tempered by privacy concerns, especially given Meta's business model centered on data collection and analysis. The technology raises profound questions about who can access brain-derived information and under what conditions, even as it remains distant from everyday use.