AI turns thoughts into text without brain surgery

Researchers from Meta AI, PSL University, and the Fondation Adolphe de Rothschild Hospital have developed a system that converts brain signals into written text without surgery. Published in Nature Neuroscience, the technology combines deep learning with non-invasive EEG and MEG scans. In tests with 35 volunteers, the Brain2Qwerty algorithm decoded characters from brain activity, with MEG achieving a 29% average error rate versus 65% for EEG. The system successfully decoded sentences not in its training data. Researchers aim to expand Brain2Qwerty beyond text input to control prosthetics and digital devices, narrowing the gap between safe non-invasive methods and surgical implants.