The Brain2Qwerty v2 System
Meta’s Brain2Qwerty v2 system translates brain scans into coherent sentences without requiring invasive surgery. This research aims to assist patients suffering from anarthria, locked-in syndrome, amyotrophic lateral sclerosis (ALS), and other paralyzing neurodegenerative conditions. The project opens up new possibilities for individuals who struggle to communicate due to brain damage.
About the Study
The Brain2Qwerty v2 model was trained at the Basque Center on Cognition, Brain and Language in San Sebastián, Spain, involving nine healthy volunteers aged 25 to 56. These participants typed over 2,500 sentences across ten sessions, with their brain activity monitored using magnetoencephalography (MEG). As a result, Brain2Qwerty v2 achieved a word recognition accuracy of 78%. More than half of the sentences decoded from brain activity contained no more than one word error. For comparison, the previous version of the system, Brain2Qwerty v1, reached only 48% accuracy. Decoding precision improved as the volume of training data increased.
The system uses pattern recognition technology similar to chatbots like ChatGPT and Meta’s Llama. In the initial stages, brain waves are converted into tokens representing individual characters, after which another artificial intelligence system arranges these characters into full words. Finally, a large language model transforms the set of characters and words into coherent sentences. The study’s results mark the first successful application of an LLM method to convert noisy brain activity into structured sentences.
Meta believes this research has the potential to genuinely improve the lives of millions of people affected by brain injuries that hinder communication. As the researchers noted:
“If long-term training on non-invasive MEG data eventually eliminates the need for neurosurgery in these cases, it will represent a transformative shift in patient care.”
Meta hopes that this work, conducted openly, will advance neuroscience toward faster detection, diagnosis, and treatment of neurological disorders.
Meanwhile, the American company Neurable is licensing brain-reading technology to create consumer devices that also require no surgery, unlike Neuralink, founded by Elon Musk. Neurable’s technology relies on EEG sensors, highlighting the growing interest in developing non-invasive methods for interacting with the brain.
The development of technologies like Brain2Qwerty v2 could significantly change approaches to treating and communicating with patients who have severe neurological disorders. Successful implementation of such methods may open new horizons in medicine, particularly in rehabilitation and psychological support. It also reflects increasing interest in non-invasive technologies, underscoring the importance of research in this area for improving the lives of people with disabilities.
In addition to advancements in brain-computer interfaces, recent studies reveal that even under general anesthesia, the brain continues to recognize speech and anticipate words. This intriguing finding suggests that our cognitive processes may remain active in ways previously unimagined. For more insights on this fascinating topic, check out how the brain reacts during anesthesia in our detailed article about brain activity during anesthesia.