Meta has unveiled a new system called Brain2Qwerty v2, which aims to directly convert human brain activity into text without requiring brain surgery. The company states that this research primarily targets individuals who have lost the ability to communicate due to brain injuries.
Record brain signals using MEG
This system relies on a helmet-like brain imaging device called a magnetoencephalography (MEG) scanner. It first records brain signals generated by subjects while typing, then feeds the raw neural signals into an end-to-end AI model to reconstruct the sentences the user intended to input.
Meta stated that the model also incorporates the semantic capabilities of large language models to enhance recognition accuracy even when brain signals are noisy, by leveraging contextual information.
Average word accuracy increased to 61%.
According to Meta’s disclosed data, Brain2Qwerty v2 was trained on data from nine volunteers. Each participant wore an MEG device and actively typed for approximately 10 hours, resulting in a total of about 22,000 training sentences.
Meta stated that the system achieves an average word accuracy rate of 61%, significantly higher than the approximately 8% level of previous non-invasive methods. The company also noted that decoding accuracy continues to improve as training data increases, indicating further potential for optimization in this approach.
- Approximately 22,000 training sentences
- There were 9 participants in total.
- The accuracy of previous non-invasive methods was approximately 8%.
Open-source code and research data synchronization
Meta stated that it will release the training code for Brain2Qwerty v1 and v2, and its research partners will also publish the v1 dataset. This work is part of Meta’s Digital Brain Project, which includes a $5 million fund to support the development of open neuroscience datasets.
Meta researchers, in a companion paper published in Nature Neuroscience, noted that current high-performance brain-computer interfaces still largely rely on implanted electrodes, making large-scale adoption difficult due to surgical risks and long-term maintenance challenges of implanted devices.
The competition in brain-computer interfaces continues to intensify
Meta believes that the accuracy of this non-invasive solution has approached levels previously achievable only with implantable technologies, potentially narrowing the gap between invasive neural prosthetics and non-surgical communication systems.
This release also comes amid growing momentum in brain-computer interface research. Elon Musk’s Neuralink and Merge Labs, backed by OpenAI CEO Sam Altman, are both advancing technologies aimed at restoring communication abilities for patients with neurological disorders.
Meanwhile, more research teams and startups are attempting to use AI to enhance the performance of non-invasive systems. Previously, Neurable launched an AI-powered EEG headset capable of monitoring focus and cognitive fatigue; MIT spin-off AlterEgo also released a wearable device that converts subtle neuromuscular signals from the face and throat into text and commands.