On June 29, Meta's research division introduced the Brain2Qwerty v2 system, capable of reconstructing typed text based on brain activity recordings without surgical intervention. This is a significant breakthrough in the field of neural interfaces: the average word-level recognition accuracy reached 61%, while previous non-invasive methods demonstrated results of around 8%.

The model was trained on approximately 22,000 sentences. Nine healthy volunteers participated in the experiment, each spending 10 hours in a magnetoencephalography (MEG) scanner while simultaneously typing heard phrases on a keyboard. Instead of manual neural event extraction, Brain2Qwerty v2 uses end-to-end deep learning to decode data directly from raw brain signals.

The system works with continuous brain activity recordings and reconstructs the entire sentence. A fine-tuned language model that accounts for semantic context is used to refine the result. The best participant demonstrated 78% word-level accuracy, with more than half of their sentences decoded with one error or none at all.

Limitations and Prospects

At the current stage, the development is not ready for clinical or everyday use. Brain2Qwerty v2 was tested on healthy volunteers who were actually typing text. For people who have lost speech or the ability to move, this approach requires separate validation. Additionally, the system does not yet function as full-fledged real-time text input: the model processes the entire sentence, so the user does not see each word immediately after typing it.

The project is part of the Digital Brain Project, an initiative to create open neuroscience datasets with a budget of $5 million. Previously, in September 2022, Meta published research on a "brain decoder" that converted thoughts into speech using AI.

My expert opinion: The breakthrough of Brain2Qwerty v2 is not just an improvement in accuracy from 8% to 61%. It is a demonstration that non-invasive methods, such as MEG, combined with deep learning, can compete with invasive implants. However, the path to a commercial product is long: scaling to patients with speech disorders, reducing processing time, and ensuring stability in everyday use are key challenges. Nevertheless, this step brings us closer to an era where thought becomes a command without any surgical intervention.