Meta has introduced a revolutionary development in neurotechnology — the Brain2Qwerty v2 system, which reconstructs typed text from brain activity recordings without surgical intervention. This is a significant step forward compared to previous non-invasive methods, which demonstrated word-level accuracy of around 8%. The new system achieves an average accuracy of 61%.

How It Works: Deep Learning Instead of Manual Schemas

Brain2Qwerty v2 is based on end-to-end deep learning, allowing data to be decoded directly from raw brain signals. The developers abandoned the creation of manual schemas for identifying neural events, which is a key difference from earlier approaches. The system was trained on approximately 22,000 sentences. The experiment involved nine healthy volunteers, each spending 10 hours in a magnetoencephalography scanner, typing heard phrases on a keyboard.

Results and Limitations

Brain2Qwerty v2 works with continuous recordings of brain activity and reconstructs the entire sentence. A fine-tuned language model that accounts for semantic context is used to refine the result. The best-performing participant achieved an impressive 78% word-level accuracy, with more than half of their sentences decoded with one error or none at all.

However, it is important to understand that the development is not yet ready for clinical or everyday use. First, testing was conducted on healthy volunteers who actually typed the text. For people who have lost speech or the ability to move, this approach requires separate validation. Second, the system does not function as full real-time text input: the model processes the entire sentence, so the user does not see each word immediately after typing it.

Broader Context

The project is part of the Digital Brain Project — an initiative to create open neuroscience datasets with a budget of $5 million. This continues Meta's work in neural interfaces: as early as September 2022, the company published research on a "brain decoder" that used AI to convert thoughts into speech.

My Expert Opinion: Although 61% word-level accuracy represents significant progress compared to the 8% of previous non-invasive methods, the path to practical application remains long. The main challenges are adaptation for people with disabilities and achieving real-time operation. Nevertheless, this project demonstrates that the boundary between thought and action is becoming increasingly blurred, and investments in open neuroscience data are a strategically sound move for the entire industry.