Meta published a new system called Brain2Qwerty v2, which attempts to translate human brain activity directly into text without brain surgery. According to the company, the study was aimed primarily at people who had lost their ability to communicate because of brain damage.

Log brain signals with MEG

The system relies on a helmet-like brain imaging device, the Brain Magnetograph Scanner (MEG). It records the brain signals that the person is typing, then enters the original nerve signal into the end-to-end AI model, and rebuilds the sentence that the user wants to enter.

Meta indicates that the model also combines the semantic abilities of the large-linguistic model so that, in the event of a convoluted brain signal, recognition can be improved according to context.

Average word accuracy increased to 61%

According to data disclosed by Meta, data training based on nine volunteers was completed for Brain2Qwerty v2. Each participant wears an MEG device for an active typing period of about 10 hours, with a total of approximately 22,000 sentences.

Meta states that the average word accuracy rate of the system was 61 per cent, which is significantly higher than the previous method of about 8 per cent. The company also indicated that, as training data increased, the accuracy rate of decodering continued to rise, indicating that there was room for further optimization of the route.

  • The training statement is about 22 million.
  • 9 participants
  • Old or no innovative methods are about 8% accurate.

Synchronized open code and research data

Meta indicates that the training codes of Brain2Qwerty v1 and v2 will be made public and that its research partners will also publish v1 data sets. The project also has a $5 million fund to support the development of open neuroscience data sets.

Meta researchers, in a companion paper published in Natural-Neuroscience, noted that most of the current high-performance brain interfaces are still dependent on implanted electrodes, which makes it difficult to expand on a large scale, for reasons including surgical risks and long-term maintenance of implanted equipment.

The brain interface continues to rise.

Meta believes that the accuracy rate of this unsolved programme is close to what was normally achieved by implanting technology, and that it is likely to narrow the gap between invasive neuropsychiatrics and non-surgery communication systems.

This release is also occurring at a time when brain interface studies continue to heat up. Neuralink in Mask and Merge Labs, supported by OpenAI CEO Sam Altman, are promoting technologies to help people with nervous system disorders recover their communication.

At the same time, more research teams and start-ups are trying to use AI to improve systemless performance. Previously, Neurable had introduced AI brain earphones that monitored concentration and cognitive fatigue; the MIT break-up project AlterEgo had also released wearable devices to convert facial and larynx neural muscle signals into text and instructions.