Meta has introduced an advanced AI system capable of converting brain activity into text with remarkable accuracy. Developed in collaboration with the Basque Center on Cognition, Brain, and Language, the technology was tested on 35 participants using magnetoencephalography (MEG) and electroencephalography (EEG).
By training AI on brain signals, the system achieved an impressive 80% accuracy in decoding typed characters from MEG data, performing twice as well as traditional EEG methods. This breakthrough builds on Meta’s previous research into decoding speech and image perception from brain activity and could eventually help individuals with speech impairments regain communication abilities.
The core technology, called Brain2Qwerty, captures neural activity through a high-precision MEG scanner that records over 1,000 brain snapshots per second. This allows the AI to analyze and map brain signals to keystrokes, effectively predicting words and characters. The deep learning model behind it has been trained on thousands of typed characters, making it increasingly accurate in understanding the connection between brain signals and language formation.
While the results are promising, there are significant challenges. The MEG scanner required for this technology is massive, weighing half a ton and costing around $2 million. Additionally, it is not portable and requires a magnetically shielded environment for optimal performance. Even minor head movements can reduce its accuracy, making practical use difficult outside of a controlled lab setting.Despite these hurdles, the Brain2Qwerty system offers hope for future applications in brain-computer interfaces, particularly in assisting individuals with neurological disorders. If the technology evolves to become more compact and accessible, it could revolutionize human-computer interaction and speech restoration. Although widespread use is still a long way off, Meta’s investment in AI-driven neuroscience points to a future where thoughts could be seamlessly transformed into text.
