Key facts
- Meta's Brain2Qwerty v2 is a non-invasive AI system that decodes brain activity into text.
- The system uses a magnetoencephalography (MEG) scanner and deep learning models.
- It achieved 61% average word accuracy, significantly higher than prior non-invasive techniques.
- Meta has released the system's training code and is supporting open neuroscience datasets.
- The technology aims to help individuals who have lost the ability to communicate.
Meta has unveiled Brain2Qwerty v2, a novel non-invasive artificial intelligence system designed to translate brain activity directly into text. The system utilizes a helmet-like magnetoencephalography (MEG) scanner to record neural signals, which are then processed by an end-to-end AI model that reconstructs intended typed sentences. This approach is enhanced by fine-tuning large language models on neural data, allowing for the use of semantic context to interpret noisy brain recordings.
During training, Brain2Qwerty v2 processed approximately 22,000 sentences from nine volunteers, each recorded for 10 hours while actively typing. Meta reported an average word accuracy of 61%, a substantial leap from the roughly 8% accuracy achieved by previous non-invasive methods. This performance level approaches that of brain-computer interfaces requiring surgically implanted electrodes, but without the associated surgical risks and maintenance challenges.
Meta is making the research open by releasing the training code for both Brain2Qwerty v1 and v2, with its research partner releasing the v1 dataset. The company also established a $5 million fund to support open neuroscience datasets, aiming to accelerate research in identifying, diagnosing, and treating neurological disorders.
The advancement arrives at a time of heightened activity in brain-computer interface research, with companies like Neuralink and Merge Labs also developing technologies to aid communication for individuals with neurological conditions. While some pursue invasive implants, others, like Neurable and AlterEgo, are focusing on AI-powered non-invasive systems.