Meta Can Now Turn Brainwaves Into Typed Sentences with Brain2Qwerty v2
Following the release of Brain2Qwerty v1 last year, Meta has unveiled Brain2Qwerty v2 to accelerate its efforts to decode brain activity.
News
- UAE Deploys First AI Agents for Weather Forecasting
- Meta Can Now Turn Brainwaves Into Typed Sentences with Brain2Qwerty v2
- CrowdStrike Surges 95%, Palo Alto 113% as AI Fuels Cybersecurity Spending
- UAE Launches National Accelerator to Build Homegrown Cybersecurity Startups
- Employee Resistance to AI Becomes UAE Firms' Top Workforce Risk: Report
- Apple’s AI Hardware Strategy Faces Pressure on Two Fronts
[Image: Chetan Jha/MITSMR Middle East]
A big proponent of emerging technologies, Meta last year unveiled its ambitions to decode brain activity into typed characters with Brain2Qwerty v1. Now, it’s accelerating those efforts with Brain2Qwerty v2, which decodes full sentences instead of single characters.
To accelerate neuroscience research and development, the tech giant released full training code for both versions, while its innovation partner, the Basque Center on Cognition, Brain, and Language (BCBL), released the v1 dataset.
With Brain2Qwerty V2, Meta will be able to achieve levels of accuracy previously reserved for techniques that require brain surgery. The latest version of Brain2Qwerty leverages approximately 22,000 sentences from nine volunteer participants, each recording 10 hours of data wearing a magnetoencephalography (MEG) device while actively typing. For comparison, v1 was trained on 1/10 of the data used for v2.
“We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions that prevent them from communicating,” it said in the official release.
While invasive procedures, such as stereotactic electroencephalography and electrocorticography, have proven that the neuroprosthesis feeding signals to an AI decoder can restore communication, they tend to be difficult to scale. “Our noninvasive approach can help bridge that gap,” it says.
Meta is relying on end-to-end deep learning to directly decode raw brain signals, bypassing hand-crafted pipelines for neural event detection. By fine-tuning large language models on neural data, the system can leverage semantic context, bridging the gap between noisy brain recordings and coherent language.
According to the company, this new system achieved an average word accuracy of 61%, compared with 8% for other non-invasive methods. In word accuracy, v2 outperformed v1 by a significant margin, scoring 78% compared to 48%.
Meta’s Brain2Qwerty series is part of a broader ambition, the Digital Brain Project, to build open foundational models of the brain, which includes the Tribev2 model for perception encoding, NeuralSet to process brain data at scale, and NeuralBench to systematically evaluate models.
