A Non-Invasive Breakthrough
A team from Meta Artificial Intelligence, PSL University, and the Adolphe de Rothschild Foundation Hospital has unveiled Brain2Qwerty, a system that converts neural activity into written words without any surgical intervention. The findings were published in Nature Neuroscience on August 4, 2026, at 10:00 p.m. For years, brain-computer interfaces have relied on implanted electrodes; a non-invasive approach like this could make such technology far more practical in everyday settings.
How Brain2Qwerty Works
Brain2Qwerty uses two non-invasive techniques: electroencephalography (EEG) and magnetoencephalography (MEG). EEG captures electrical impulses through sensors placed on the scalp, while MEG detects the incredibly weak magnetic fields produced by neurons using a helmet equipped with ultra-sensitive sensors. In the study, 35 healthy volunteers memorized and typed short sentences on a regular keyboard. A deep learning algorithm then learned to identify and anticipate the characters each person intended to enter, before any actual movement occurred.
Across the experiments, the system recorded an average character recognition error rate of 29% with MEG and 65% with EEG. For the most successful participants, the MEG error rate fell to 18%. Moreover, the AI was able to decode sentences that were not part of its original training data. According to the researchers, this narrows the gap between non-invasive techniques and the precision of surgically placed neuroprostheses.
The next step is to refine Brain2Qwerty for use in controlling prostheses and other digital devices. Such advances could dramatically improve communication for people with severe speech or motor impairments, allowing them to interact with those around them more easily. Progress in neuroscience and artificial intelligence may also lead to new therapeutic options for patients with movement or language disorders, while opening fresh avenues for brain research. The potential to deploy this technology across multiple domains, from digital device control to prosthetic limbs, makes it particularly promising.
In a similar vein, advancements in brain-computer interfaces are not limited to text generation. A recent study highlighted how robots can now be controlled by thought alone, showcasing a significant leap in non-invasive technologies that enhance human-machine interaction. These developments underline the growing potential of neural interfaces to transform various aspects of daily life.