80% accuracy! Meta develops a non-invasive brain-computer interface that can type with just thinking

Social media giant Meta recently announced that they have developed a new device that enables text input by reading neural signals from the human brain. The research results were introduced in detail by Meta scientists in two studies, using advanced brain scanning technology and deep learning AI models to successfully decode the EEG signals people use when typing, and even reconstruct the complete sentences.

Specifically, the technology relies on a scanner called magneto-electroencephalography (MEG), which can capture weak magnetic signals from the brain. Compared with brain-computer interface technology that requires implantation into the body, the advantage of this device is that it does not require invasive surgery and can work without direct operation of the brain . However, the huge drawbacks of the device are equally obvious: Not only does it weigh nearly half a ton and cost as much as $2 million, it can only be used in a dedicated shielded room to prevent the earth's magnetic field from interfering with the reception of signals. In addition, during use, the subject must keep his head still, and any movement may lead to loss of signals.

Nevertheless, this technology still demonstrates extraordinary potential. According to the researchers, the system is able to detect the keys pressed by a "skilled" typist with up to 80%. Although this accuracy is not perfect, it is enough to construct a complete sentence by decoding brain signals. To achieve this, the research team designed a deep learning system called "Brain2Qwerty" to learn the keys they press by observing thousands of characters entered by users.

Although the current technology is still quite far from practical application, Meta researchers are full of expectations for this discovery. They believe that this study validates the theory that the human brain follows hierarchical structure in language formation, which is of great benefit to further research on artificial intelligence. Jean-Rémi King, head of Meta's Brain and AI team, said that understanding the working principles of the human brain may provide new ideas for the development of machine intelligence.