Facebook's "brain reading" tech works but it's still giving up on it

Facebook and a team from the University of California – San Francisco have published a new study on brain-computer interfaces (BCIs) and their suitability for, in this case, restoring speech by enabling the user to 'think' words into existence. The system involved using a BCI to translate a man's thoughts into words in near-real-time when he attempted to speak them.

According to Facebook, this is the first time this sort of system has been demonstrated, highlighting the potential use of brain-computer interfaces in helping treat medical conditions — the same sort of ambitions Elon Musk has expressed about his Neuralink effort.

The study notes that UCSF researchers decoded brain signals in order to translate the man's thoughts into words. The man who participated in this effort lost his ability to speak more than 16 years ago due to multiple strokes. With this new system, Facebook explains that he only needed to try to talk in order to make words appear on a computer screen.

Facebook has funded this effort with the goal of seeing whether it would be possible to develop a BCI that can type out 100 words per minute using neural signals. Facebook has its Reality Labs Brain Computer Interface project that launched in 2017, while UCSF researchers have worked to develop the implantable communications prosthesis, collected the data, designed, and oversaw the study.

This was, according to Facebook, the final phase of the project, which started in 2019 at the UCSF Chang Lab with the aforementioned participant. Facebook said in a statement:

We've learned a lot from Project Steno, particularly as it applies to how algorithms can use language models to improve accuracy for brain-to-text communication.

Despite this, Facebook says that it plans to pivot to wrist-based electromyography-powered neural interface devices going forward. Put simply, such devices can decode brain signals to control devices — which could eventually be used to develop things like high-speed typing based ultimately on brain signals.