A mind-reading brain implant with password protection
A brain implant can decode a person’s internal thoughts, but it only works if the user thinks of a pre-defined password.
The device, a brain–computer interface (BCI), was able to decipher up to 74% of imagined phrases. It only begins decoding when the user thinks of the keyword, ensuring that private thoughts are not accidentally translated.
The study, published in Cell on August 14, represents a significant advance in the development of BCIs capable of accurately decoding internal speech, says Sarah Wandelt, a neural engineer at the Feinstein Institutes for Medical Research, who was not involved in the research. The password also protects users’ privacy, a crucial feature for practical applications.
BCIs translate brain signals into text or audio and have shown promise for restoring speech in people with paralysis or limited muscle control. Most require users to attempt speaking aloud, which can be tiring. Last year, Wandelt and colleagues developed the first BCI capable of decoding internal speech, based on signals from the supramarginal gyrus, a brain region essential for language.
“We wanted to investigate the risks of accidentally decoding thoughts that users do not intend to express,” says Erin Kunz, a neural engineer at Stanford University and co-author of the new study.
Kunz and colleagues analyzed brain signals collected by microelectrodes in the motor cortex, which is responsible for voluntary movements, from four participants with speech difficulties—one due to a stroke and three due to amyotrophic lateral sclerosis. Participants were instructed to either attempt to speak words aloud or imagine saying them.
The recordings showed that both attempted speech and internal speech originated from the same region and produced similar signals, although those from internal speech were weaker.
Using these data, the researchers trained artificial intelligence models to recognize phonemes, the smallest units of speech. Language models then combined these phonemes into words and sentences in real time, using a vocabulary of 125,000 words.
Aqui está a tradução para inglês americano:
The device correctly interpreted 74% of the imagined phrases from two participants who were instructed to think of specific sentences, a level of accuracy similar to the previous BCI for attempted speech, according to Kunz.
In some cases, the BCI also decoded numbers that participants silently counted on a screen, indicating that it can detect spontaneous internal speech.
To prevent the decoding of unintended thoughts, the researchers included a password in the system. When a participant imagined the password “Chitty-Chitty-Bang-Bang,” the BCI recognized it with over 98% accuracy.
The study is exciting for revealing neural differences between internal and attempted speech, says Silvia Marchesotti, a neuroengineer at the University of Geneva. She emphasizes the importance of exploring speech signals in regions beyond the motor cortex.
The researchers plan to expand these analyses and improve the system’s speed and accuracy. “If we examine other parts of the brain, we may be able to address more types of speech disorders,” says Kunz.
