Brain implants, powered by artificial intelligence, are improving rapidly and giving those who have lost their ability to speak a voice again.
In a pair of studies published this week in the journalTrusted Source NatureTrusted Source researchers from Stanford University and the University of California San Francisco both showed off their work on brain-computer interfaces (BCIs), so-called “neuroprosthetics,” that have allowed two women with paralysis to speak again with unparalleled speed and accuracy.
The BCIs read brain activity related to speech and feed the data into a language learning model, which is then output in usable speech either through on-screen text or computer-generated voice.
The Stanford research team’s workTrusted Source involves Pat Bennet, now 68, who was diagnosed in 2012 with amyotrophic lateral sclerosis (ALS), otherwise known as Lou Gehrig’s Disease. ALS is a neurodegenerative disease that causes weakness and paralysis. Muscular control deteriorates over time, including in muscles that involve speaking, swallowing, and even breathing. There is no known cure for ALS. Most often, arm and leg muscles begin to show signs of weakness, before the disease progresses to other parts of the body. However, Bennet’s ALS development was atypical. Today she is still able to move, use her fingers, and even dress herself, though perhaps not so nimbly as before her diagnosis. But she can’t speak. ALS affects Bennet’s lips, tongue, mouth, jaws and larynx — all the tools needed for speech. She can still produce certain sounds, phonemes, but can’t do so accurately or consistently.
But her brain is still working: it is still sending signals down those pathways, trying to wake up her mouth and tongue and produce speech. But there’s a disconnect somewhere down the line. Stanford researchers have now, essentially, cut out the middleman by implanting popcorn-kernel size electrode arrays onto the speech motor cortex of the brain. This device, a BCI, then interfaces with computer software that allows her to speak.
Erin Kunz, a PhD student at Stanford University’s Wu Tsai Neurosciences Institute, and co-author of the research paper, was there when Pat spoke for the first time.
“She was thrilled,” Kunz told Healthline. “We’ve done almost, I think we’ve done 30-plus days of running this with her and even after day thirty, it’s still just as exciting seeing it in real time.
Their work has come a long way. The BCI they use today along with artificial intelligence that learns from language patterns, allow Bennet to speak quickly and accurately, relatively speaking. The team says they’ve achieved a 9.1% word error rate, using a smaller 50-word vocabulary — 2.7 times more accurate than previous state-of-the-art BCIs — and a 23.8% word error rate on a 125,000-word vocabulary. The algorithm they use to take brain signals and turn them into a speech output is able to decode 62 words per minute, more than three times as fast as previous models, and approaching conversational speed of 160 words per minute.
While it is still early, the research demonstrates a proof-of-concept and also a significant improvement over previous iterations of the technology. Kunz hopes their work will eventually give people like Pat more autonomy and improve their quality of life, their friendships, and maybe even allow them to work again.
You must be logged in to post a comment.