news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

A recent study shows that with the support of artificial intelligence technology, brain-computer interface technology has made significant progress.

Doctors at the University of California, Davis, implanted electrodes in the outer layer of the brain of Casey Harrell, an amyotrophic lateral sclerosis (ALS) patient, to decipher what he was trying to say. Researchers found that the treatment exceeded expectations, setting a new benchmark for implanted speech decoders and demonstrating the powerful potential of these devices for people with speech disorders.

ALS is a progressive neurodegenerative disease that affects the nervous system. It affects the nerve cells (motor neurons) that control voluntary muscle movement, causing the muscles to gradually weaken and atrophy, eventually leaving patients without the ability to walk, speak, swallow, and even breathe. As the disease progresses, the muscles that control speech and swallowing are gradually damaged, resulting in slurred speech, and weakness and atrophy of the hand muscles make it difficult for patients to express themselves in writing, eventually leading to a deterioration of speech.

Media reports said Harrell's treatment team surgically implanted four electrode arrays in the outer layer of his brain, which look like tiny beds of nails. This is twice the number implanted in the language area of ​​ALS patients in another study. The 64 probes in each array captured the electrical signals sent by neurons when Harrell tried to move his lips, jaw and tongue to speak.

Three weeks after the surgery, the researchers successfully connected the implant to a bank of computers via cables attached to two metal posts in Harrell's skull.

After only a brief period of computer learning Harrell's speech, the device began recording with 99.6 percent accuracy what he was trying to say across his 50-word vocabulary.

The device actually bypasses Harrell's disease, the researchers say, by relying not on his weakened facial muscles but on the part of his brain's motor cortex that initially formulates speech commands.

“The key innovation is targeting more arrays very precisely to the areas of the brain that we could find that are most expressive of language,” says Sergey Stavisky, a neuroscientist at the University of California, Davis, who led the study.

By the second day of the trial, the device was able to cover 125,000 words with 90% accuracy and, for the first time, generate sentences written by Harrell himself. Moreover, the device spoke these sentences in a voice that was very much like his own: using podcast interviews and other old recordings, the researchers used AI to deeply imitate Harrell's voice before his illness.

As the researchers continued to train the device to recognize his voice, it got better and better.Over eight months, Harrell produced nearly 6,000 different words, the study reported. The device's accuracy rate remained at 97.5 percent.

That score beats the accuracy of many phone apps that transcribe speech. It also marks an improvement over previous studies, in which the implants were accurate about 75 percent of the time, misinterpreting one in four words.

And, unlike Musk’s Neuralink device, which helps people move a cursor on a screen, Harrell’s implant enables him to explore an infinitely larger and more complex realm of speech, a system that researchers say patients could use every day to communicate with family and friends.

The achievement is partly thanks to the artificial intelligence models used by language tools like ChatGPT, the researchers said.At any given moment, Harrell’s implant picks up the activity of a group of neurons, translating their firing patterns into units of sound, such as vowels or consonants. Computers then aggregate strings of such sounds into a word and strings of words into a sentence, selecting the outputs they think are most likely to match what Harrell is trying to say.

But it’s unclear whether the same implants would help people with more severe paralysis. Harrell’s speech, though degraded, has not disappeared. And while the technology works, it’s expensive and most insurance policies don’t cover it, making it unaffordable.