
Brain-computer interfaces link a person’s brain to external devices or computers using sensors placed around or inside the head. Such systems have been used for about a decade to help people with paralysis and neurodegenerative diseases. In recent years, companies have added large language models to improve decoding of brain activity beyond conventional signal processing and data analysis. China and the United States are leading development, with small human trials underway in China and some AI-powered devices planned for public sale. NeuroXess in Shanghai has run clinical trials using an implant on the skull with sensors on the cerebral cortex, connected by wire to a chest-embedded transmitter and battery. A trial participant controlled a computer cursor to operate appliances, and a language model enabled real-time Mandarin decoding at 300 characters per minute for a person with epilepsy.
"BCIs, which link a person's brain to an external device or a computer using sensors placed around or inside the head, have been used in people who are paralysed and those with neurodegenerative diseases over the past decade."
"In the past few years, companies, mostly in China and the United States, have added large language models to their brain devices. This enables scientists to decode brain activity more accurately than can be achieved using conventional signal-processing and data-analysing technologies, says Li Haifeng, a neuro-computing scientist at Harbin Institute of Technology in China."
"NeuroXess in Shanghai is one company in China that has run small clinical trials, including on their AI-powered brain implant can assist people with paralysis. The implant is placed on top of the skull, and its sensors are fitted on the brain's outer layer, called the cerebral cortex. The system is then connected by wire to a data transmitter that doubles as a battery, which is embedded in the recipient's chest."
"The company has also developed a large-language model to enable a brain implant to decode Mandarin in real time at a rate of 300 characters per minute. This is faster than the average talking speed of a person who speaks Mandarin as a first language, which is around 220 characters per minute. The AI model generated words and phrases for a 35-year-old woman with epilepsy, says Tiger Tao, co-founder and chief scientist at NeuroXess."
#brain-computer-interfaces #artificial-intelligence #large-language-models #clinical-trials #neurotechnology
Read at Nature
Unable to calculate read time
Collection
[
|
...
]