Beijing, September 14 morning news, according to reports, artificial intelligence technology now makes non-invasive brain decoding technology a step further! Although the technology does not yet enable people who are unable to communicate verbally to talk and laugh like normal people, it allows scientists to accurately decode the content of their speech.
Artificial intelligence technology can decode words and sentences from participants’ brain activity with incredible accuracy, up to 73%, which means AI is a step closer to “non-invasive decoding” of brain language and consciousness from brain activity data!
The artificial intelligence technology can decode words and sentences from the participants’ brain activity with incredible accuracy, but still cannot achieve 100% accuracy, people only need a few seconds of brain activity data, that is, through artificial intelligence technology to infer what people hear, they found in a preliminary study, the participants from the selective test to find the correct answer probability of 73% on average.
Giovanni Di Liberto, a computer scientist at Trinity College Dublin who was not involved in the study, said artificial intelligence has performed beyond what many thoughts were possible.
On Aug. 25, the media reported that Facebook parent company Meta is developing a new artificial intelligence technology that could eventually be used for tens of thousands of people around the world who are unable to communicate by voice, typing, or gesture, including people with minimal consciousness, atresia syndrome or “vegetative states,” now commonly referred to as unresponsive people. They are now commonly referred to as unresponsive wakefulness syndrome.
Most current technologies to help people with speech disorders are known to be somewhat physically invasive, requiring high-risk brain surgery to implant electrodes, said Meta AI researcher and neuroscientist Jean-Remi King, adding that the newly developed AI technology promises to provide a viable solution to help people with communication disorders, rather than using invasive methods. Instead of using invasive methods.
Jean-Remi and colleagues have developed a computational tool that can detect words and sentences in 56,000 hours of recorded speech in 53 languages, also known as a language model, to learn how to identify specific features of language at both the subtle level (e.g., letters or syllables) and the broader level (e.g., words or sentences).
The team applied the artificial intelligence system with language models to four institutional databases that included the brain activity of 169 volunteers in which participants listened to different stories and sentences, such as Ernest Hemingway’s “The Old Man and the Sea” and Lewis Carroll’s “Alice in Wonderland,” while staff used a magnetic brain wave tracer (MEG) or The participants’ brains were scanned using an electroencephalograph (EEG), which measures the magnetic or electrical components of brain signals.
Then, with the help of a computational method that helps explain the physical differences between actual brains, the team tried to decode what each participant heard using 3 seconds of brain activity data, instructing the AI system to match the speech in the story recordings with the brain activity patterns that the AI calculated corresponded to what people heard, and then predicting, based on more than 1,000 possibilities, what the participants of what the participants were likely to hear in a short period of time.
The researchers found that the top 10 possible answers selected by participants were 73 percent accurate when tested with the brainwave tracer, but that accuracy dropped to less than 30 percent with the EEG results, making the brainwave tracer a very good performer. But we don’t foresee any real-world applications for the system,” said Liberto, “What use could it be? Magnetoencephalography is a bulky and costly machine, and applying the technology to clinics will require technological innovations to make the device less expensive and easier to use.
Jonathan Brennan, a linguist at the University of Michigan, Ann Arbor, says, “In this latest study, it’s important to understand what ‘decoding’ really means, a term often used to describe the process of deciphering information directly from the source, and here AI technology can do this because the system can provide a limited range of possible answers with dramatically improved accuracy. For language, if we want to extend that AI system to practical applications, this is very difficult to achieve because the language applications are infinite.”
What’s more, the AI can decode information about participants who are passively listening to the audio, which is not directly relevant to nonverbal patients, and in order to make it a meaningful communication tool, scientists need to learn how to decipher from brain activity what the patient is trying to say, such as hunger, discomfort expressions or simple “yes” or “no” expressions.
In fact, this AI technology is decoding speech perception, not speech generation, and although speech generation is the ultimate goal of scientists, there is an urgent need to further improve the related science and technology for the time being.
