A team of scientists has announced that a sensor-equipped helmet, combined with artificial intelligence, can translate a person’s thoughts into text.

In this study, participants read passages of text while wearing hats, and their brain electrical activity was recorded through the scalp. These electroencephalogram (EEG) recordings were converted to text using an AI model called DeWave.

Lin Ching Tian Researchers from Australia’s University of Technology Sydney (UTS) say the technology is non-invasive, relatively cheap and portable.

The system is far from perfect, with an accuracy of about 40%, but recent data currently under peer review shows an improvement in accuracy of more than 60%, Lin said.

In a study published in NeurIPS conference In New Orleans, Louisiana, the DeWave program does not use spoken language, but instead has participants read sentences aloud. However, in the researchers’ latest study, participants read the text silently.

Last year, the team he led was jerry tan Researchers at the University of Texas at Austin reported similar accuracy in converting thoughts into text, but used MRI scans to interpret brain activity. Using EEG is more practical because the subject does not have to remain still in the scanner.

UTS team member Charles Zhou said the DeWave model was trained by looking at many examples where brain signals matched a particular sentence.

“For example, when you think of saying ‘hello,’ your brain sends a specific signal,” Zhou says. “DeWave learns how these signals relate to the word ‘hello’ by looking at many examples of these signals for different words and sentences.”

Once DeWave had a good understanding of the brain signals, the team connected it to an open-source large-scale language model (LLM) similar to the AI ​​that powers ChatGPT.

“This LLM is like a smart writer who can craft sentences. We tell these writers to pay attention to the signals from DeWave and use that as a guide to craft their sentences. ” says Zhou.

Finally, the team trained both DeWave and a language model together to further improve their ability to write sentences based on EEG data.

Researchers predict that further improvements to the system could revolutionize communication for people who have lost language due to stroke or other conditions, and could also have applications in robotics.

craig gin from the University of Sydney said he was impressed by Lin’s team’s work. “It’s great progress,” he says.

“People have long wanted to convert brainwaves to text, and the team’s model shows amazing accuracy. A few years ago, EEG-to-text conversion was complete and utter nonsense. .”