A group of scientists from Australia have managed to convert human thoughts into written words. They achieved this thanks to a helmet with sensors and an artificial intelligence model called DeWave. Its creators emphasize that the technology is easy to transport, non-invasive and not expensive.
The study was conducted by the GrapheneX Center at the University of Technology Sydney (UTS). Scientists asked 29 volunteers who silently read passages from the text while wearing a hat. The device allowed their brain activity to be recorded using an electroencephalogram (EEG). These signals were then decoded thanks to AI.
Chin-Teng Lin, one of the project leaders and director of the GrapheneX Center, said that the interface is in full development. Lin emphasized that they have already achieved The accuracy level of the artificial intelligence helmet increases from 40% to more than 60%.
The system is “the first to incorporate discrete encoding techniques into the brain-to-text translation process,” Lin said in a statement. Unlike other mechanisms, such as Elon Musk’s Neuralink project, implantation of electrodes does not require surgery. An artificial intelligence helmet also does not require an MRI machine, which is large, expensive and impractical, the scientist emphasized.
How was the artificial intelligence helmet developed?
The technology behind the AI headset could help people who are unable to speak due to illness or injury, such as stroke or paralysis. The team also notes that this could enable seamless communication between humans and machines.
DeWave, the AI that powers it, was trained by observing examples of brain signals corresponding to specific sentences. “For example, when you think about saying ‘hello,’ your brain sends certain signals,” said Charles Zhou, a member of the development team. New scientist. “DeWave learns how these signals relate to the word “hello” by looking at many examples of these signals for different words or sentences.”
After DeWave clarified these signals, the team connected it to a large language model (LLM), similar to the technology behind ChatGPT. “This LLM is like a smart writer who can put together sentences. We advise this writer to pay attention to DeWave signals and use them as a guide to create proposals,” Zhou said.
Using EEG signals obtained through an AI headset rather than electrodes implanted in the brain means the signal is noisier. However, the researchers are confident that they will be able to match the accuracy level of traditional language translation or speech recognition systems, which is close to 90%.
Talk to the machines
The possibility that an AI helmet could serve as a tool for communicating with machines is not far-fetched. In fact, this research is a continuation of the development of previous brain-computer interface technology developed by the same university in partnership with the Australian Defense Force.
In this first study, the team achieved use human brain waves to control a kind of robot dog. A group of sergeants used augmented reality glasses and a biosensor placed on the back of their heads.
Those who used the system were required to blink in a certain way. Thus, the biosensor read brain waves and sent the robot dog a command to perform a certain movement. “Integration with large language models also opens up new horizons in neuroscience and artificial intelligence,” Li emphasized.
Source: Hiper Textual
