A machine that can read our thoughts may sound like something out of a science fiction novel, but a new artificial intelligence (AI) system called DeWave has made this a reality.

Australian researchers have developed this technology, which translates silent thoughts from brainwaves into text using electroencephalogram (EEG) caps to record neural activity. Scientists at the University of Technology Sydney (UTS) achieved over 40% accuracy in early experiments and hope that DeWave's AI can enable communication for those who are unable to speak or type.

This non-invasive system does not require implants or surgery, unlike Elon Musk's Neuralink chip in development. It was tested on datasets from participants who read text while their brain activity and eye movements were monitored. By matching EEG patterns to eye gaze points that indicate recognized words, DeWave learned to decode thoughts.

Chin-Teng Lin, the chief investigator at UTS, stated that DeWave introduces "an innovative neural decoding approach." He said in a statement, "This research represents pioneering efforts in directly translating raw EEG waves into language, marking a significant breakthrough in the field."

Professor Lin continued, "This is the first time that discrete coding techniques have been integrated into the brain-to-text translation process, introducing an innovative neural decoding approach. The integration with large-scale language models also opens up new frontiers in neuroscience and artificial intelligence."

DeWave may one day assist paralyzed patients. Verbs from neural signals are the easiest for AI to recognize, while specific nouns are sometimes translated into synonyms. Researchers propose that semantically related concepts can produce similar EEG patterns, which presents a challenge for the research.

This technology, which only requires a snug-fitting EEG cap, could one day enable smooth communication for paralyzed patients or direct control of assistive devices. However, extensive research is still needed to improve the system's accuracy to a level comparable to speech recognition, around 90%.

Combined with rapidly advancing language models, similar brain-computer interfaces may one day allow people to communicate or interact with technology simply through their thoughts.