SYDNEY, June 16 (Xinhua) -- Australian researchers have created a non-invasive artificial intelligence (AI) system that decodes brainwaves into text using a 128-electrode electroencephalogram (EEG) cap and deep learning algorithms.
The pioneering non-invasive mind-reading technology combines neural signal analysis with large language models to reconstruct phrases like "I am jumping happily, it's just me" from silent speech, the Australian Broadcasting Corporation (ABC) reported on Monday.
The AI model has been trained on a small set of words and sentences to simplify the process of recognizing each word, said Distinguished Professor Lin Chin-Teng, a leading researcher in AI and brain computer interfaces at Sydney's University of Technology, who led the development of the AI model.
"We can't get very precise because with non-invasive, you can't actually put it into that part of the brain that decodes words," Lin was quoted as saying by the ABC.
The AI is used to filter out noise and clarify the brain signals because signals from different brain sources overlap on the skull's surface, he added.
The technology shows promise for stroke rehabilitation, speech therapy in autism, and restoring communication for paralysis patients, the ABC reported.
Beyond healthcare, the team aims to boost cognitive abilities like memory and focus and to integrate the technology with wearables such as Augmented Reality (AR) glasses, it said.
Lin describes their approach as "AI-human co-learning." The system currently achieves about 75 percent accuracy in decoding thoughts to text, with a goal of reaching 90 percent, matching implanted models, according to the report. ■
