Prototype of BrainGPT mind reading: reading ideas and converting them to text

CTOnews.com, December 13 (Xinhua) there is new hope for stroke patients and other patients who are unable to communicate in a conventional way. Scientists are developing BrainGPT to read the wearer's ideas and convert them into readable text content.

A research team from Sydney University of Science and Technology recently developed BrainGPT, which allows users to express their thoughts without wearing an eye tracking camera or other additional hardware, as long as they wear an EEG cap and connect to a computer.

The team customized the DeWave software, inviting 29 volunteers to record electrical signals generated by the brain while silently reading text paragraphs.

Through AI model training, DeWave analyzes and maps the words and phrases corresponding to EEG signals, and then detects similar EEG signals, which can be automatically converted into text content.

The translation score of the system is currently about 40 per cent on the BLEU (bilingual evaluation substitute) scale, which is a measure of the accuracy of machine translation text, but the team hopes to continue to improve the technology and increase the number to 90 per cent in the future. The demo video attached to CTOnews.com is as follows: