BrainGPT Converts Thoughts into Text

Researchers have invented a mind-reading cap capable of non-invasively decoding thoughts into text.

Wearable

Image credits: University of Technology Sydney

For stroke victims and other "locked-in" individuals who are unable to communicate through traditional channels, there might be new hope. It takes the shape of the novel BrainGPT technology, which is still in the early stages and can read users' thoughts and translate them into legible text.

In addition to being unable to speak, paralyzed people in a locked-in state are also unable to communicate via hand or head movements. While some of them can move their eyes to use eye-tracking communications systems, others are not even able to do so, reports NewAtlas. The mind reading cap is being developed by Technology Sydney in Australia. It could one day help people unable to speak due to illness or injury, while also providing a way for humans to interact directly with machines.

In other words, the AI-based algorithms at DeWave discovered which particular EEG patterns correlated with which textual words and sentences. It recognized that the user was thinking the appropriate word or phrase - or at least, it frequently did - when it later picked up those signals while the user was not reading the text.

As a gauge of machine translation accuracy, the system currently has a translation score of roughly 40% on the BLEU(BiLingual Evaluation Understudy) scale. That being said, the team intends to increase that percentage to roughly 90% after additional technological advancements.

During the cap tests, subjects were instructed to read text passages silently while their electrical brain activity was monitored with an electroencephalogram (EEG). Next, with an accuracy of 40–60%, an artificial intelligence model named DeWave was employed to translate the ideas into written language.

"This research represents a pioneering effort in translating raw EEG waves directly into language, marking a significant breakthrough in the field," said the lead scientist, Prof. C.T. Lin. "It is the first to incorporate discrete encoding techniques in the brain-to-text translation process, introducing an innovative approach to neural decoding."