Portable AI System Translates Brain Thoughts into Language

[baoshi.rao]

According to a report by Phys.org on the 11th, scientists at the University of Technology Sydney in Australia have developed the first portable, non-invasive AI system capable of decoding silent thoughts and converting them into tangible text. This technology could assist individuals who are unable to speak due to illness or injury (including stroke or paralysis) in communicating, and it also holds the potential to enable seamless communication between humans and devices such as bionic arms or robots. The latest research represents a pioneering effort to directly translate raw electroencephalogram (EEG) signals into language, marking a significant breakthrough in the field.

Image credit: University of Technology Sydney

In the latest study, participants wore a cap that recorded scalp EEG activity while they silently read text passages. The EEG waves were segmented into different units to capture specific features and patterns from the human brain, a task accomplished by the DeWave model developed by the researchers. The DeWave model translates EEG signals into words and sentences by learning from vast amounts of EEG data.

The researchers noted that this is the first time scientists have incorporated discrete encoding techniques into the process of translating brain thoughts into text. To achieve this, they introduced an innovative neural decoding method. The integration of this latest technology with large language models also opens new frontiers in neuroscience and artificial intelligence.

Previous technologies for converting brain signals into language either required the implantation of electrodes in the brain, such as Elon Musk's Neuralink, or relied on magnetic resonance imaging (MRI) machines to scan the brain. However, MRI machines are bulky, expensive, and difficult to use in daily life. Additionally, these methods struggled to convert brain signals into word-level segments without the aid of eye-tracking, limiting their practical applications. The new technology, however, can function with or without eye-tracking.

The latest study involved 29 participants, suggesting its decoding technology may be more powerful and adaptable than previous methods tested on only one or two individuals, as brain waves vary between people.

Using caps worn on the head rather than electrodes implanted in the brain to receive EEG signals means the signals contain more noise. However, the results show the new system outperformed previous benchmarks in EEG translation. This research has been selected as a highlighted paper for the NeurIPS conference on December 12 in New Orleans, USA, which showcases leading artificial intelligence and machine learning research achievements.