Researchers in Australia have developed a non-invasive, portable system that uses artificial intelligence (AI) to turn silent thoughts into text.
The technology could potentially help people who are unable to speak to communicate, including patients who were paralysed following an accident or illness.
It could also provide a new type of interface between humans and machines, allowing seamless communication.
Other systems have been developed that are able to translate brainwaves into language, including Elon Musk’s Neuralink.
The researchers at the University of Technology Sydney (UTS), from the GrapheneX-UTS Human-centric Artificial Intelligence Centre, pointed out that Neuralink relied on surgery to implant electrodes in the brain, while other systems used large and expensive MRI machines.
They also tend to require additional elements, such as eye-tracking, which can restrict the practical application and deployment of the technology.
Meanwhile, the UTS system uses a non-invasive cap that is worn on the head and records electrical brain activity through the scalp using an electroencephalogram (EEG).
It can also be used with or without additional eye-tracking functionality.
AI model translates EEG signals into words and sentences
As the EEG signals are captured through the scalp, they are segmented into different units, capturing specific patterns and characteristics from the human brain.
This is done by a specially developed AI known as DeWave, which has already been trained on large volumes of EEG data.
It then decodes the EEG signals, translating them into text that can be displayed on a screen.
In a study that was presented as a spotlight paper at this month’s NeurIPS conference, 29 participants demonstrated the technology.
According to the researchers, the technology is likely to be more adaptable and robust, as previous technologies had only been tested on one or two individuals, and EEG waves differ from person to person.
The volunteers silently read passages of text while wearing the ‘mind reading’ cap.
The system was able to translate the text with an accuracy score of around 40% on a scale known as BLEU-1.
BLEU (Bilingual Evaluation Understudy) is a separate algorithm commonly used for evaluating the quality of machine-translated text.
The study’s first author, Yiqun Duan, explained that the AI is currently better at correctly identifying verbs rather than nouns.
With nouns, there was a tendency to produce synonyms, such as ‘the man’ instead of ‘the author’, and this affected the overall accuracy.
However, despite these challenges, it is still producing meaningful results.
Today’s news was brought to you by TD SYNNEX – the UK’s number one solutions distributor.
