Researchers have developed an innovative system capable of converting individuals’ thoughts into written texts by using a helmet equipped with sensors and artificial intelligence.
Interestingly, data is collected using sensors placed on the head, without the need for surgery or introducing any foreign materials into the body.
The system still needs improvement to enhance its accuracy, but it could revolutionize communication for individuals unable to speak.
A team of scientists has announced the possibility of converting human thoughts into written words using a helmet containing sensors and artificial intelligence technologies. Participants in the study were asked to read text paragraphs while wearing the helmet, which recorded brain electrical activity.
Subsequently, the recorded brain electrical activity of participants (EEG) was converted into texts using an artificial intelligence model known as DeWave.
Chien Ting Lin from the University of Technology in Sydney praised the benefits of the new technology that helps in turning human thoughts into writing. He pointed out that it is non-invasive, cost-effective, and easy to transport.
Nevertheless, Lin acknowledged that the current system is still imperfect, with its current accuracy reaching only 40%. However, he expressed optimism that accuracy could be improved in the future, as the system is currently under peer review.
The study was presented at the NeuroIPS conference in New Orleans. In initial trials, participants read sentences aloud despite DeWave’s system not relying on speech. In recent experiments, participants read their sentences silently.
Last year, a team led by Jerry Tang from the University of Texas achieved similar accuracy in converting thoughts into texts using functional magnetic resonance imaging technology instead of brain electrical schematics. However, the use of brain electrical schematics is considered more feasible as it does not require individuals to lie inside an imaging device.
Charlie Cho from the University of Sydney says they trained the DeWave model on a large set of examples, linking brain signals to specific sentences. For instance, when a person says “hello,” the brain sends specific signals.
When the DeWave model learns how to link these signals to sentences, it can use them to generate new sentences.
After the DeWave team learned how to interpret brain signals, they connected it to a huge open-source language model similar to the artificial intelligence model used in the ChatGPT conversational robot.
Cho says: “The large language model acts like a writing tool for the brain, using signals sent by DeWave to construct sentences.”
The team trained the DeWave and the large language model together to enhance their ability to convert thoughts into texts.
Although the system is still in the development stage, researchers expect it to contribute to improving communication for individuals who cannot speak, such as patients who have suffered a stroke, and it can also be used in robotics applications.
