The innovative company Meta has published the results of a study on the so-called “brain decoder.” Using the achievements of artificial intelligence, it converts thoughts into speech. This unique development is based on non-invasive methods of the brain’s activity recording. The research used magnetoencephalography and electroencephalography. As a result, the scientists collected over 150 hours of recordings of 169 volunteers without health problems.
The algorithm recorded the reaction of the brain of the subjects to listen to audiobooks and individual phrases in Dutch and English. Then, certain words were selected from the text. In this way, a so-called dictionary was created, which started the reverse process of interpreting thoughts and turning them into text.
The algorithm’s accuracy reached 73% when using a set of 793 words, often used in everyday life.
It turned out that EEG and MEG are less reliable than brain-implanted sensors. Therefore, the researchers had to collect more data to improve the accuracy of the given AI model.
Scientists believe that this development will be useful for people who have lost the opportunity to speak and write. At the same time, it is an important contribution to studying the human brain.
In the future, scientists want to increase the dictionary of the algorithm to make it even more accurate.