The first machine to translate brain waves into sentences was built.

In their study, scientists note that there are now a large number of similar devices,

however, most of them are inefficient - they can only decode fragments of spoken words, and their accuracy is only a few percent.

Machine Learning Specialist Dr. JosephMackin of the University of California (UCSF) and his colleagues tried to improve the accuracy of such machines using artificial intelligence. They trained algorithms to convert brain patterns into sentences in real time with a word error rate of only 3%.

As part of the device tests, four volunteersread sentences out loud, while electrodes recorded their brain activity. After that, the data entered the computing system, which created graphs of regularly occurring signals in this information.

These repeating patterns are associated withrepetitive characteristics of speech, such as vowels, consonants, or commands to different parts of the mouth, scientists note. Another part of the system decrypted this data back into sentences.

So far, artificial intelligence can decryptonly speech with a length of no more than 30-50 sentences with a low number of errors. However, the device’s interface already allows you to decode individual words, and not just sentences - this means that in the future the device can be taught to recognize words that have never been found in educational systems.

Earlier, HiTech described in detail how the human brain works and why artificial intelligence can help in the study of this organ.