New research has allowed "pulling" whole sentences out of a person's head
28 Jun Earlier, thought-reading systems could only guess the single-digit number that a person was thinking about, and deeper and more complex thoughts were beyond the capabilities of technology. Now, a team of scientists at Carnegie Mellon University (CMU) has developed a way that allows you to read more complex concepts using brain scans and even combine whole sentences.
Even the simplest sentence is very complex: each word is a separate concept, and their placement in a certain order can change the meaning of both individual words and the whole sentence. Scientists have found that the “building blocks” that the mind uses to construct thoughts are made up of concepts, not based on the words themselves. This means that brain processes work in a universal way, regardless of the person's language and culture.
The study examined how the brain encodes complex thoughts and how an fMRI scanner, with a little help from machine learning algorithms, can decode them. The researchers collected 240 "difficult sentences", such as "The witness shouted during the trial." All of these proposals consisted of several building blocks. In total, the study used 42 building blocks that pointed to concepts such as person, setting, size, social interaction, and physical action.
Each of these different types of information is processed in different parts of the brain, so the system can choose a general category of what is on the person's mind. To test their skill, the researchers asked seven participants to read sentences by recording brain activation patterns. After training the algorithm with 239 sentences and comparing the results, he was able to put together the last sentence based solely on brain data.
The team ran this test 240 times, systematically leaving each sentence in turn, and found that the algorithm was able to predict the missing sentence from the brain activation pattern with 87 percent accuracy.
Mikhail Romkin
