Functional magnetic resonance imaging (fMRI)

History

2022: First time successful non-invasive method of reading thoughts

In mid-October 2022, scientists report that they have developed a method that uses recordings of functional magnetic resonance imaging of the brain to restore a continuous language. The findings are the next step in the search for better brain-computer interfaces, which are being developed as an assistive technology for those unable to speak or type. It is argued that this is the first application of the non-invasive method of reading thoughts.

A team at the University of Texas at Austin details an algorithm that can read words a person hears or thinks during functional magnetic resonance imaging (fMRI) of the brain. Although other teams have previously reported some advances in reconstructing language or images based on signals from implants in the brain, the new decoder is the first to use a non-invasive method to achieve this goal.

For the first time, a non-invasive method of reading thoughts was successfully used

It is difficult to use fMRI data for this kind of research, since they are quite slow compared to the speed of human thoughts. Instead of recording outbreaks of neurons that occur on a millisecond scale, MRI machines measure changes in blood flow in the brain as indirect indicators of brain activity, it is worth noting that such changes take seconds. According to scientists, the reason why the installation used in this study works is that the system does not decipher word for word, but rather determines the high-level meaning of a sentence or thought.

Based on 16 hours of fMRI recordings of the subjects' own brains, the decoder made a number of predictions about what fMRI readings would look like. Using these guesses was key to the decryptor being able to translate thoughts that did not relate to any of the known audio recordings used in the learning process, the researchers said. These assumptions were then checked against a real-time fMRI record, and the prediction that most closely matched the actual readings determined the words that the decoder eventually generated.

In order to determine how successful the decoder was, the researchers evaluated the similarity of the speech generated by the decoder with the stimulus presented to the subject. They also evaluated the language generated by the same decoder that was not verified against the fMRI record. The scientists then compared these estimates and tested the statistical significance of the difference between the two. The results showed that the algorithm's guessing and validation procedure eventually creates an entire story based on fMRI recordings, which the researchers said coincides "pretty well" with the actual story told in the audio recording. However, she also has disadvantages, for example, she does not preserve pronouns very well and often confuses the first and third person. Decoder, knows quite exactly what is happening, but does not know who is doing it.

Because the decoder uses non-invasive brain fMRI recordings, it has greater potential for real-world application than invasive methods, although the costs and inconveniences associated with using MRI machines are an obvious problem. Magnetoencephalography, another non-invasive but more portable brain imaging method more temporally accurate than MRI, could potentially be used with a similar computational decoder to provide nonverbal people with a way to communicate, the researchers said.

From a privacy perspective, it is also notable that a decoder trained on one person's brain scans failed to recover another person's language, returning virtually unusable information in the study. Therefore, a person will have to undergo long training before his thoughts are accurately deciphered.^[1]

Notes