A computer that can reproduce what you think. It’s a terrifying idea, but at the same time it’s very useful for people who can’t speak, for example. Now American scientists have succeeded in making something like that. Using a so-called “semantic decoder,” someone’s brain activity can be scanned and converted into text.

Researchers University of Texas We created a language decoding system that does not require surgical implants. All it takes to train the system is a session in an fMRI scanner, in which the subject listens to an hours-long podcast, to calibrate the AI ​​model. Then, the system can convert ideas into text. It’s not about words from a predefined list, anything is possible when it comes to text. If the test-taker is open to mind-reading the scanner, the system can pick up the story the person is listening to or the story they have in mind. Thus, a new form of communication was born.

From complex ideas to text
“This is a huge leap forward in non-invasive methods. Previously, only single words or short sentences could be captured in this way,” says researcher Alex Huth. “We were able to set up a model that decodes complex thoughts and ideas over a longer period of time and converts them into a continuous stream of text.”

Results Not word for word, but a description of what was read or thought about. To be sure, the system is not yet perfect: in just over half of the cases, it succeeds in producing text that approaches or even reproduces exactly the meaning of the original words.

Take, for example, the text: “I don’t have my driver’s license yet,” which was said to a study participant while he lay in an MRI scanner. The AI ​​language decoder analyzed brain activity and translated it as: “You haven’t even started driving lessons yet.” In a subsequent test, the system decoded the words it had listened to: ‘I didn’t know whether to scream, cry, or run away. Instead I said leave me alone! Like, “She started screaming and crying, and then all she said was, I told you to leave me alone.”

Calibration and focus
This mind-reading technique is great news for people who have lost their ability to communicate normally due to a stroke or other condition, but it also raises questions about potential abuse and invasion of privacy. Are our thoughts still safe in the future? The researchers explain that the semantic decoder produces the correct texts only if the subject cooperates fully. First, the system must become aware of the individual’s specific brain activity by lying for 15 hours in a brain scanner and listening to the audio broadcast, so that the unique pattern of the brain waves can be stored. Then, the AI ​​system can only mock thoughts if the subject is not thinking of something else at the same time — for example, reciting the seven-stroke table while listening to a text — or actively suppress the thoughts. Without calibration or focus, the results are useless.

Abuse is imminent
“We take the concern that this technology could also be used for malicious purposes very seriously and have done everything we can to prevent this,” said researcher Jerry Tang. “It should not be the case that this type of technology will be used on people who have not given permission and have not been helped.” The system can’t be used outside of the lab at the moment, as it only works with a bulky fMRI scanner, but the team believes mind reading should also be possible with more portable brain scanning devices, such as Near infrared spectroscopy (fNIRS). fNIRS measures where and when there is more or less blood in different brain regions. This is exactly the same signal that fMRI measures,” Huth explains.

However, the question remains whether totalitarian regimes or other malign parties could get away with a future version of this technology. This is why Tang is in favor of preemptive measures. “I think it’s important, while this technology is still in its infancy, to put in place policies that protect people and their privacy. There is a need for laws that describe what these devices are used for,” says Tang.