Published on
Updated
Reading 3 mins.
Scientists have developed a decoder which, via brain imaging and artificial intelligence, manages to translate a person’s thoughts into language without them speaking. A discovery whose applications are dizzying.
The main objective of this “language decoder” is to help patients who have lost the use of speech to communicate their thoughts via a computer. Although for medical purposes, this new device nevertheless raises questions about the invasion of “mental privacy”, concede the authors of the study published in Nature Neuroscience.
A system that requires no electrodes
To fend off criticism, they point out that their tool only works after brain training by spending long hours in an MRI (magnetic resonance imaging) machine.
Previous brain-machine interfaces, intended to allow people with major disabilities to regain autonomy, have already proven their worth. One of these interfaces was able to transcribe the sentences of a paralytic unable to speak or type on a keyboard. But these devices require invasive surgery, with the implantation of electrodes in the brain, and focus only on the areas of the brain that control our mouths to form words.
“Our system operates at the level of ideas, semantics, meaning”, said at a press conference Alexander Huth, neuroscientist at the University of Austin in Texas, co-author of the study. And this in a non-invasive way.
Detect the different stimulated regions of the brain
During the experiment, three people spent 16 hours in a functional medical imaging device (fMRI): this technique makes it possible to record variations in blood flow in the brain, thus reporting in real time the activity of the areas cerebral during certain tasks (speech, movement, etc.).
Volunteers were played podcasts telling stories. This allowed the researchers to determine how words, sentences and their meanings stimulated different regions of the brain.
They then fed that data into an artificial language-processing neural network using GPT-1, the predecessor to the chatbot ChatGPT.
The network was trained to predict how each brain would react to the speech heard. Each person then listened to a new story inside the fMRI machine, to test whether the network had guessed correctly.
A decoding that goes beyond language
Result: even if it paraphrased or changed the order of the words often, the decoder managed to “reconstruct the meaning of what the person heard”, explained Jerry Tang (University of Austin), first author of the study.
For example, when a user heard the phrase “I don’t have my driver’s license yet.“, the network model responded”she hasn’t even started learning to drive yet” .
The experiment went further: even when the participants imagined their own stories or watched silent films, the decoder was able to capture the essence of their thoughts.
These results suggest that “we decode something that is deeper than language, then we convert it into language”Mr. Huth continued.
A threat to our privacy?
This is a real step forward compared to previous brain-machine interfaces, reacted David Rodriguez-Arias Vailhen, professor of bioethics at the Spanish University of Granada, who was not involved in the study.
This brings us closer to a future where machines will be “capable of reading minds and transcribing thoughts”, commented the scientist. But he warns that this could happen against people’s will, for example when they are sleeping, and therefore jeopardize our freedom in the future.
Risks that the authors of the study anticipated, by demonstrating that the decoder did not work on the brain of a person on which it had not been trained. The three participants also managed to easily outwit the machine: while listening to a podcast, they had to count by seven, imagine and name animals, or tell another story in their head… So many tactics that “sabotaged ” the work of the decoder.
The authors of the study nevertheless called for the establishment of regulations aimed at protecting privacy.