US scientists use AI, brain scans to 'read minds', decode thoughts

• 

Scientists at the University of Texas in Austin have found a way to read minds and decode thoughts through the application of artificial technology and brain scans. 

Although the main objective was to be an aid to those who cannot communicate, the scientists were aware of the controversy of violating "mental privacy". Tests showed that the decoder does not function on those who haven't permitted it to be trained on their brain activity over a long time frame inside a functional magnetic resonance imaging (fMRI) scanner.

As previous research shows that people unable to speak or type can spell out words and phrases via brain implants, the "brain-computer interfaces" target the part responsible for controlling the mouth for word formation in the brain.

A neuroscientist at the University of Texas at Austin, Alexander Huth, describes his team's language decoder as "work[s] at a very different level," adding, "Our system really works at the level of ideas, of semantics, of meaning."

According to the study in the journal Nature Neuroscience, this research counts as the first of its kind to have the ability to reconstruct continuous language without resorting to a brain implant.

Read more: Shut down AI or 'everyone on Earth will die', researcher warns

To reach the results, the study required three people to spend 16 hours inside an fMRI machine and listen to spoken narrative stories mainly derived from podcasts such as the New York Times' Modern Love. Doing so gave the chance to the researchers to map out the way in which words and meanings produced responses in parts of the brain known for language processing.

The data was then put into a neural network language model using the predecessor of the AI technology, GPT-1, which was later used in the hugely controversial ChatGPT.

To test GPT-'s accuracy, which aims at anticipating how each brain reacts to perceived speech, each of the individuals listened to a story in the fMRI machine. One of the study's authors, Jerry Tang, claimed that the decoder was able to "recover the gist of what the user was hearing."

For instance, when the phrase "I don't have my driver's license yet" was heard by the individual, the model showed the response, "she has not even started to learn to drive yet", however, it did struggle with identifying personal pronouns. 

Even when the individuals were in their own thoughts or were watching silent films, GPT-1 was still capable of grasping the "gist". Huth added that this indicated that "we are decoding something that is deeper than language, then converting it into language."

Due to fMRI scanning being slow in capturing singular words, it does gather a "mishmash, an agglomeration of information over a few seconds... so we can see how the idea evolves, even though the exact words get lost."

Bioethics professor at Spain's Granada University not included in the research, David Rodriguez-Arias Vailhen, stated that this research goes beyond was has been previously achieved as it illustrates a closer future where machines are "able to read minds and transcribe thought" - but he warned that it could work against us since it could happen against someone's will as when they are asleep.

Read more: UN says AI poses 'serious risk' for human rights

The researchers did admit that when the individuals upon request were counting by sevens, naming things, and imagining animals, the decoder was "sabotaged" as they expressed hope to be able to expedite the process to decode brain scans in real time. 

They called on regulations and guidelines in an attempt to protect mental privacy, with Rodriguez-Arias Vailhen saying, "Our mind has so far been the guardian of our privacy... This discovery could be a first step towards compromising that freedom in the future."

AI has had its advantages, with doctors, scientists, and researchers having developed an artificial intelligence model that can precisely detect cancer, which they believe might speed up illness diagnosis and treatment, as reported by The Guardian last month. 

The researchers utilized CT images from 500 individuals with big lung nodules to create an AI system based on radiomics. The technology can extract critical information from medical photos that the human eye cannot see.