Have You Ever Dreamed of a Superpower?

Have you ever dreamed about having a superpower? Maybe you imagined yourself traveling into the future to see intergalactic battles between rebels and the empire. Or maybe you wanted to soar through the sky like Superman.

But what if, instead of time traveling and flying, I told you another superpower may no longer be a fantasy?

In May 2023, researchers from Texas showed that with non-invasive brain imaging, you can read people's minds.

Imagine knowing what another person is thinking just based on their neural activity. It sounds both cool and dystopian, doesn’t it? But more importantly, how does it work? Well, it turns out the answer is AI.

In this study, Tang and colleagues scanned participants’ brains using an fMRI scanner while participants were listening to 16 hours of storytelling podcasts. With that data, researchers created activation maps for each participant that showed how their brains reacted to different words, phrases, and sentences.

Using the collected data, the team trained the AI to encode text into an activation map. But by itself, this neural network struggled to translate the brain activity into words. So, they decided to incorporate a large language model.

The language model was trained on a large corpus of books, Reddit comments, and podcasts, which allows the model to predict the next likely word within a sequence. For example, the language model can predict that the most likely words in the sentence “Each morning I have a cup of… ‘coffee’ or ‘tea’.”

Combined together, the decoder works as follows: First, the language model generates possible words, and then the encoder predicts fMRI recordings for these words. The system then compares if, for each possible phrase, the predicted brain activity matched the observed brain activity or not, and then just keeps the best word.

After the training was complete, participants revisited the scanner to see if the decoder could read their minds. While subjects were listening to stories, the system started producing words, sentences, and eventually ideas that matched what the person heard.

For example, when a participant heard “I don’t have my driver’s license yet,” the brain activity was translated as “she has not even started to learn to drive yet.” And that’s not everything. Researchers even tried to decode participants’ thoughts.

As an individual was watching a silent animation where a dragon kicked a girl down, the system translated it as “I see a girl that looks just like me get hit on her back and then she is knocked off.”

Alexander Huth, one of the researchers on the team, said: “That really demonstrates that what we’re getting at here is something deeper than just language. … [The system] works at the level of ideas.” 

Although at the moment the system mainly gets the gist and not the precise phrasing, researchers believe one day it could help restore communication to people who lost the ability to speak due to neurodegenerative disorders or injuries.

Even though the technology is still in its infancy, some people are concerned it can be used on someone without them knowing. For example, by an authoritarian regime to control people. But researchers assured that you can’t use the system that way.

Firstly, there are structural and functional differences between our brains. So, to use it on anyone, that person has to lie perfectly still and pay attention to stories for at least 15 hours. And that’s not to mention that decoding happens inside the large fMRI scanner, which is very difficult to go unnoticed.

Secondly, without your cooperation, it won’t be possible to train the decoder in the first place, as participants could resist it by simply thinking of animals.

So, if you had plans to use this technology to win a few negotiations or maybe become a social butterfly, well, I’m sorry to tell you, but you’ll probably have to wait for quite some time. But by that point, there will already be regulations to protect our mental privacy.