Call it a Chat-scan.
Just in case artificial intelligence didn’t spark enough privacy concerns: Israeli scientists have devised a mind-reading program named Brain-IT that can tell what people are thinking just by analyzing their brain scans — and then recreate that image with shocking accuracy.
The technological telepath “produces reconstructions that more closely resemble the viewed images both structurally and semantically,” the researchers wrote in a paper submitted to the International Conference on Learning Representations.
These precise reproductions entail everything from stop signs to an olive and tomato pizza with the exact number of slices people had envisioned, like the digital equivalent of a counterfeit bill.
They demonstrated these in a series of composites that juxtaposed the original image (left) with the one the AI recreated based on their brain scans (right).
Coincidentally, AI mind readers are nothing new. In 2017, researchers at Purdue University in Indiana devised a model that could predict what image a subject was thinking of, but could only produce grainy recreations, Popular Mechanics reported. Meanwhile, in 2022, Japanese scientists developed an AI model dubbed Stable Diffusion in 2022 that could more precisely approximate what a subject was but relied on text to image tech.
By contrast, Brain-IT recalls the images to a T — and in just one hour, rather than multiple hours like prior iterations, Weizmann Institute of Science’s Michal Irani, who spearheaded the tech’s development, said in a statement.
Researchers built the system by feeding it over 70,000 images that were collected while eight subjects looked at various pictures.
As this was a relatively small sample size by AI brain scanner standards — generating a larger one would require cramming vast numbers of participants in MRI machines for long periods — researchers were forced to rely on an “encoder.”
Using this shortcut, the tech could both decode the images from brain scans and vice versa to build a large sample.
“We realized that by translating back and forth – from a random image that had never been viewed in an fMRI machine, to a predicted brain scan, and then back to the image we started with – the models would effectively build themselves a massive dataset,” said Irani. “In this way, during training, the models would learn to generate scans that encode images effectively, even though those fMRI scans had never actually been performed.”
During this mind-blowing experiment, the tech identified “128 functional regions” — some known and some new — that perform similar functions in all people.
For instance, some of these image synthesizers responded to images of food, while others were activated by sports pics.
Meanwhile, different areas corresponded to indoor and outdoor thoughts.
Irani said they could use the tech to help people suffering from paralysis communicate, as well as study the brain more efficiently and cost-effectively.
“None of us can afford 40 hours of imaging for a new subject,” said Tommy Sprague, a neuroscientist at the University of California, Santa Barbara, in a statement. “It’s something like $600 to $1,000 an hour.”
In the future, the tool could be used to probe even further into our psyche, the team hopes.
“If we overcome all these obstacles, it’s possible that in the future we may even be able to read dreams,” said Irani.
Of course, some techsperts worry that the mind-reader could make private thoughts a relic of the past — an ongoing concern with the omnipresent tech.
“But if there’s a way to surreptitiously extract information about what you’re thinking about, then …150 years of sci-fi can come true anytime, and that’s worrisome in a lot of ways,” said Sprague.
