New AI Can Figure Out What You’re Thinking About From a Brain Scan

New AI technology could soon be able to "mind-read" individuals by analyzing their brain scans to determine what they are looking at, and even generate a highly accurate recreation of their mental image. Neuroscientist Tommy Sprague of the University of California, Santa Barbara, described the results as "very impressive," but warned that the capability to secretly extract information about someone's thoughts could lead to a reality similar to what is depicted in science fiction.

While this isn't the first research to accomplish a similar feat, the team from the Weizmann Institute of Science claims their method produces a higher-fidelity representation of the subject's mental picture.

For their study, published in a recent conference, researchers trained an AI model on high-resolution brain scans of eight subjects viewing thousands of different images. They achieved this by creating a "brain decoder" that predicted both the structure and content of the image being viewed. This decoder was then combined with an "encoder" model that predicts brain activity from the actual images, allowing for an iterative process where the AI continuously refines its predictions until the recreated image closely resembles the original.

Despite its potential benefits for aiding those with communication impairments, the possibility of surreptitiously accessing an individual's thoughts raises significant concerns regarding brain privacy. Currently, the technology necessitates the use of high-resolution fMRI machines, which are large, expensive, and inaccessible to most.

However, if similar techniques could be adapted for smaller, wearable EEG devices, the implications for potential misuse could be far-reaching. Neuroscientist and philosopher Marcello Ienca of the Technical University of Munich, Germany, expressed his apprehension that these advancements could be co-opted for "ethically and societally problematic commercial uses."