Introducing something like a ‘Chat-scan.’
In an intriguing development, Israeli researchers have come up with a program called Brain-IT that can interpret thoughts by analyzing brain scans, and then produce surprisingly accurate visual depictions of those thoughts.
The technology can generate reconstructions that closely match both the structure and meaning of what the person viewed, as detailed in a paper submitted to the International Conference on Learning Representations.
This innovative method can recreate everything from stop signs to specific types of pizzas, even including the exact number of slices imagined by the individual—a digital representation that mirrors conceptually and visually what was thought of.
During their experiments, researchers showcased original images alongside AI-generated recreations based on participants’ brain activity.
Interestingly, the concept of AI interpreting thoughts isn’t entirely new. Back in 2017, Purdue University researchers developed a model capable of predicting the imagery a subject had in mind, but with only rudimentary, grainy outcomes. Then in 2022, Japanese scientists created an AI model named Stable Diffusion, which provided more clarity but operated based on text-to-image technology.
Brain-IT, however, nails the details with remarkable precision and pulls this off in just an hour—much quicker compared to its predecessors, explained Michal Irani from the Weizmann Institute of Science, who led the project’s development.
The researchers trained this system using over 70,000 images gathered while eight subjects viewed various visuals.
Though this sample may seem limited by typical AI standards—expanding it would require long hours of imaging hundreds of participants—researchers utilized an “encoder” to aid their work.
This approach allowed the tech to decode images from brain scans and vice versa, creating a robust dataset.
By converting images back and forth—from a random unseen image to its predicted brain scan and back—the models could construct extensive data during the training phase, despite the original scans never being conducted.
Throughout this fascinating process, the tech identified 128 functional regions in the brain, some of which were already known, while others were newly discovered, all essentially performing similar functions across individuals.
For instance, certain areas lit up in response to food-related images, while others were triggered by sports visuals. Similarly, distinct regions corresponded to thoughts about indoor versus outdoor contexts.
Irani suggested that Brain-IT could facilitate communication for individuals with paralysis and lead to more efficient brain studies.
“Funding for 40 hours of imaging isn’t feasible for most,” remarked Tommy Sprague, a neuroscientist from the University of California, Santa Barbara, highlighting the high costs involved in such research.
This technology may also potentially allow deeper insights into our mental processes in the future.
“If we tackle the hurdles ahead, we might even unlock the ability to read dreams someday,” speculated Irani.
Yet, some experts express concern that such advancements could render private thoughts a thing of the past—something that has been a persistent worry with the rise of advanced technologies.
“If it’s possible to secretly extract someone’s thoughts, then we’re stepping into a world reminiscent of 150 years of sci-fi narratives, which raises numerous ethical worries,” Sprague added.
