What a person sees can increasingly be reconstructed from their brain activity. Researchers at the Weizmann Institute of Science in Israel have developed Brain-IT, an artificial intelligence model that can generate images from fMRI brain scans that come astonishingly close to the actual object being viewed. The institute itself describes this as a new science of “mind reading.”
However, this technology does not read thoughts in the traditional sense. Brain-IT analyzes the brain activity of individuals while they view images. Functional magnetic resonance imaging (fMRI) records changes in blood flow and oxygen supply to different areas of the brain, from which patterns of activity can be inferred.
Professor Michal Irani’s team developed a method that identifies functionally similar regions across different brains. Brain-IT assigns measured brain signals to shared functional clusters and then links this information to features of the perceived image. A diffusion model then uses these data to generate a visual reconstruction.
Just one hour of brain data
The key breakthrough lies in how quickly Brain-IT can be adapted to a new person.
Previous methods sometimes required dozens of hours of individual brain scans. According to the researchers, Brain-IT achieves results comparable to previous models trained on around 40 hours of data while using only one hour of fMRI data from a new test subject. In experiments, recognizable reconstructions were produced with as little as 15 minutes of data.
The model does not merely try to determine whether someone is looking at, for example, a person, an animal, or a building. It combines semantic information about the image’s content with structural elements such as spatial arrangement. Brain-IT is reportedly better than previous methods, particularly in terms of color, composition, and detail.
The process also works in reverse.
What is particularly interesting is that the scientists are not only translating brain activity into images.
The team also works with a so-called brain encoder, which takes the opposite approach: an image is shown, and the model attempts to predict what patterns of activity that image would trigger in the human brain.
According to Irani, during its development the system identified 128 functional regions, or clusters, that perform comparable tasks when processing visual information across different people—even though their precise anatomical locations vary from person to person.
This research goes beyond merely producing spectacular image reconstructions. It could provide new insights into how different human brains process visual information.
From seeing to genuine “mind reading”?
Brain-IT currently primarily reconstructs images that a person is actually seeing. The term “mind reading” should therefore not be confused with freely reading arbitrary thoughts.
But the direction of development is remarkable.
The scientists want to go beyond static images. In the future, methods for more complex visual and other sensory information are conceivable. This brings us closer to the question of whether similar systems might one day be able to reconstruct internal thoughts or other mental content.
The Weizmann Institute itself explicitly places Brain-IT under the heading “The New Science of Mind Reading” and describes the model as an AI capable of “reading” brain activity and reconstructing what a person sees.
Medical opportunities—but also a new frontier for privacy
The medical benefits of such technologies could be enormous. Brain-computer interfaces, for example, could help people who can no longer speak or communicate because of severe paralysis. Systems that convert brain activity into speech or other signals already exist.
Brain-IT now demonstrates how quickly this field is advancing: instead of building a new system for each person using enormous amounts of personal brain data, the AI attempts to recognize shared patterns across different human brains and transfer that knowledge to new people.
But this also raises a sensitive long-term question: How private will our brain data remain if artificial intelligence becomes increasingly capable of inferring perceptions—and potentially more complex mental content—from it?
Brain-IT is still a long way from being able to secretly read a person’s thoughts. Current experiments require high-quality fMRI scans obtained in a large magnetic resonance imaging scanner. So there can be no talk of an unnoticed “mind scanner.”
The direction of the research is clear, however: artificial intelligence is beginning to translate measurable brain activity back into information understandable to other people with increasing accuracy.
What today is still a reconstruction of an image that was actually seen may therefore be only the beginning of a development that, in the long term, will touch one of humanity’s most intimate frontiers: the processes within our own brains.
