Research AI Rebuilds from Brain Scans What People See

A thought remains hidden until it is spoken. A new AI tool shortens that path. Based solely on the analysis of brain scans, it recognizes what a person is looking at and generates the matching image on that basis, with what the report describes as remarkable precision.

The tool was developed by Michal Irani together with her colleagues at the Weizmann Institute of Science in Rehovot, Israel. The brain scans come from magnetic resonance imaging (MRI), a method that makes brain activity visible through changes in blood flow. The model is trained on such data until it recognizes connections between brain activity and images. Irani presented the tool, which she calls the “Universal brain encoder,” last month at the “Cognitive Computational Neuroscience” conference in New York.

What sets it apart is the direction it works in. Most brain research methods measure what is happening in the head. This tool can also work the other way round: from the image a person is looking at, its brain activity can be predicted. The encoder is meant above all to shed more light on how the brain processes sensory impressions. The report explicitly emphasizes this basic function rather than a finished application.

What this could become in practice is open, and the report deliberately presents the possibilities as hopes. People with locked-in syndrome, who can barely move or speak at all, might perhaps be able to communicate again with such a method. Scientists could also try to reconstruct the content of dreams. Neuroethicist Judy Illes, who was not involved in the work, calls the therapeutic approach “incredibly exciting.” At the same time, a brain that can be read raises the question of mental privacy, to which the report devotes a separate section. The report does not say when, or whether, the method will make its way into clinics.