A team led by Michal Irani at the Weizmann Institute of Science has built an AI system that can guess what someone is looking at by analysing a brain scan and then generate a picture that closely matches the original. The system works both ways: it can also predict the brain activity that would result from viewing a given image. The breakthrough was shown at the Cognitive Computational Neuroscience conference in New York in September 2026.
How the system works
The researchers trained a “decoder” on fMRI data from eight volunteers who each saw about 9,000 images in a high‑resolution scanner. The decoder has two branches – one predicts the layout of colors and shapes, the other predicts the content such as “a bunch of bananas”. A diffusion model – a type of AI that refines noisy pixels into a clear picture – then creates the final image. To overcome the limited amount of paired scan‑image data, they also trained an “encoder” that predicts brain activity from any image, letting the two models improve each other in a loop.
Why it matters
The new decoder needs only about one hour of fMRI data from a new person, whereas earlier tools required roughly 40 hours – a cost difference of $600‑$1,000 per hour of scanning. This makes the technology more practical for neuroscience labs and could eventually help people who cannot speak, such as those with locked‑in syndrome, to communicate by turning brain activity into pictures.
Why it matters
For neuroscientists, the reduced scanning time means studies can be done cheaper and faster, opening up more experiments on visual processing. For patients with severe motor impairments, the method hints at a future way to convey thoughts without speech, though practical devices are still years away.
