New AI ‘Brain Decoder’ Reconstructs What People Are Viewing From Brain Scans
Scientists at Israel’s Weizmann Institute of Science have built an AI tool that can look at a brain scan and reconstruct, with striking precision, the image a person was viewing, according to a report by MIT Technology Review.
The “brain decoder”, developed by Michal Irani and her colleagues in Rehovot, analyses functional magnetic resonance imaging (fMRI) data to recreate the picture a subject was looking at. It also works in reverse: given an image, it can predict what that person’s brain activity would look like.
Earlier brain-decoding systems could generally identify the subject of a picture but failed to reproduce its structure and position. The Weizmann tool uses two branches — one predicting an image’s structure and colour placement, the other identifying its content — and feeds those predictions into a diffusion model that reconstructs the image.
The researchers trained the system on high-resolution fMRI data from eight people who each viewed about 9,000 images. An accompanying “brain encoder” let them effectively expand the training set by predicting brain activity for images never actually shown in a scanner; roughly 70 per cent of the training images were not originally paired with scans.
The researchers hope the tool will reveal more about how the brain works and could one day help locked-in patients communicate, according to MIT Technology Review. Other scientists warn that similar techniques could eventually expose people’s mental imagery without consent — although for now the technology requires a large scanner and a willing volunteer lying still inside it.
Sources
- MIT Technology Review, “An AI ‘mind-reading’ tool can reconstruct what you’re looking at based on a brain scan” (1 October 2026): https://www.technologyreview.com/2026/10/01/1145588/ai-mind-reading-reconstructs-what-youre-looking-at/
- Weizmann USA, “The New Science of Mind-Reading”: https://weizmann-usa.org/news-media/news-releases/the-new-science-of-mind-reading/