Researchers at the Weizmann Institute of Science developed Brain-IT, an AI system that reconstructs recognizable images from a person’s brain activity while they view pictures inside an MRI scanner. Functional MRI scans captured activity patterns as participants viewed different images, and the system produced reconstructions of objects, food, faces, landscapes and more complex visual displays.

Brain-IT does not retrieve thoughts at random: a participant must be in the scanner and looking at an image for the system to estimate what they saw. Researchers focused on how quickly the system could adapt to new participants, since older methods typically needed many hours of brain-scan data from each person before producing useful reconstructions.

Brain-IT reportedly needs approximately one hour of learning about a new participant. To achieve this, researchers focused on patterns shared across people’s brains rather than treating every brain as completely distinct. They divided brain activity into tens of thousands of small components and matched signals to general visual characteristics, including color and location, as well as higher-level categories such as faces, food and sports.

Researchers at the Weizmann Institute of Science developed Brain-IT, an AI system that reconstructs recognizable images from a person’s brain activity while they view pictures inside an MRI scanner. Functional MRI scans captured activity patterns as participants viewed different images, and the system produced reconstructions of objects, food, faces, landscapes and more complex visual displays.

Brain-IT does not retrieve thoughts at random: a participant must be in the scanner and looking at an image for the system to estimate what they saw. Researchers focused on how quickly the system could adapt to new participants, since older methods typically needed many hours of brain-scan data from each person before producing useful reconstructions.

Brain-IT reportedly needs approximately one hour of learning about a new participant. To achieve this, researchers focused on patterns shared across people’s brains rather than treating every brain as completely distinct. They divided brain activity into tens of thousands of small components and matched signals to general visual characteristics, including color and location, as well as higher-level categories such as faces, food and sports.