Recording dreams? AI could help us do that using brain scan data

Teaching AI to recreate images using brain scans

Since the algorithm is capable of generating images based on prompts, the researchers focused their attention on making the AI work with brain signals. To do so, they gathered information from different parts of the brain, such as the occipital lobes and temporal lobes that are involved in image perception.

The researchers used data from functional magnetic resonance imaging (fMRI) scans which are sensitive enough to detect differences in blood flow to active regions of the brain. The dataset used for training the AI contained fMRI data of four individuals as they each viewed a set of 10,000 photos.

Brain activity from the occipital lobe provided information to recreate the layout and perspective of the images while that from the temporal lobe provided data about people, objects, or scenery in the image. The researchers then used the data set to form links between text descriptions of the images and fMRI data.

The models were then used to test a portion of the brain scans from the four participants that were not used in training earlier. The researchers found that the AI algorithm could reconstruct around 1,000 images from these brain scans with 80 percent accuracy. This is the same efficiency that other researchers have previously achieved using more intensive approaches.

The models were custom trained for the four participants of the study, making this approach impractical when it comes to scaling it for a large number of users. The researchers are now working on refinements to the methods to make them easier to be adopted by a larger group of users.

When widely available, the technology could be used to record the thoughts and dreams of people and even allow them to create art by simply imagining it. The possibilities are endless.