AI can pluck images directly from a person鈥檚 brain. Given an fMRI scan of someone looking at a picture, an algorithm can reconstruct the original image from the scan alone (bioRxiv, ). Although the results aren鈥檛 yet perfect, they are still often recognisable and hint at what may be possible in the future.
Guohua Shen at Japan鈥檚 Advanced Telecommunications Research Institute and his colleagues experimented with three types of images: 鈥渘atural鈥� pictures of things like bats, snowmobiles and stained glass; simple shapes like squares and plus signs; and alphabetical letters.
The shapes and letters were fairly recognisable, but the reconstructions of the natural images tended to be blurry and difficult to parse. You can see the images at .
Advertisement
The team trained their algorithm using 6000 images that were paired with associated fMRI scans. With enough data, AIs like this might make it possible for computers to know what we are thinking about (see main story).
鈥淭hese decoding methods could be used for human-computer interaction in the future,鈥� says Haiguang Wen at Purdue University in Indiana. 鈥淵ou could know what a person is dreaming or thinking just by analysing the exciting brain signals.鈥�
This article appeared in print under the headline 鈥淧icture perfect鈥�