New study by Kyoto University, Japan: AI reads brain waves and rebuilds human thinking

Researchers at the Japan's ATR International Electrical Communications Apex Technology Institute and Kyoto University in Japan say they created an AI that can read human brain waves. They build a neural network that can not only read, but also reconstruct your mind.

Specifically, according to the Zme Science website, "the team created an algorithm that can interpret and accurately reproduce (reproduce) images or imagined images that a person sees." ”

A structure diagram for deep image reconstruction. The pixel value of the input image is optimized so that the DNN feature of the image is similar to the decoding feature of the fMRI activity. A depth generator network (DGN) can choose to combine with DNN to produce a natural image in which optimizations are performed in the DGN input space. CREDIT:BIORXIV (2017). doi:10.1101/240317

The title of the study, "using deep image reconstruction of human brain activity" (Deep image reconstruction from human brain), wrote that researchers were able to replicate an image based on a scene that a person was observing. These AI created images are not exactly the same as people actually see, but only vaguely represent people's thinking. However, AI can still use brain waves to reconstruct these images.

Although it may be decades before it is actually used, this technology is a step closer to creating a system that can read and understand human thinking.

Deep image reconstruction: Natural images

Trying to tame a computer to decode a mental image (mental image) is not a new idea. In fact, the study has been going on for many years, and researchers have been trying to reconstruct brain images, such as films, photographs and even dreams, since 2011. However, previously all systems were limited in scope and capacity. Some can only handle narrow areas such as face shape, and some can only reconstruct images through predesigned images or categories (e.g., "Bird", "cake", "person", etc.). So far, all of these technologies require stored data; the way they work is to match the brain activity of the subjects with those previously recorded in the image.

Read the original