Developing a platform that allows a computer to read and interpret the user's thoughts is a goal many companies are trying to achieve. Elon Musk's Neuralink is working on a special brain-computer interface, but what brings us here today is the project by Guohua Shen, Tomoyasu Horikawa, Kei Majima, and Yukiyasu Kamitani from Kyoto University, who managed to decode thoughts using artificial intelligence. Their decoding far surpasses all previous methods and includes multiple layers of color and structure.
One of the first things Facebook does is ask "what are we thinking". An uncomfortable question if there ever was one, and to be honest, I don't think anyone on the social network really wants to know. However, that same question is much more profound for science, and whoever manages to create an effective bridge between the biological and the digital will probably change the world forever. Consider the possibility for a moment: computers reading thoughts. Accurate interpretation of orders and commands. All those "magical" science fiction interfaces turned into reality. We are still far away, but a team from Kyoto University has just taken a big step.
What we can observe in the animated GIFs is an advanced decoding of thoughts, assisted by artificial intelligence. Previously, machine learning was applied to interpret thoughts and generate images from magnetic resonance readings, but these were always simple concepts, like binary images (black or white) and basic geometric figures. In contrast, the technique by Guohua Shen, Tomoyasu Horikawa, Kei Majima, and Yukiyasu Kamitani decodes much more complex images with defined color and structure.
The technique in detail
Kamitani explains that their previous method interpreted images as a series of pixels, but the human brain does not work that way. In fact, visual processing is hierarchical, extracting multiple levels of aspects or features with variable complexity. The artificial intelligence model is used as a "substitute" for the hierarchical functioning of the human brain.
Another very interesting detail is that the experts managed to decode thoughts based on images "remembered" by the participants, instead of images fed directly. In this case, the artificial intelligence had many problems with reconstruction, because it is more difficult to develop an image from a simple memory. The precision of this technology can only improve over time, and some of its potential applications are as attractive as they are chilling.
