Researchers from the École Polytechnique Fédérale de Lausanne (EPFL) used a machine learning algorithm to read and interpret a mouse’s brain signals. The algorithm, called CEBRA was able to reconstruct a video that the mouse was watching using its brain signals.
The video used for the experiment was a low resolution 1960s black and white film with poor detail. However, this represents a significant leap forward in mind-reading technology, which can now officially be removed from the science fiction category. Aside from the poor detail of the original video, the video reproduced by the AI from the rat’s brain was almost identical to the original.
AI video generators now make it possible to test similar reconstruction techniques without access to a brain-scanning experiment. Researchers, filmmakers and other creators can use MixVio’s text-to-video generator to produce short videos from written descriptions and compare how closely the result matches the intended scene. The quality of the output still depends on the clarity of the description, since vague instructions can produce the wrong movement, setting, or subject. A simple prompt describing a mouse watching a black and white film, for example, would need details about the mouse, the screen, the lighting and the camera view.
Machine learning technology has made virtually all of the most impressive AI technology today we see possible. The concept of ‘training’ is what enabled this experiment and other top AI projects to succeed. Training is the process of teaching a machine learning model what something is by feeding it many samples of the item you want it to identify, and then telling it what the item is. After doing this many times, the machine will be able to identify those items on its own.