Facebook has just announced a new artificial intelligence-based deepfake detection tool. It not only separates the fake images from the real ones, but also deduces parameters used in their generation to identify the source.
You will also be interested
[EN VIDÉO] The 10 most dangerous threats of artificial intelligence Artificial intelligence has enabled advances in health, research and industry. But in the wrong hands, it can also be used for criminal purposes or for disinformation.
Deepfakes are constantly improving to the point where it becomes impossible for humans to tell right from wrong. These fake images and videos are widely used for disinformation because they allow people to attribute entirely invented actions or words. To fight this scourge, Facebook has partnered with the University of Michigan to create a detection tool of a new kind.
There are already tools based on artificial intelligence capable of identifying the traces left by known deepfake generation models. However, the number of models is steadily increasing, limiting the interest of these detection tools. The one developed by Facebook uses reverse engineering to trace the source of the image. It builds on thedigital print of the image left by the model used. By generalizing this, they are able to detect never-before-seen patterns.
A system that can identify disinformation campaigns
Once the footprint has been determined, the tool is able to discern certain parameters used in the deepfake generation, such as the number of blocks and the type of operations used in each block. It thus creates a representation of the model, even if it is new.
This advance is not limited to the simple detection of deepfakes. Based on this representation, the tool is able to compare a series of images to determine if they are from the same source. This function will allow, for example, social networks quickly detect a disinformation campaign.
Interested in what you just read?
