Deepfakes abound on the web and some are very difficult to identify . The technology behind these fakes allows for increasingly realistic end results, and that's a real concern. Facebook is one of the many companies that is working on creating tools to detect them, and in the last hours it has made noticeable advances.
The social network partnered with Michigan State University (MSU) to develop a method that not only seeks to identify deepfakes, but also to trace their origin . For this it relies on reverse engineering. What does this mean? That from the analysis of an image, technology can determine if it was generated artificially and detect which generative model was used to produce it.
The Facebook method is still under development and is far from a real or mass implementation. In any case, it represents an important step forward in the face of a growing problem.
Facebook wants to be able to find out the origin of a deepfake
The technology behind the creation of deepfakes is constantly evolving, and that makes it very difficult to find clues about its source. However, Facebook maintains that its method can detect subtle patterns that are “imprinted” on the fake images by the device that creates them. As each generative model has its peculiarities, the new technology would identify common characteristics between different deepfakes and establish whether they have a common origin.
While there are already techniques that detect which known model was used to create a deepfake, Facebook goes further. The project aims to identify the “architectural features” (known as hyperparameters) of unknown models, to later distinguish between the patterns that remain in the final image.
“Suppose a bad actor is generating a lot of deepfakes and uploading them to different platforms for different users. If it were a new AI model that nobody saw before, we could say little about it in the past. But now we can say: 'Look, the image that was uploaded here, and the image that was uploaded there, come from the same model. 'And if we manage to get hold of the laptop or computer used to generate the content, we will be able to say: ' This is the culprit ', “Tal Hassner, Facebook's research leader, told The Verge.
It is interesting how research is progressing to detect “fingerprints” in deepfakes and trace their origin, although the technology used to create them is unknown. However, Facebook navigates unexplored terrain , and its specialists recognize that it is a kind of “cat and mouse” game.
