Although the use of deepfakes has become popular in areas such as advertising, these are still a problem to consider in terms of disinformation, through social networks. For this reason, Facebook has funded research with various academic institutions to protect itself against these threats from their origin, through Artificial Intelligence (AI).
Deepfake montages are images or videos in which the face, or even the voice, has been manipulated to introduce certain characters in contexts that do not correspond to them. Although they have been used in entertainment contexts, they have malicious uses.
In this sense, the most recent work by Facebook is a partnership with Michigan State University, whose team of academics has created a method to reverse engineer these types of files, that is, they analyze the images to identify the characteristics of the model. machine learning that created it.
Although reverse engineering is not an innovative approach in the field, the developers of the technology and the company defend its use within this field, since they allow discovering the fingerprint in the final image, which is a key element to identify its source. They are “subtle but unique patterns,” describe researchers Xi Yin and Tal Hassner, in one.
The technology, the specialists explain, is inspired by forensic techniques used to identify which camera model was used to take a photograph from the patterns in the image.
Likewise, they highlight that identifying the features of models to perform deepfakes are relevant, as this opens the possibility of discovering the properties of the model that has generated the false image, and on the other hand, it opens the possibility of tracing similarities between a set of deepfakes. distributed in the network.
With this model “now we could say: ‘Look, the image was uploaded here or there, and they all come from the same model,” Hassner told The Verge, a medium to which he also said that in this way it would be easier to identify those responsible for the spread of deepfakes.
It should be mentioned that the project is still in a research stage and has not been deployed, but the initiative is part of the technology’s efforts to prevent the spread of different forms of disinformation within its platform.
In fact, the same company tested various tools for this purpose and the most relevant one only managed to detect manipulated videos 65.18% of the time, which shows that the identification of this type of files continues to be a problem for companies. social networks.
And it is that although the technology is not new, and has even been used by millions of users around the world within applications to change their age or gender, its malicious uses are related to misinformation or even harassment on the internet.
Given this, other companies such as have established agreements with the media and social networks, including Twitter, to develop security protocols in which the origin and authentication of the content is defined, in order to “restore trust in digital content. through methods that authenticate sources and track the evolution of the information we consume ”.
