Facebook is studying a new method to recognize deepfakes

Thanks to a reverse engineering system, still in the research phase, fake videos can be detected starting from the common traits of the Ai that generated them

(photo: Unsplash) Although it is not yet such a threat Facebook is continuing to invest in improving deepfake recognition systems on its platforms. Deepfakes are images or videos in which faces, but also voices, are superimposed and combined with existing scenes, thanks to artificial intelligence (Ai), thus creating real fakes.

The The latest work sponsored by Facebook is the result of collaboration with scholars from Michigan State University (MSU) and has made it possible to create a method for decoding deepfakes with reverse engineering techniques. Thanks to this new model, it is possible to analyze the images generated by an Ai and, by means of some parameters, identify the characteristics generated by the automatic learning models, even the unknown ones. With this method there is the possibility of identifying a sort of fingerprint that allows you to indicate whether a movie is true or false.

At this moment, the work is still being researched and is not ready for be distributed. However, the research could help Facebook identify deepfakes shared on their social networks in fturo. This content can include disinformation but also non-consensual pornography, a very common use of deepfake technology. In this case the faces of the actors and actresses are replaced with those of celebrities or well-known people.

How the new method works is well explained to The Verge by Facebook researcher, Tal Hassner. “Let's assume an attacker is spawning many different deepfakes and uploading them to different platforms for different users,” says Hassner. "If this is a new model of artificial intelligence that no one has seen before, then there is very little we could have said about it in the past." Now we are instead able to say: "Look, the image that was uploaded here, the image that was uploaded there, they all come from the same model", in that case we can say: "This is the culprit", he Hassner explained.

Preliminary information on this research certainly shows progress in the efforts to discover deepfakes, but it must be emphasized that these models are still the alder of infallibility. When Facebook held a deepfake detection contest last year, the best algorithm detected only 65.18% of the videos manipulated by artificial intelligence.

This difficulty is mainly due to the fact that the field of Generative ai is extremely active and progresses quickly. New techniques are published every day and it is almost impossible for any filter to keep up, as Hassner himself confirmed.


Media - 9 hours ago

The new European copyright directive is reality, but all states are late

adsJSCode ("nativeADV1", [[2,1]], "true", "1", "native", "read-more", "1"); Social Network - 23 hours ago

Facebook first tested its Clubhouse in the US

adsJSCode ("nativeADV2", [[2,1]], "true", "2" , "native", "read-more", "2"); Tech - Jun 16

Using deep learning to turn a photo into a video

Topics

Deepfake Facebook Artificial intelligence globalData.fldTopic = "Deepfake, Facebook, Artificial intelligence"

This opera is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.