By akademiotoelektronik, 07/11/2022

Why correctly identify a Deepfake video is a miracle

Would you be able to say if a video is a deepfake with a naked eye?The answer is probably not if we believe a study carried out by British researchers from the University of Oxford, the Brown University and the Royal Society.Let us first remember that these are artificial intelligence techniques allowing to create false videos or very realistic vocal extracts, and in particular to make someone say the words that he did not make.

To see if Internet users are able to recognize a deepfake, scientists separated the participants from their research in two groups.The first watched five unsuccessful videos, while the second watched four videos without special effects, as well as a deepfake.Then they had to detect which content is a false.

An alarming constant for the future

Deepfake used hit the column in March 2021.It was a false video of Tom Cruise created by the artist Chris Ume, and broadcast on Tiktok.Very realistic, we can notably see the actor tell jokes, or perform certain magic tricks on short formats and adapted to the Chinese social network.

The results are final, since even if the participants were warned that a deepfake was in the sample, 78 % of them could not spot it.It even happened that those to whom the false video was directly pointed out, refuse to believe it.

Cited by The Independent, researchers explain: "Individuals are not more likely to notice anything unusual when they are exposed to a Deepfake video with neutral content".

They add: "The difficulty of manually distinguishing the real videos of the false threatens to fully reduce the informative value of the video media.As people internalize the capacity of the Deepfakes to deceive, they will rationally give less confidence to all online videos, including authentic content.»»

A particularly alarming observation in particular with regard to the fight against disinformation.This loss of confidence would also be an obstacle to the current labeling strategy, where social networks indicate the fact that content has been changed.Furthermore, we also know that Deepfakes automatic detectors are still very largely handling, and we cannot therefore count on them to solve the problem.

Les meilleurs smartphones 2022 Dernière vidéo Abonnez-vous