And you, would you fell into the trap? Since Thursday evening in the United States, a video of the figure democrat and speaker of the House of representatives Nancy Pelosi has become very viral. You see the elected u.s. on stage during a speech at the American Center for Progress. Instead of expressing himself clearly, his speed is slow, and the political woman seems to be struggling to articulate. Very quickly, members of the Republican party, relatives of Donald Trump as his lawyer staff Rudy Giuliani or supporters of the alt-right american have shared the video originally published on the account Facebook conservative Politcs Watch Dogs. These latter accusing Nancy Pelosi of being "drunk" or have "addiction problems". Less than two days later, the extract has been seen more than 2.3 million times and shared nearly 50,000 times. Problem? The video is not authentic. The Washington Post has analysed and has been able to establish that it was 75% slower than the original version, thus creating a distortion of the word of the elected democrat. "Not that the tone of the voice is altered, the sound also seems to have been modified" note the daily american. YouTube has decided to remove all copies of the video on its platform. For its part, Facebook has responded and said that the video was the subject of an investigation. "If it is considered misleading, and its scope will be voluntarily reduced within the news feed," said the social network.

" SEE ALSO - "The videos are faked ("deep fake") are the great challenge of the future for technologists"

" The videos are faked ("deep fake") are the great challenge of the future for technologists " - Look on Figaro Live

Difficult to know who is behind this virality instant, if it was organised beforehand by opponents of Nancy Pelosi or if she just has naturally affected millions of people. Its publication occurs in a particular context. The president of the United States is in open war with the head of the democratic majority in the House of representatives. "I'll tell you, I have watched in a long time, it is no longer the same person, she is lost," said Thursday Donald Trump before addressing his opponent as "mad". He took the opportunity to share another video, pinned on his Twitter account, which is extremely well fitted for mock up once more to the democratic party. She had also been targeted at the beginning of the month of may, during a speech before the association of manufacturers of road and u.s. carriers. Once again, the speed of his speech had been deliberately slowed down and the video had been viewed nearly 28 million times compared to only 200,000 for the original.

This type of videos, change at the margin and widely shared on social networks, are multiplying in the United States and Nancy Pelosi is not the first to pay the costs. In November, a collision between a journalist of CNN and Sarah Sanders, an assistant of the White House had been deliberately slowed down to make it look like a violent. The extract has been used to justify the ban on the press conferences of Donald Trump's the journalist in question.

The fact that these videos are little changed, only slowed down, can meet such a success on the Web is an issue. "Now, the same manipulations simple and crude can be used to undermine an opponent or to score political points" notes the Washington Post. "It is simply amazing that a simple change can have as much impact and can be believed by so many people. Too many of us are ready to believe the worst if it is delivered by individuals who do not share our opinions," said Hany Farid, a researcher and expert in digital forensics at the university of california, Berkeley interviewed by the Washington Post.

Behind the light fixtures cheap flat a threat far greater: that of the deepfake sophisticated. That is to say the possibility for a software-doped artificial intelligence to change the voices and the faces of the protagonists, with no note. A video of Barack Obama, discusses the president Donald Trump "idiot total and absolute" and shared knowingly by the media, Buzzfeed in may 2018 for a demonstration by the absurd had highlighted this phenomenon. More recently, researchers in artificial intelligence working for Samsung have unveiled the fruit of their research, and they are disturbing. They have been able to create a sort of avatar be able to move and speak from a single portrait photo.

Previously, the deepfake was reserved for researchers or engineers emeritus. The technique of "networks adversatifs generators" (GAN) was the most prevalent. It consisted of a competition of two algorithms to obtain the best imitation possible. One copy of a first video by adding a new face while the other judge of the credibility of this replacement and trying to improve it. Other technological tools allow you to reproduce the movements of the lips or the body.

The last example from Nancy Pelosi is more than troubling because it shows the extent to which changes very cheap, far the latest advanced technology, enough to fool a very large audience.