The Limited Times

Now you can see non-English news...

Manipulated videos: Facebook wants to delete deepfakes - if they are serious

2020-01-07T10:47:12.021Z


If Melania Trump suddenly looks like Vladimir Putin: Facebook has announced new rules for dealing with so-called deepfakes. The time for this was probably chosen deliberately.



Facebook sets itself new guidelines for dealing with so-called deepfakes: According to its own statements, the network wants to take videos from the platform in the future, which are processed in such a way that an "average person" does not notice it and people could believe that someone said things that weren't really said. On the other hand, technical processing that serves to improve the sound and image quality is still permitted.

Specifically, according to Facebook, it's about videos in which, for example, using artificial intelligence or machine learning, content is merged, replaced or overlaid so that the result looks authentic.

With such deepfakes, software replaces faces in videos with others, for example. Or words are put into the mouth of people who have not said them (a detailed article on the topic can be found here). In this way, funny videos can be created - for example, if you have your own face inserted into a Hollywood film scene - but just as well problematic clips, for example in the context of porn films or falsified political speeches.

"Great challenge for our industry"

In his blog post, Facebook describes Deepfakes as "a major challenge for our industry and society if its use increases". However, these videos are still rare on the Internet.

How exactly Facebook wants to identify deepfakes as such, which is often not that easy, is largely left open. He only refers to an industry initiative called Deepfake Detection Challenge and a collaboration between Facebook and the Reuters news agency.

Facebook moderators are also likely to face some challenges with deepfake video material, because the new directive provides for exceptions for parodies and satire, which leaves room for interpretation.

Depending on the definition, the satire category, for example, could include the deepfake clip, in which Mark Zuckerberg once declared himself to be the world ruler. It was uploaded to Instagram by artists Bill Posters and Daniel Howe in summer 2019. It was an excerpt from a 2017 Zuckerberg video talking about Russia's attempts to influence the US election. But in the Instagram clip, Zuckerberg seemed to say something puzzling: "Imagine that: A man with total control over the stolen data of billions of people, all their secrets, their lives, their future - I owe it all to Specter . "

Elections are held in the United States in November

Facebook's decision to delete certain deepfakes comes as no surprise. On Wednesday there will be a hearing in the American House of Representatives on the subject of "manipulation and deception in the digital age", at which a representative of Facebook will presumably ask critical questions. So it is certainly helpful if she can refer to newly adapted rules that show that deepfakes have already been tackled.

In addition, there will be the US presidential election in November 2020, in the course of which numerous people from politicians to secret services will keep an eye on how Facebook is used to influence voters.

Last year, Facebook had clashed with the Democrats, among other things, after the network refused to delete a rigged video that House Representative Nancy Pelosi showed. With the clip, the playback speed of the recording was reduced and then the pitch was adjusted - Pelosi looked like drunk.

Would the Pelosi video be deleted by now?

How exactly Facebooks intends to deal with such material in the future cannot be read from the blog post on Deepfakes, since in the Pelosi case material was manipulated technically, but no other words were put into the politician's mouth. Machine learning also played no role in this manipulation.

Facebook generally refers in its blog post to its factchecking system, through which various external partners of the network could classify photos and videos as "wrong" or "partially wrong". These then appeared significantly less prominently in the news feed, it is said. And if you want to share content marked in this way, you will get an indication that it is problematic content.

Source: spiegel

All tech articles on 2020-01-07

You may like

Trends 24h

Latest

© Communities 2019 - Privacy

The information on this site is from external sources that are not under our control.
The inclusion of any links does not necessarily imply a recommendation or endorse the views expressed within them.