Yesterday, Twitter officially released the first anti-Deepfake strategy draft, which is currently collecting public comments.
Editor’s note: This article is from WeChat public account “Heart Heart” (ID: Almosthuman2014), original Synced, author Zhang Qian, egg sauce.
Twice yesterday, Twitter officially released the first anti-Deepfake strategy draft. The company said that if Deepfake content threatens someone’s personal safety or could cause serious harm, they will remove the content from Twitter. Currently, Twitter is still collecting public comments on this draft.
Remember the face-changing app “ZAO“? Once the app is launched, it detonates the social network, pulling the face of the face to a new low, and you can easily convert your face by downloading an APP.
It’s a good time to change your face, but once you are abused, you can’t always be cool. A investigation report shows that 96% of the current Deepfake videos are still pornographic, which will cause great problems for women. In addition, politicians in the United States and other countries are also worried that Deepfake will be used. We have been very vigilant in discrediting politicians and intervening in general elections, and urged major technology giants to introduce corresponding measures.
In response, yesterday, Twitter published a version of the details on combating the Deepfake strategy draft.
This social media says that when someone finds a Deepfake file that tries to guide the public or confuses them, they do the following:
-
Also remind you next to public tweets that involve Deepfake videos, images, and more;
-
Warning prompts before users attempt to “share” or “like” Tweets in Deepfake files;
-
Add a link (pointing to a news story or Twitter Moment) so people can refer to more sources to determine if this is indeed a Deepfake file.
Twitter said that if a Deepfake content threatens someone’s personal safety or can cause serious harm, they will consider removing the content.
At present, the company is conducting a survey to solicit feedback from users. The survey results show that users believe that if a Deepfake content poses a threat to someone’s mental health, privacy, dignity, property or other aspects, the content should also be deleted.
This tube is deepfake
With the blessing of AI technology, the number of things like Deepfake video has soared, only doubled last year. This makes it easier for people to make audio and video files that are fake, but when this technology becomes available to everyone, things get more and more dangerous. Because it leads to “not necessarily seeing the truth.”
At the end of 2017, Deepfake just got on the Reddit, netizens got together and there was a lot of AI face-changing videos in a short time – but the problem was that the “Deepfakes” community slowly became porn. The gathering place was finally completely closed.
In July of this year, there was a section called “DeepNude was once hot, but it was quickly removed. Its function is simple but terrifying: APP can use the neural network to automatically “take off” the clothes on the female in the photo to reveal its nude.
In addition to the above undescribed applications, Deepfake is politically easy to use by unscrupulous people.
For example, in a political election campaign, Deepfake can make a document of something that a candidate has never said or done, unknowingly or distorting the facts and deliberately undermining the candidate’s reputation.
A few months ago, a speech by Nancy Pelosi, a US Democratic congressman at the Center for American Progress, was rumored on Facebook. In the video, Pelosi’s speech rate was reduced to about 75% of the original version, and the tone was also artificially raised, yin and yang, like a drunken. This spoof video has had a bad influence on the psychology of the parties. But at the time, Facebook refused to remove the video.
This incident triggered US politicians’ dissatisfaction with these large social media. It is reported that after Facebook decided to keep Pelosi’s “drunk” video, Zuckerberg tried to call Pelosi to discuss the response to social media fake video, but was rejected by Pelosi. Since then, the US government has repeatedly called on social media to strengthen supervision of content such as fake videos.
In early October, two members of the Senate Intelligence Committee, Mark Warner (D-VA) and Marco Rubio (R-FL), called on large technology companies (including Facebook, Twitter, YouTube, Reddit and LinkedIn) to work together to develop a Plan to team up against D on the platformEpefake. In line with this plan, Twitter announced its next step later in the month and sought public feedback on the strategy, so we can see the draft announced yesterday.
In addition to Twitter, other tech giants are not idle
In fact, in addition to Twitter, technology giants such as Facebook and Amazon Google have also taken a series of measures to counter the proliferation of Deepfake fake videos.
In September of this year, Facebook announced that it will spend $10 million to launch a Deepfake Testing Challenge with Microsoft, Partnership on AI and several universities. They test the effectiveness of the detection tool by using fake videos generated by tools such as GAN.
In October, Amazon also announced support for the challenge, providing “technical support and guidance” to machine learning experts, and will provide $1 million in AWS points to teams that need cloud services to complete inspections. In addition, they will also provide data sets and other help for the competition. In addition, the project involves scholars from Cornell University of Technology, Massachusetts Institute of Technology, Oxford University, the University of California at Berkeley, the University of Maryland, Parker, and the State University of New York, Albany.
Of course, in the face of this kind of charity event, Google, who is not doing evil, is not far behind. In October of this year, Google announced that the open source large Deepfake video dataset< /a> to support community research on Deepfake detection. In order to make this data setIn the past year, Google has worked with hundreds of paid and unpaid actors to shoot hundreds of videos.
Then, they created thousands of Deepfake videos based on these videos using the publicly available Deepfake generation method. These real and fake videos together form the Deepfake video dataset.
Even though Deepfake represents an advancement in the image synthesis technology of people – but at a stage where supervision and review are not enough, everyone is still facing “the muddy sand.” Before we get on the right track, we may have to go through several rounds of games with Deepfake.
Reference link:
https://techcrunch.com/2019/11/11/Twitter-drafts-a-Deepfake-policy-that-would-label-and-warn-but-not-remove-manipulated-media/