Recently, an artificial intelligence face-changing app called ZAO was launched in the country. Users only need to upload a frontal photo and can operate with one click, change the faces of actors in some film and television clips to their own, generate videos immediately, and spr

Recently, an artificial intelligence face-changing app called ZAO was launched in the country. Users only need to upload a frontal photo and can operate with one click, change the faces of actors in some film and television clips to their own, generate videos immediately, and spread them through social media, as if they instantly fulfill their dream of becoming a star. The face-changing APP

quickly became a hit and was very popular, but it also quickly attracted the attention of relevant departments. Just this week, the Ministry of Industry and Information Technology summoned relevant companies on network data security issues.

At the same time, an AI technology based on the "face-changing" software has also attracted people's attention. It is called: DeepFake, deep fake.

Foreign media reports: We cannot believe everything on the Internet, yes, deep forgery, deep forgery! Yes, what exactly is this? Deep forgery technology is to use artificial intelligence to synthesize fake videos. This is a high-tech means that can stuff certain words into someone's mouth.

Deepfake, a new word composed of "Deep learning" and "Fake".

Simply put, this is an artificial intelligence technology that can fake facial expressions in real time and render them into 2D synthetic videos.

As more than a year ago, in December 2017, a user named "DeepFakes" released a "fake video" on the Reddit website, replacing the faces of actors in adult pornographic movies with those of well-known actresses such as Scarlett Johnson, and the pictures are fake and real. This has attracted a large amount of traffic for artificial intelligence face-changing technology. "DeepFake" is deeply faked and has gradually become the synonym for this technology. The algorithm of the same name is also open source on Github, which is a hosting platform for open source and private software projects.

Online video anti-counterfeiting expert Dr. Wardle: Hello, today I want to discuss with you a new technology that will influence celebrities. (Remember) Obama said Trump is a fool, or Kardashian's rap "because I'm always half-naked?" Deep forgery! Deep forgery! Deep forgery! This is also deep fake. I am not Adele, I am an expert in (striking) network fakery. Deep falsification is used to describe video or audio files synthesized by artificial intelligence technology. It started out as a very basic facial change, but now it has become a movie stunt. With such a hit technology, my goodness, we can't believe anything. Yes, deep forgery is a terrible dystopia. They will only become easier and cheaper to make.

Li Hao is an assistant professor at the University of Southern California and a co-founder of "Needle Animation". The company launched software that allows users to instantly customize their three-dimensional virtual avatars while playing virtual reality games and shopping.

"Needle Animation" Co-founder Li Hao: Now I have made a incarnation of you.

American Public Broadcasting Corporation reporter O'Brien: A kind and neat O'Brien.

This trick can be used on anyone, and of course it also includes important people in the political circle.

American Public Broadcasting Corporation reporter O'Brien: Now I become our president (Trump), Japanese Prime Minister Shinzo Abe.

"Needle Screen Animation" Co-founder Li Hao: Of course, this technology can be used to do something really bad. But the main purpose is by no means the same. It is used for entertainment, an interesting tool for fashion and lifestyle, bringing us more experience.

However, with the development of technology, such synthetic images or videos are becoming more and more realistic and deceptive. The developers are not without concerns about this. Li Hao, co-founder of

"Needle Screen Animation": We all assume that there will be a critical point, and then it will be impossible to distinguish between true and false. I mean, visually, I think you can already do it very close, it just depends on how much effort you spend on it. But as far as content anyone can create, I think it's very close to this tipping point.

In January 2018, a software called FakeApp was launched, claiming that it can achieve "one-click face change".The videos that can be generated include spoofing US President Trump or sticking his face to Hollywood stars.

Although some social news sites such as Reddit websites have explicitly prohibited the dissemination of face-changing videos and pictures on their own platforms, more than 90,000 users still spread such videos on Reddit.

Computer Science Expert Farid: I'm worried about the weaponization of this technology and how it affects our entire society.

Deep forgery can not only change faces, but also fake sounds.

According to the Wall Street Journal, in March this year, some criminals used "deep forgery" technology to synthesize the voice of the CEO of a certain company and successfully defrauded 220,000 euros.

Is it really so difficult to see through the synthesized sound of computers?

In 2018, three doctors from the University of Montreal co-founded a company called "Lyrebird". The company has developed a "voice synthesis" technology. As long as you record the voice of the target person for more than one minute and throw the recording to the "Piano Bird" for processing, you can get a special key. Using this key, you can use the voice of the target person to generate anything you want to say.

Founder of "Pinbird": You need to record your own sound for several minutes.

Bloomberg reporter Vance: Thousands of letters are beating on the screen of amateur writers. When you start eating like this, there will be problems. You better quit the political circle and stop doing it. I don't know how it works and now create my digital sound. It takes at least one minute to create your digital sound, one minute, my god. After recording the

voice, the "Pinbird" will automatically run and synthesize your digital sound.

At this time, you only need to enter what you want to say to the computer and you can say it in your newly synthesized voice.

Reporter Vance's digital synthesized voice: Artificial intelligence technology seems to be developing very fast, should we be afraid?

Bloomberg reporter Vance: I did hear it, it's really interesting. I just picked something I never said.

Moreover, "Pianobi" can also add "emotional" elements to the synthetic sound, making the sound sound more realistic.

Bloomberg reporter Vance: Now, in order to experiment with my computer synthesized sound, I want to call my dear mother to see if she can hear it. Hi, mom, what are your arrangements today?

s' mother: The house was out of power this morning, and we were wandering around the house.

VS: I just got off work and was waiting to pick up my son home.

Wans' mother: OK.

VS: I think I may be infected with the virus.

Wans' mother: Then you feel uncomfortable, right?

VS: Haha, I was just making trouble with you, you were just talking to the computer.

Vens's mother: I feel like talking to you, it's amazing.

VS: Do you think it’s scary or good.

VS's mother: If something very important involves, it is really scary. But now it's you, right?

VS: I don't know.

VS's mother: sounds like you.

VS: is it?

VS's mother: Yes, yes, it sounds like you.

Bloomberg reporter Vance: Obviously, some people are frightened by this technology. Because we have made the truth blurred.

"Pinbird" founder: Of course, there is indeed a risk that someone will use this technology to do bad things. However, technology cannot stop developing, so we decided to choose ethical practices to show technology to people, let people know the feasibility of this technology and make them more alert.

Cybersecurity company Pindrop conducted an online survey of 500 companies in the United States, Britain, Germany and France in May 2018. The results showed that from 2013 to 2017, the number of voice fraud cases increased by 350%, and 1 out of every 638 fraud calls was artificial synthesized sound.

Researcher at the University of Zurich, Switzerland, Udrich: For a long time, human voices have been the biggest challenge for (counterfeiters), because each sound is extremely complex and unique, and they are almost impossible to be faked. But in recent years, tremendous advances have been made, and the combination of video material and fake sounds has posed a huge threat.

Indeed, with mature sound counterfeiting technology and fake images, it is really not difficult to make a video that is fake and has ulterior motives.

"Deep Fake" fake video: President Trump is a complete fool. You see, I will never say these words, at least in public speeches (not), but others will say, such as Jordan Peele. This is a dangerous era, and we must be more vigilant about the Internet we trust in the future. In this era we need a credible source of news, which sounds simple, but how we should develop, the information age will determine whether we survive or become some kind of bad dystopia. Thanks.

This video looks like a speech by former US President Obama, but in fact it is a performance by comedian and filmmaker Peel, with the sound and pictures synthesized through software.

Computer Science Expert Farid: Artificial intelligence system synthesizes Obama's mouth to align with the audio stream, making the video look like President Obama said something he had never said before. This is called lip-synchronous depth forgery.

technology has no good or evil. Deep fakery can be used in interesting videos, games and advertising cooperation, but malicious use can also bring serious consequences. It may destroy our perception of reality and make people dare not believe in videos or videos anymore.

The Carnegie International Peace Foundation also warned that deep forgery is having destructive effects and will also cause harm to counter-terrorism and national security.

Computer Science Expert Farid: The nightmare situation is that a video of President Trump came out, in which he said, "I fired a nuclear weapon against North Korea." Someone hacked into his Twitter account, and the news was spread wildly. In just 30 seconds, a global nuclear war broke out. Do I think this is possible? No, but it's not entirely impossible, and it will scare you to death, right? But this is not impossible, and it is really worrying.

From "deep learning" to "deep forgery", photos, sounds, and videos can be easily faked, while identifying deep forgery is much more difficult than making deep forgery.

When "seeing may not be true, listening may not be true", what should we do?

Zuckerberg "Deep fake" fake video: This is all attributed to the Spectre vulnerability, which tells me whoever controls the data will control the future.

In June this year, such a video appeared on the image sharing website Instagram. Zuckerberg, founder of Facebook , gave a speech on the power of big data.

But soon, Zuckerberg denied the rumors that he had never said these words. In fact, this was synthesized by an Israeli technology company using deep forgery technology.

A little embarrassing is that Facebook once claimed that the company will not delete fake news, but will reduce its influence on the website and display information from fact-checking personnel.

Instagram president Mosri also said in an interview that there is currently no large-scale data or standards in the industry to detect these fake videos.

Instagram President Mosri: At present, we have no policy to deal with deep forgery, and we are trying to evaluate whether we are willing to do so. If you want to do it, how should you define deep forgery? I don't think (this kind of incident) is good.

Host Gail King: You can limit this technology and you have influence.

Mosri: Actually, what I want is not just to pull it off, I think the problem is how we do this in a principled way.

In response to this, Fortune magazine called for the time being for technology companies, academia and government to work together to find solutions.

On June 13, the U.S. House Intelligence Committee held a hearing on deep forgery of artificial intelligence, and publicly talked about the risks brought to the country, society and individuals by in-depth forgery technology, as well as the prevention and response measures that should be taken.

At present, many countries around the world have also begun to carry out legislation to strengthen constraints on the application scenarios of facial information collection and recognition technology. But all efforts are inseparable from the cooperation of social media platforms.

Computer Science Expert Farid: But in most cases, social media platforms seem careless about how to deal with such illegal content, harmful content, false information, fake news, election tampering, involuntary pornography, etc., because these contents attract attention and are good for business.

Recently, technology companies such as Facebook, Microsoft and many academic institutions jointly launched a "Deep False Image Challenge", hoping to improve existing tools and strengthen the detection capabilities of deep falsified pictures, sounds and videos.

Researchers at the University of California, Berkeley are also studying how to deal with deep forgery.

Computer Science Expert Farid: The way we deal with this problem is to establish a soft biometric model. These biometrics refer to fingerprints, iris, and face. It is not that unique, but its purpose is to capture subtle facial expressions, head movements, characteristics unique to each individual but disrupted when making fake videos.

Computer Science Expert Farid: First of all, we will measure various head movements and facial expressions. You can see in this video that we are capturing a blue box, capturing how his head rotates in three-dimensional space. The red dot is positioning his facial expression, and we can actually see when he will raise his eyebrows and open his mouth wide. Of course, the green laser in his eyes will tell us where he is looking. In every frame of the video, we will make some measurements of his facial expressions, facial movements, and head movements, and then we use these to build a soft biometric model.

Computer science expert Farid sighed that more and more people are making fake videos, and the technology of counterfeiting is developing very rapidly. In contrast, the strength of the anti-counterfeiting team seems backward and weak.

There is a research team in the US Pentagon, also known as the US Defense Advanced Research Projects Agency, which has been studying how to resist the threat of "deep forgery".

Computer scientist Turek: This makes us distrust pictures and videos.

, software that detects fake videos can describe lips movement and compare it with the sound in the video.

Computer scientist Turek: When you see these red dots, it means that the speaker's voice is actually inconsistent with the movement of his lips.

Let’s watch this video again. It seems that the two of them are sitting together, but by measuring the light angle on their faces, it can be determined that this is a synthetic video.

Computer scientist Turek: It estimates a 3D model of the face with this model, and the software also estimates facial reflection characteristics and lighting angles. Here, we mainly use the lighting angle to see whether those (features) are consistent.

This is a picture in the surveillance video, and the detection software tries to predict the direction of the object's movement.

Computer scientist Turek: It detects the discontinuity of object movement, which sends us a signal to carefully view the picture or video, (inferred) Maybe the picture is cleared from here.

detection found that this video was passive and another car was cut off from the screen.

Computer scientist Turek: This is a game of cats and mice. The more you expose fake pictures or videos, the more pressure it puts on counterfeiters.

Two years ago, the United States released the "Artificial Intelligence and National Security" report, clearly listing artificial intelligence forgery technology as a key technology that threatens national security.

As a major country in developing artificial intelligence, China has also strengthened policy and technology supervision to deal with the possible security risks caused by new artificial intelligence technologies.

After all, when important biometric information such as sound, fingerprints, and faces begin to appear "high imitation", there will be more and more risks and hidden dangers.