In recent years, with the advancement of artificial intelligence technology, some criminals have begun to use AI technology to fuse other people's faces and voices to create very realistic synthetic images to carry out new online frauds. Such scams often cause serious consequences to the victims in a short period of time. loss.
How should we see clearly this type of AI “deep fake”?
Is there any good way to identify "AI face-changing"?
Let’s first take a look at a recent “AI face-changing” fraud case that occurred in Xi’an, Shaanxi.
Ms. Zhang, a financial officer in Xi'an, Shaanxi Province, had a video call with her boss, who asked her to transfer 1.86 million yuan to a designated account.
Victim Ms. Zhang
: The boss asked me to transfer this money quickly. This money person was very anxious because his voice and video image looked the same as others, so I was even more convinced that this money was from him, and then I directly transferred it to him. Transferred this amount.
After the transfer, Ms. Zhang sent the electronic voucher to the company's internal financial group in accordance with regulations. However, unexpectedly, after the boss in the group saw the information, she asked her about the source of the funds?
Victim Ms. Zhang
: Then we called the boss to verify it. The boss said that he had not sent me the video, and he had not mentioned the transfer.
After realizing that she had been deceived, Ms. Zhang quickly called the police for help. The police immediately contacted the anti-fraud center and contacted the relevant banks to stop the payment urgently. In the end, most of the defrauded funds of 1.56 million yuan were saved.
AI "multiple face-changing" fraud case defrauded 200 million Hong Kong dollars
More than one million in cash is a huge amount of money for most people. Fortunately, the police handled it in a timely manner and recovered most of the losses. Let's look at another case, this case's "face-changing" technology It is more difficult and the amount involved is larger.
Recently, the Hong Kong police also disclosed an AI "multiple face-changing" fraud case, involving a total amount of up to HK$200 million.
In this case, employees of the Hong Kong branch of a multinational company were invited to participate in a multi-person video conference initiated by the chief financial officer of the headquarters.
According to the requirements, he made multiple transfers and transferred 200 million Hong Kong dollars to five local bank accounts. He then inquired with the headquarters before he learned that he had been deceived.
Police investigation revealed that in this case, only one of the victims in the so-called video conference was a "real person", and the other so-called participants were all fraudsters who had their faces changed by AI.
After reading these two cases, you must be a little curious, what is the technical principle behind AI face-changing?
At a technical level, how does it achieve accurate recognition and replacement of faces and create realistic effects?
Let’s listen to what the experts have to say.
How to technically achieve accurate face recognition and replacement
Xue Zhihui, an expert from the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: The AI face-changing process mainly includes several key steps such as face recognition and tracking, facial feature extraction, face transformation and fusion, background environment rendering, and image and audio synthesis.
The core behind it includes three parts. First, use deep learning algorithms to accurately identify face images in videos and extract key facial features such as eyes, nose, mouth, etc.
Secondly, these features are matched, replaced and fused with the target face image.
Finally, by rendering the background environment and adding synthesized sounds, a highly realistic fake face-swapping video is generated.
Quick AI "face-changing" can be completed with just one photo
In order to understand how realistic AI face-changing can be, the reporter cooperated with professional technicians and experienced the AI face-changing technology in depth.
The technician first took a photo of the reporter's face with a mobile phone. After importing it into the AI artificial intelligence software, the reporter was surprised that although the person in front of the computer camera was a technician, the output was indeed the reporter's photo. It could almost be said that "One-click face change" does not require complex environment and decompression operations.
What’s even more surprising is that as the technician’s facial expression changes, the reporter’s face in the photo also changes accordingly.
Reporter: Why does my photo move when the technician's face moves?
Xue Zhihui, an expert from the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: First, through video collection, the face in the picture can be tracked and positioned. After locating it, in the second step, he can make a facial feature of the face. The collection and extraction of points mainly include obvious facial features related to the mouth, nose and eyes.
After collecting it, the third step is to transform, fuse and reshape the original photo.
Reporter: What else can be achieved through this photo?
Xue Huihui, an expert from the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: Currently, through this photo, technicians have made this photo move and come alive, and if this photo is further saved, it can be stored Lots of photos then.
This photo can be combined into a short video and released later.
Artificial intelligence face detection technology is mainly implemented through deep learning algorithms, which can identify facial features and accurately analyze them.
One person's facial expressions can be extracted from a photo or video and matched to another person's facial features.
Experts told reporters that if you want to use artificial intelligence face-changing technology during real-time video calls, one photo is not enough, and you need to collect nearly a thousand photos from different angles.
Xue Zhihui, an expert from the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: If you want to communicate point-to-point in real time, you need to collect more photos and complete the training of the deep learning algorithm model. After training this model, we will feed it to our video. Once inside, you can make real-time changes and conversions.
In this case, real-time face changing can be achieved.
Voice communication is also possible. We need to collect some voices from our target group in advance, and then train the model to restore the voices of the target group.
AI-generated video with extremely high degree of simulation is difficult and requires a lot of investment
According to experts, implementing point-to-point video calls for the purpose of fraud requires AI artificial intelligence to generate videos with extremely high degree of simulation.
It is not easy to achieve the effect of fakes and real ones for fraud.
Fang Yu, a member of the Digital Economy and Security Working Group of the Security Professional Committee of the China Computer Society: For fraud, it is actually mainly through point-to-point video calls. At this time, if face-changing technology and voice synthesis are used to conduct real-time face-changing fraud, you want to complete these Technical operations require strong technical support.
Xue Huihui, an expert from the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: There needs to be a lot of financial investment behind it, including the collection of pictures, including professional algorithm personnel, etc. It requires a long cycle, including some computing power algorithms. .
Investment in all aspects requires continuous iteration and operations over a long period of time to achieve a very realistic effect and achieve the actual effect of fraud.
At present, AI "face-changing" is more used in entertainment video recording
In addition to some criminals trying to use AI technology to commit fraud, in fact, in recent years, AI technology has been increasingly used in the secondary creation of short videos, which are of an entertainment nature. More and more celebrity AI face-changing videos appear on the Internet. , many netizens lamented that in such videos, "the mouth shapes and gestures are all right, so natural, I almost thought they were real."
For example, this software can record and generate a video of the reporter transforming into a racing driver after taking a photo of the reporter's face.
Fang Yu, a member of the Digital Economy and Security Working Group of the Security Professional Committee of the Computer Federation of China: What we often see in AI technology currently is face-changing in short videos, by doing some specific actions, dancing, etc.
These videos actually look a little unnatural, and they are purely entertainment face-swapping.
The reporter found that there are dozens of face-changing software in the mobile application mall, all of which can achieve the purpose of face-changing.
Xue Zhihui, expert of the Artificial Intelligence Security Governance Committee of the China Cyberspace Security Association
: From the perspective of entertaining the public, there are many software and tools on the market that can achieve the effect of AI face-changing, but the simulation level is only six The public can see it directly from the seven-point appearance.
But if you want to generate a fraud video to speak, you need to generate our highly simulated point-to-point video.
AI technology "changing face and voice" may have legal risks
However, AI technology is also a "double-edged sword". Even if AI is used for face-changing or voice-changing for entertainment, there are still legal risks.
Legal experts said that using AI technology to change the face and voice of others or even translate it into other languages and publish videos may be suspected of infringement. There are three main aspects:
First, suspected copyright infringement, such as cross talk, sketches, etc., are "works" protected by the Copyright Law of the People's Republic of China.
For example, netizens use AI software to "translate" crosstalk, sketches, etc. into other languages, which must be authorized by the copyright owner, otherwise there will be infringement issues.
The second is suspected infringement of portrait rights. According to the Civil Code of the People's Republic of China, no organization or individual may infringe on the portrait rights of others by vilifying, defacing, or using information technology means to forge.
Without the consent of the portrait right holder, the portrait of the portrait right holder shall not be produced, used, or disclosed, unless otherwise provided by law.
The third is suspected of infringing the right of voice. According to the provisions of the Civil Code of the People's Republic of China, the protection of natural persons' voices shall refer to the relevant provisions on the protection of portrait rights.
In other words, you need to obtain the consent of the voice owner before you can use the voice of others.
Learn a few tricks to easily identify AI "face-changing and voice-changing"
The emergence of AI face-changing technology has caused hearing to be false and seeing to be false.
So how can we prevent it?
Experts said that in fact, no matter how realistic AI artificial face replacement is, there are still some ways to identify the authenticity of the video.
Fang Yu, a member of the Digital Economy and Security Working Group of the Security Professional Committee of the China Computer Federation
: In fact, judging from the real-time deepfakes I have seen so far, there are actually some ways to conduct some verifications.
For example, we can ask the other party to identify and fake the video in real time by waving in front of their face during the video conversation, because it needs to generate and process the video in real time and use AI to replace it. Face.
Then during the process of waving, it will cause interference with the facial data. The final effect is what we see. During the process of waving, the face he forged will have a certain amount of jitter or some flash. Or some unusual circumstances.
The second is that in peer-to-peer communication, you can ask questions that only the other party knows to verify the authenticity of the other party.
Raise awareness of prevention to avoid leakage of personal biometric information
Experts said that in addition to some tips for identifying AI face-changing scams, each of us should increase our awareness of prevention, take relevant preventive measures in our daily lives, and develop good online habits.
First of all, we should protect daily information security and strengthen the security protection of biometric data such as faces, voiceprints, and fingerprints. In addition, we should do a good job in security management of the software and hardware of personal mobile phones, computers, and other terminal devices.
Second, do not log into websites from unknown sources to avoid being invaded by viruses.
Third, authorization management must be done for applications that may collect sounds, images, or even videos and positioning.
Not giving others the opportunity to collect your information can also keep AI face-changing scams away from you to a certain extent.
The development of AI technology requires multi-level regulatory regulations
In addition to improving self-prevention awareness, how to strengthen supervision of AI technology has also become an issue of concern to more and more people.
AI technology itself is not a problem, the key is how do we use it?
How to form effective supervision?
According to experts, the development of AI technology requires multi-level supervision and regulation.
First, at the source, it is necessary to further strengthen the protection of citizens’ personal information, especially the technical and judicial protection of biometric and other private information.
Second, management can be strengthened at the technical level.
For example, video dissemination websites or social software can be identified using professional identification software, and the AI-generated videos can be stamped with an indelible "AI-generated" watermark.
At present, this digital watermark detection technology needs to be further popularized.
Third, at the legal system level, it is necessary to further improve laws and regulations related to artificial intelligence and other fields.
On August 15, 2023, my country officially implemented the "Interim Measures for the Management of Generative Artificial Intelligence Services".
The "Measures" "draw red lines" from many aspects, aiming to promote the healthy development and standardized application of generative artificial intelligence.