Beware of new forms of fraud: "AI face-changing" | Digital literacy

Audit expert: Zheng Yuanpan, professor at Zhengzhou University of Light Industry

When it comes to AI face-changing technology (AI faceswap), everyone may have a familiar yet confused feeling. However, whether it is the prominent sign "This video is suspected of using intelligent synthesis technology, please identify it with caution" under some short videos, or the hairstyle and face-changing software that everyone sometimes uses, it shows that AI face-changing technology has been integrated into people's lives.

The human face decorated with brilliant colors

Source: pixabay

From this perspective, AI face-changing technology has gone a long way in life and entertainment, but the application of AI technology is far more than that.

As the saying goes, where there are people, there are rivers and lakes. The seemingly interesting and fun AI face-changing technology actually hides many black and gray industrial chains behind it - fraud criminal gangs that change faces to become beauties, sellers who spread fictitious photos and videos , and some individuals who change other people's faces for their own benefit, as well as ghost animal spoofs and rumors about public figures, etc., all of which let us see the other side of this magical technology.

Correct understanding is the premise of reasonable use. How is AI face-changing technology defined, how does it achieve face-changing, and in what fields can it be applied?

1 AI face-changing technology

AI face-swapping first came into the public eye in 2007, when a user named deepfakes on the Reddit community used the technology to replace the faces of adult video heroines with some well-known Hollywood actresses and published them. Eventually, his account was blocked due to the adverse impact.

However, the user later directly disclosed the source code of the technology, from which various face-changing mini-programs and apps were derived . The face-changing trend in China started with the sudden popularity of Yang Mi's version of Huang Rong on Bilibili in 2019 and the explosion of the "ZAO" software. Until now, face-changing videos are still a traffic password on Bilibili.

According to research, current digital image tampering techniques can be roughly divided into six categories: synthesis, retouching, computer generation, variation, enhancement and painting .

Woman in the Shadow of Computer Language

Source: pexels

AI face-swapping is a plus version of computer tampering with facial images with the help of deep learning algorithms. Based on face recognition technology and graphics forgery technology, it uses some deep learning algorithms to more efficiently and realistically replace the source face in the picture or video with the target face.

From a professional perspective, the deep learning algorithms of face-changing technology mainly include generative adversarial networks (GAN), convolutional neural networks (CNN), recurrent neural networks (RNN) and encoder-decoder.

Among them, CNN and RNN are feedforward neural networks and recursive neural networks with different functions respectively; GAN is a generative network that learns data distribution through the game between generative models and discriminative models; and encoder-decoder are tools for encoding images into computer data signals and restoring signals into images respectively.

In addition, the technologies used in the replacement process, in addition to the familiar face recognition technology, also include graphics forgery technology , which can use 3D models to track facial key points and affine transform the source face to the target face.

Generally speaking, AI face-changing technology is defined as deep fake technology based on facial image recognition, and deep fake technology includes identity replacement, facial reenactment, attribute editing, face generation, etc. The face-changing technology that is well known to the public is just one form of identity replacement, and is only a small part of deep fake technology.

2. Technology Application

At present, China's AI face-changing technology is mainly used in the field of film and television entertainment , such as face-changing of characters in film and television dramas, face-changing production of short videos, makeup and face-changing software, etc.

In addition, AI face-changing has also played a significant role in documentaries, real-life maps, surveillance and other areas involving privacy protection.

Virtual avatars through AI technology

Source: a software advertisement

At the same time, with the full development of deep fake technology, it also provides users with a more realistic gaming experience in the fields of games and virtual reality VR . At the same time, deep fake technology makes it possible to video call with deceased relatives or meet them again in VR, allowing us to see another beauty of this type of technology in the future.

Grandpa talks to his deceased grandmother through AI face-changing

Source: Stills from "Grandpa Bravely Enters the Metaverse"

3. Facing the Challenge

Although deep fake technology brings many new possibilities to people, the challenges it brings cannot be ignored.

For the general public, when deep fake technology cannot be used correctly, challenges such as privacy protection and property safety will follow. If criminals use deep fake technology to spread rumors, it may cause social panic. At the same time, the development of things cannot be separated from the constraints of the law. The emergence of deep fake technology is undoubtedly a challenge to the updating and formulation of relevant laws.

In general, deep fake technology, as a branch of artificial intelligence, reflects the development of artificial intelligence towards higher intelligence. How to guide the coordinated development of artificial intelligence with reality and ethics requires continuous thinking. This problem may always exist, and it may be properly solved at some point. Let's wait and see.