Shulou(Shulou.com)06/02 Report--

In the tide of the world attacking the "evil" of AI, AI fraud may already be the number one reason. In this article, we will talk about why AI needs a "name correction" from "deep forgery" to "deep synthesis".

Since the birth of Deepfake (Deep forgery) in 2017, people have been amazed that AI is simply talented in counterfeiting. Since then, this fact has been further confirmed by the rapid development of "AI generated content" technology, especially the GAN algorithm. Not only AI face change, but also AI automatically generates text, voice, image, video and other digital content.

In addition to the flood of erotic videos brought about by AI face changes, people are further worried that AI content generation technology will bring new challenges in terms of privacy violations, threats to information security, manipulation of political elections, and so on.

People tend to assume that if AI-generated content is allowed to spread across the Internet, it will further erode the boundaries of authenticity in the Internet world.

(Zuckerberg, who was faked by AI, "satirizes" his Facebook platform.)

Where is the truth after Deepfake?

If it is difficult for ordinary people to tell what is true and what is false, then the truth and trust that make up the cornerstone of society will collapse, but we do not seem ready to live in a "distrustful society".

In the critique of practical reason, the German philosopher Kant demonstrates the law of "why people can't lie", which reveals the paradox and absurdity of "untrusted society".

If "everyone can lie" is a general law of society, then everyone will no longer trust what another person says, so that the speaker's lie will not succeed. Lying and no one believes, it is caught in a self-contradictory situation, on the contrary, "no one can lie" should be the general rule of normal society.

That is to say, only in a trusting society where "everyone should be honest" can liars make a profit by successfully deceiving others, and will go bankrupt because the lies are exposed. In the "untrusted society", it is difficult to judge whether the information is true or false, so we can only acquiesce that everything is "false" so that we will not be deceived. But the corresponding price is that there is no trust, cooperation is difficult to achieve, communication is no longer possible, and eventually society will fall apart.

Of course, this is only the most extreme interpretation of the theory. The real world will always form a huge grey space under the theoretical world, the cornerstone of human nature will remain unchanged, the evolution of technology will be unstoppable, and the weakness of human nature will always be exposed in the tension of the tear between the two. It seems that there is no better way for each generation of new humans to continue to learn and evolve to adapt to the new challenges brought about by the acceleration of technology.

Going back to the "AI generated content" technology represented by "Deepfake", it will neither drag our society into an abyss of "no trust", nor will it make our human nature better or worse. In the coming post-truth world of "compatibility of reality and reality, no distinction between true and false", it will only make us more complex and anti-vulnerable to adapt to this change.

So. This slightly derogatory Deepfake (deep forgery) technical term needs to be remolded into a technology-neutral word-Deep Synthesis (deep synthesis).

Rectifying the name of "Deep Synthesis": the Technical neutrality of AI

Every breakthrough in science and technology may bring unexpected "by-products". Just as after Einstein discovered the mass-energy equation, he could not stop the emergence and use of the proton bomb, even if he was reluctant to do so. Just after the "evil" of Deepfake was released on the US news site Reddit, Yann LeCun, the leader of AI, also reflected on Twitter:

"to be honest, if we had known that convolution neural network (CNN) would give birth to Deepfake, would we have published it?"

Then LeCun answered the question himself. LeCun said that even if we didn't publish it first, CNN would have been invented by someone else or organization. After it was announced in 2002, people didn't know how to use it.

In other words, the value of CNN can only be tapped by the continuous exploration of technicians. Now CNN is being developed into a variety of applications, which not only have many positive effects on the world, such as medical diagnosis, autopilot, content filtering and security monitoring, but also may cause some negative effects, such as invasion of privacy, fraud, prejudice, discrimination and so on.

To put it simply, AI is innocent, and the problem still lies with humans who use AI technology.

For example, the flooding of AI face-changing video caused by Deepfake technology, it is almost a "inevitable" process for AI technology to be applied to the sex industry. On the one hand, the modern sex industry has always been a pioneer in the application of the latest science and technology, on the other hand, the development of AI in image content generation technology has just ushered in a breakthrough critical point. In the final step, only this user named "Deepfakes" will have the last "brainstorm".

In fact, Deepfake contributes to the popularity of "AI content generation" technology, but it also brings indelible stigmatization. In view of the fact that the development of "AI content generation" technology has long gone beyond the scope of AI face-changing, the technology business community is trying to use "deep synthesis" to correct the name of this technology.

First of all, the term Deepfake (deep forgery) is obviously overgeneralized. it is only an early representative of "AI face-changing" technology and is not enough to include all "AI-generated content" technologies. Using Deep Synthesis (Deep Synthesis) can better refer to all AI generation algorithms and synthesis techniques that cover the automatic generation of image, video, voice, text, music and other content.

Second, Deepfake has not been widely recognized by the technology community, but has just been called catchy by the media. Moreover, the "abdominal black" constitution of Deepfake will have a direct negative impact on the application and promotion of AI technology.

"Deep synthesis", a more neutral technical name, will replace Deepfake to exercise the responsibility of AI content development. So, how can "deep synthesis" support this important task?

The strength of "Deep Synthesis": technological acceleration and Commercial Landing

In fact, the "deep synthesis" technology is realized with the help of a deep learning algorithm model that can learn independently. The two main technologies used are the artificial neural network of "automatic encoder" and the artificial neural network of "generating countermeasure network" (GAN).

The former is used for the synthesis of training data, while the latter is composed of a generator and a discriminator, one is used to generate new data and the other is used to identify it. After countless confrontations between the two, the composite data is finally generated, including the face-changing video created by Deepfake.

(publication of GAN-related papers)

From 2014, GAN proposed to today, has experienced from CGAN, BigGAN, StyleGAN and other versions of the update, in which the annual related research papers are also accelerating growth, which shows that the academic attention and development prospects of GAN algorithm.

(high-definition pictures generated by BigGAN with various categories)

Correspondingly, the quality of image generation is improved by leaps and bounds, in which we can not only achieve face synthesis, but also achieve image superposition and fusion, or directly generate brand-new high-definition pictures, so that it is difficult for human eyes to distinguish between true and false.

For example, a GAN-based AI artist released by MIT and IBM Watson last year can learn the painting style of Renaissance painters and directly turn pictures of modern humans into Renaissance paintings.

The technical advantage is that the GAN neural network reconstructs the picture according to the skills it has learned, that is, it draws new pictures, rather than using style transfer to change the color of the original picture.

In fact, deep synthesis technology can go further. In addition to single image and audio synthesis, multi-dimensional content synthesis has become a trend, which can combine speech recognition, face recognition and lip search to synthesize face voice. this allows a person to naturally and fluently say what he has never said before.

In addition, in addition to face synthesis, whole-body synthesis and 3D synthesis of virtual human technology have also become the focus of the current application. During the past two sessions, Sogou and Xinhua News Agency launched the world's first 3D AI synthetic anchor, which can already drive facial expressions and lips in real time, body movements, and hyperrealistic detail presentation to compare with real-life dynamic effects.

In the commercialization of "deep synthesis" technology, many industries and enterprises have seen its application scenarios and broad market. At present, "deep synthesis" has begun to play a role in many industries, such as film and television entertainment, social communications and so on.

For example, in film and TV drama production, the most direct help is to improve the efficiency of audio and video production and expand the creative imagination. In some special cases, we can also use synthesis technology to synthesize the voice of the actors who have lost their voices, "digital resurrection" for the deceased actors, and even directly create virtual digital people to produce film and television dramas.

In terms of entertainment application experience, the most basic facial special effects applications, face-changing videos and motion pictures have appeared in our lives many times; virtual idols, virtual anchors, and virtual customer service have become more realistic and credible with the maturity of deep synthesis technology.

In terms of social communications, instead of worrying that deep synthesis technology will expose personal privacy, deep synthesis technology can help us build our own "digital avatars" in social networks, just like the virtual images created by everyone in the "number one player". Become your own pass in the online world.

In addition, in the fields of e-commerce marketing, educational art, medical research and so on, the simulation data and virtual content brought by deep synthesis technology can bring new application scenarios for these industries or directly promote the technological progress in this field.

Obviously, these positive values of deep synthesis technology are giving it a more confident voice and development prospects. But this AI technology, which has been questioned and feared by human beings as soon as it appears, is still worthy of our serious attention to its application boundaries and rules.

The governance of "deep synthesis": how to lock the evil dragon of "false content"

Just as all acquisition must pay a price, if we want to enjoy all the convenience and spiritual enjoyment that deep synthesis technology brings to us, we must bear the cost of fully virtualizing the digital world. The social risk of "false content" brought by deep synthesis technology will exist for a long time.

First of all, the deep synthesis of open source technology and software greatly reduces the threshold for ordinary people to forge and manipulate audio and video; second, these false audio and video content is enough to deceive most people who "don't know the truth"; finally, these information with obvious color, alarmism or invasion of privacy is attractive enough to spread as long as it comes out from the source.

Except for a few professionals who can tell the true from the false, most people find it difficult to distinguish and resist the temptation of these false information. The risk of technology abuse in depth synthesis needs to be restricted by law, technology, industry, people and so on.

First, the legal level. The use, marking, scope of use and penalties for abuse of technology of AI deep synthesis content should be studied in depth, and # corresponding regulations should be issued to provide a basis for the legal use of deep synthesis. Second, the technical level. Content identification technology and traceability technology which evolve synchronously with deep synthesis technology should also be paid attention to. For the effective identification and marking of the synthetic content, the synthetic content can be identified from the source, so as to prevent the spread of negative false content. Third, the industry level. Deep synthesis technology is inseparable from industry self-discipline, synthetic content technology providers and platforms must promise to make tags on the synthetic content, or provide effective detection and tagging tools to ensure that the synthetic content is clearly identified. Finally, at the level of the public. Compared with the worries of authorities or mainstream elites about the proliferation of synthetic content, the general public may be the main supporters of this wave of "virtualization", or even the promoters of false information.

Today, when we are about to usher in the digital world, the cultivation of qualified "digital literacy" should become a compulsory course for citizens since childhood. However, what to teach and how to teach this course still need to be explored slowly in the development of deep synthesis technology.

Just as no technology is available until we are ready, so is AI technology. From the very beginning, we defined the starting point of AI technology as learning and imitating human ability as much as possible, so that it can eventually replace human driving those heavy, repetitive and even extremely difficult tasks.

The deep synthesis technology is not the realization process of this goal. Since we have chosen to awaken the giant dragon AI, we can no longer worry about the fact that AI is becoming more and more human.

Finally, if we look at our human species in turn, on the one hand, we have the ultimate wisdom to explore the laws of cause and effect in the world, and we always explore that "truth"; on the other hand, we invent all kinds of tools with great enthusiasm. to take on all kinds of human work.

These two capabilities also directly contribute to our industrial world today, as well as the digital virtual world we are going to enter in the future.

Optimistically, not only do we not have to worry too much about the advent of the "post-truth era", but even we will soon adapt to this beautiful new world of complete "virtualization".

For most people, the pursuit of truth is far less attractive than the pursuit of comfort.

Http://www.woshipm.com/ai/3983358.html

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.