When a digital human is more like you than you, you need to be wary of ethical and legal risks

AI "resurrected" Bao Xiaobai's daughter. Data map

From the digital people Tang Xiaoou and Bao Xiaobai's daughter, to the AI-generated "Coco Li" and "Qiao Renliang", since the beginning of this year, "AI resurrection" has become a hot topic. While the "AI resurrection" is gradually becoming commercialized and becoming a new business, the ethical and legal issues behind it cannot be ignored. On April 24, Southern Metropolis Daily held the 13th session of the Nandu Digital Economy Governance Forum in Beijing, with the theme of "AI Resurrection: Technological Comfort or Business Gimmick". Six experts from industry practice, technical security, philosophy of science and technology, and law gathered to discuss the "resurrection of AI" in the dispute.

At the meeting, well-known musician Bao Xiaobai participated in the meeting in the form of a video to talk about the experience of rebuilding her daughter's digital life. Chen Xiaoguang, deputy general manager of Digital Domain, the behind-the-scenes team behind "Virtual Human Teresa Teng", revealed the story behind letting a generation of singers sing on the same stage with Jay Chou on the spot, and shared the commercialization practice of virtual human technology in application scenarios such as organization and management. Han Jizhong, senior engineer of the Institute of Information Engineering, Chinese Academy of Sciences, introduced digital human generation technology and its risks. He said that if synthetic content is not clearly identified and is not authorized by individuals, it can create ethical and legal problems. Duan Weiwen, director of the Philosophy of Science and Technology Research Office of the Institute of Philosophy of the Chinese Academy of Social Sciences, mentioned the identity of digital avatars and identities, and emphasized that we should pay attention to the impact of AI avatars on emotions and ethics.

The ethical and legal torture behind the resurrection of AI

Since the beginning of this year, the use of AI technology to "resurrect" relatives has become a hot topic. Bao Xiaobai, a well-known musician, made up for her beloved regrets before she was born by reconstructing her daughter's AI data, hoping that she would not be sick in the digital world, and Tang Xiaoou, the late founder of SenseTime, "appeared" at the company's annual meeting in the form of a digital human and talked about a talk show. But at the same time, some people also used AI technology to "resurrect" late stars such as Coco Li and Qiao Renliang to harvest traffic, and were reprimanded by their families for "revealing scars".

These cases of "AI resurrection" have repeatedly aroused public discussion about this technology and the boundaries of its use. AI "resurrects" the deceased, on the one hand, it can concretize thoughts and bring psychological comfort to people, but on the other hand, it also faces many ethical and legal tortures. What do you think of this phenomenon?

At the seminar, the guests shared their views from different professional backgrounds. Wang Yuan, a partner at Beijing Gaoqin Law Firm, said that the use of personal data such as the voice and image of the deceased without the explicit consent of the deceased's family is suspected of infringement.

Bao Xiaobai also emphasized that when using AI technology to continue the "digital life" of the deceased, it is necessary to consider whether there is a legislative basis and full authorization, while avoiding harm to others. "Whether or not to be generated by AI is a matter for the family (of the deceased) to decide. It is not a business (institution), and people with skills and techniques can take it to the long and short, which is called secondary injury. ”

According to Article 994 of the Civil Code of the People's Republic of China, if the name, likeness, honor, privacy, or body of the deceased is infringed upon, his spouse, children, and parents have the right to request that the perpetrator bear civil liability in accordance with the law;

AI resurrection involves the reuse of the personal information of the deceased. Article 49 of the PIPL stipulates that the close relatives of the deceased may have the right to access, copy, correct and delete the relevant personal information of the deceased, unless otherwise arranged by the deceased before his death.

In addition, Wang Yuan added that in addition to the issues that may be involved in the "resurrection" of the deceased, such as personality rights, personal information protection and copyright, public order and good customs should also be considered. For example, for the sake of national interests and social interests, some historical heroes cannot be "resurrected" at will, let alone used for commercial purposes.

Han Jizhong, a senior engineer at the Institute of Information Engineering of the Chinese Academy of Sciences, noticed that some AI "resurrection" videos of celebrities have been able to distinguish the real from the fake. Ethical and legal issues arise when synthetic content is not clearly identified and is not authorized by individuals.

In Duan's view, there are many new questions behind the resurrection of AI that are worth discussing: "Who owns the personal data heritage?" Does everyone have the right to choose the AI to 'resurrect'? If he is a criminal, can we 'resurrect' him?" How can we respect the wishes of others who do not want to be 'resurrected'?"

He believes that with the continuous development of science and technology, some new values and ethical conflicts may arise, and that a new ethical relationship needs to be established in the future, and that legal considerations and responses need to be made.

Artistic re-creation, digital doppelganger...... The future of human-machine co-creation

On September 6, 2013, the virtual person Teresa Teng appeared at Jay Chou's Taipei concert, not only singing the classic golden song "What Do You Say", but also singing "Red Dust Inn" and "Thousands of Miles Away" with Jay Chou. In the past two years, at some satellite TV's New Year's Eve parties, the virtual person Teresa Teng has also appeared several times, performing with Wang Xinling, Zhou Shen and others.

The team behind the "resurrection" of Teresa Teng, which provides visual effects and virtual human technology, is Digital Domain, which was founded in 1993 and is headquartered in Hong Kong. Chen Xiaoguang, the company's deputy general manager in China, shared the story behind the "resurrection" of Teresa Teng by AI at the seminar that day.

"Rather than resurrection, we say it's an artistic reinvention – using current technology to make up for our regrets. In this way, Teresa Teng's artistic life is continued, and at the same time, people who like her or don't know her have a way to keep their memories forever. Chen Xiaoguang said.

According to him, before trying to "resurrect" Teresa Teng, Digital Domain contacted Teresa Teng's third brother and Mr. Deng Changfu, chairman of Teresa Teng's Cultural and Educational Foundation. After full communication and permission from the Deng family and their copyright owners, the next commercialization operation began. "AI resurrection is a resurrection within a certain boundary, not a wanton breakthrough. Chen Xiaoguang said.

If there is an insurmountable boundary for AI resurrection, then there is more imagination for living people to take the initiative to create digital clones. Not long ago, a friend from a technology company took the initiative to contact Wang Yuan, saying that as long as she provided relevant video materials, she could help her create a digital person, and in the future, she could use her digital twiner to live broadcast or participate online. In this regard, Wang Yuan showed interest - "The development of technology is unstoppable, so why not embrace it?"

Hu Yating, founder of AVAR, a metaverse digital brand company, believes that the future of AI and human co-creation has infinite room for extension and imagination. Digital life is invaluable to users. However, the premise of implementation is that it has relatively high-end multi-modal capabilities and invests a large amount of training data to achieve a better interactive experience. "The so-called digital life should be a complete life, not just a picture or a video. ”

Looking to the future, Han Jizhong believes that with the continuous implementation of applications such as AI resurrection, more people will choose to put personal data on the Internet for editing and processing, and form a new information flow. Looking back on the development of Internet technology, he believes that every change in information flow, from yellow pages, search engines to algorithmic recommendations, has profoundly changed the Internet ecology, and in the future, we may rely on AI agents as digital clones to help us browse, collect and process information, and even interact with people.

"If one day this can become a reality, the Internet will enter the era of human-machine symbiosis, then now the entire Internet ecology will also change. In this trend, it is necessary not only to promote inclusive development, but also to pay attention to its security risks. Han Jizhong said.

Duan Weiwen tries to depict the scenarios of future life: people use AI to write novels and make up their lives, from going out to look in the mirror to going out to "see" digital humans, and another self in the virtual world provides advice on dressing and matching, and people may even "live" again in the digital space reconstructed by AI......

"But does it have to be wonderful? It's possible that we'll soon get tired of this virtual life. Duan Weiwen said. But what the future holds, no one can say. He believes that in general, people's minds need to be as open as possible, to find rules in exploration and keep moving forward.

gather

Is there a risk that people will become addicted to getting along with avatars?

At the seminar, Duan Weiwen first pointed out that digital human technology is actually an exploration of the possible dimension of human existence and life, which can also be understood as digital technology redefines and constructs self and life.

Issues such as virtual privacy will also arise in the future

Duan Weiwen pointed out that on the one hand, digital technology has become the "gravitational field" of the current era in which human beings live, people are constrained by the digital world, and the use of technology has changed the way of thinking and the center of gravity of moral life.

On the other hand, in addition to the real world, people are living in the virtual digital world for more and more time, and the digital world is gradually becoming a living space for human beings to choose from, in which they redefine and reconstruct themselves and their lives. In his view, this means that human beings have always been combined with technology, a kind of "centaur" existence.

Today, digital technology has given humans a new "centaur" state, that is, to influence human life through a technological image. "This kind of technical image has always existed from ancient times to the present, and for primitive humans, those primitive murals are a very important technical image, which can give a group, a race, and a family life or survival meaning. ”

So, what do today's technology images refer to? Duan Weiwen thinks it's short video applications, etc. In his view, there has always been an inherent virtuality in human thinking, and the combination of virtuality and real technology in this thinking has changed human cognition of the world. From the mural totem of the primitive race to the current "Internet celebrity check-in place", it is the embodiment of the existence of "centaur".

Duan Weiwen also said that AI doppelgangers are AI-driven automation subjects. In other words, "AI resurrection" is a synthetic effect created on the basis of learning past data, and this effect can separate the human from the digital human. For example, humans can use one AI clone to give speeches and another to sell goods.

He stressed that the use of AI doppelgangers could cause some negative effects. The first is the emergence of the "Poseidon effect", in which people's behavior and self-perception are affected by the image of the avatar in the virtual world. "You can play different roles in different places, but whether you can guarantee that your personality is still the same, that's a question. ”

The second is the narcissistic complex. Digital technology allows humans to observe themselves, understand themselves, and even understand humans better than they can be. In such a situation, digital technology can either guide human life or make human beings fall into a narcissistic complex - in the process of digital human application, "finding oneself" or becoming a problem.

For example, he said, a Korean-American artist once deliberately used all the diary data written by his girlhood for AI training, and then talked to the digital human of Girls' Generation, "himself", to reminisce about his youth. In this way, the artist opens some knots and reconciles with the past – "then this is of course the best-case scenario, but in reality there are times when you can't reconcile and fall into a dead knot." ”

In his view, whether it is "AI resurrection" or digital human, it means that human beings can achieve a kind of virtual communication when they have emotional needs, so there will also be problems in virtual privacy and mental privacy in the future. "Who I like, who I hate, who I can talk to, who can open my heart, etc...... It doesn't have to happen in the real world, but it does give us a greater flip in our inner mental space. ”

The third is the identity of digital avatars and identities. He mentioned at the meeting that Kurzweil, who proposed that "the singularity of artificial intelligence surpassing human intelligence is approaching", once collected his father's data every day and used it to make a digital human, and in the process, he realized that "the digital human 'father' he made is more like my father than my father." This means that there are many stereotypes of human cognition, and the emotional and ethical impact of AI doppelgangers should be taken into account.

It is a breakthrough in the "double life" of human beings in the past

So from the perspective of the philosophy of technology, how do you view the "resurrection of AI", and if this technology continues to develop, will there be a risk that people will become addicted to getting along with virtual avatars?

In this regard, Duan Weiwen said that the most important thing is to distinguish between two issues. The first is to clarify the four types of life and their boundaries, including physical life, spiritual life, social life and digital life, in which physical life refers to human physiology, psychology, etc., and spiritual life refers to ideological concepts. If a person dies, his spiritual life may still exist, but his physical life and social life have disappeared, and society no longer recognizes the existence of the deceased, and theoretically he will no longer affect or interfere in the lives of others.

Secondly, the "AI resurrection" and digital human technology mean a breakthrough in the "dual life" of human beings in the past. He explained that human beings have long been in a state of "unsatisfactory", which can bring about three possibilities: the life of longing and the life of living in it, the life that has never happened and the life that is constantly unfolding, the life that is fleeting and the life that can be redeemed. In explaining the "redeemable life", he said, "The 'AI resurrection' seems to give us a hint that we can live all over again, can we have a regret pill?"

When it comes to the risk of addiction, Tuan Weiwen cites the story of Endymion in ancient Greek mythology as an answer. In the story, Endymion falls in love with Selene, the goddess of the moon, and Selene feels that humans have a short lifespan compared to the gods, so he asks Zeus, the god of the sky, to grant his lover immortality. Zeus decided that if Endymion wanted to live forever and stay young, he would have to fall into eternal sleep. In the end, Endymion chose to sleep for a long time, meeting Selene in his sleep every night.

Duan Weiwen said that this ancient Greek myth enlightened human beings that in order to pursue digital life, they may have to pay the price of "sleeping forever". "For the sake of desire, in order to permanently satisfy our desires for the goddess of the moon, we have to turn mortal vigils into immortal sleep, and this forms a dead knot - it hinders the natural course of our lives, and hinders the possibility of a better life for man himself. ”

challenge

Security risks such as data authorization must be addressed in advance

How does AI Teresa Teng cover Jay Chou's "A Thousand Miles Away"?How is Academician Yuan Longping's simulated voice generated?

At the seminar, Han Jizhong brought a live demonstration of the above scenario to the guests, and shared the AIGC technical principles and data security governance challenges behind it.

The Internet may enter the era of human-machine symbiosis

From "AI Yanzi Sun" to "AI Coco Li" and "AI Gao Yixiang", from "virtual singers release albums in one second" to the current hot business of "resurrecting the deceased", the development of the AIGC industry can be described as rapid progress. In Han Jizhong's view, although generative AI technology is still far from the real large-scale implementation, it is still very rapid in terms of the rich speed of commercial application scenarios. Han Jizhong believes that "digital humans may change the information flow of the Internet, triggering a new trend in the digital age, the Internet may enter the era of human-computer symbiosis, and more data related to personal daily life will be online", at the same time, the improvement of the service quality of large models also depends on high-quality training data, but whether enterprises can retain and use personal data in compliance and how to protect personal privacy and security is still a challenging problem.

In his keynote speech on "Digital Human Generation Technology and Its Risks", Han Jizhong summarized the current main modal technology paths of global AIGC: chatbot technology in text, voice cloning, timbre imitation and song generation technology in audio, image and video as human identity replacement, speaker generation, character customized generation, and image or video generation based on text guidance focusing on the environment.

For Jay Chou's audio input, first remove his timbre as much as possible, while retaining the lyrics, tone and other information, and then use the VITS model exclusive to Teresa Teng to add Teresa Teng's timbre to it. Taking the field of speech generation as an example, Han Jizhong introduced that the current field mainly includes mainstream technology forms such as timbre transfer and voice cloning. ”

"Voice cloning is mainly based on convolutional neural networks and pre-trained models for large-scale speech, for example, as long as Yuan Lao's previous speech is input to provide timbre, plus a piece of text, the corresponding audio can be cloned. Han Jizhong said. At the same time, he said, "The more sensitive topics in the music generation industry are mainly related to music copyright, and there are currently risk control loopholes in the training data sources in related business fields. ”

"The quest for AI to resurrect the dead comes at a price"

At the meeting, Han Jizhong said that in the case of AI resurrection of relatives, these tools can be used comprehensively, such as using customized Wensheng image technology to generate the head picture of relatives, using sound cloning technology to copy the timbre of relatives, using text generation technology to generate dialogue content, and using audio-driven video generation technology to drive the head movements of relatives.

"The quest for AI to resurrect the dead comes at a price. Han Jizhong said frankly that for digital people, it is technically possible to chat and communicate with relatives, and even write some memory dialogue scenes into the database as prompt words for large models to generate dialogues that restore the scene at that time. In terms of the corresponding technical requirements, in order to be vivid, "AI resurrection relatives" needs to collect a certain scale of high-quality facial images and voice data of the deceased at the data level. ”

He emphasized that according to the "Provisions on the Administration of Deep Synthesis of Internet Information Services", if the training data contains personal information, the relevant provisions on personal information protection shall be observed. If an AI service provider collects personal data, such as sounds or images, from a deceased person without the explicit consent of the data subject (i.e., the family or legal heirs of the deceased), it is a violation of the privacy rights of the deceased. And even if authorized, the service provider may still use the deceased's data in a way that is beyond the scope of the original authorization, such as unauthorized use for model training, "which may cause serious harm to the family of the deceased." ”

At the same time, as the effect of synthetic content becomes more and more realistic, if it is not clearly marked, it is easy to cause confusion and exacerbate the proliferation of online false information, Han Jizhong pointed out sharply. According to the "Provisions on the Administration of Deep Synthesis of Internet Information Services", where deep synthesis services provided by deep synthesis service providers are likely to cause confusion among the public, they shall be conspicuously marked in a reasonable location or area of the information content generated or edited.

In addition, Han Jizhong also mentioned that how to identify whether the digital human technology released by the company uses a certain privacy data is also in the exploration stage of technology, and the current detection technology also needs to cooperate with enterprises to provide test interfaces, and there are still certain difficulties in pure black box testing, so "it is difficult to find out whether it has used some unauthorized privacy data in the case of some enterprises not cooperating." To this end, our team is also carrying out relevant research, and if there is any progress, we will report and communicate with you."

Written by: Nandu reporter Li Ling, Fan Wenyang, Lv Hong, Mo Qianru, trainee reporter Yang Liu