Following the recommendation of the algorithm, the deep synthesis technology will also usher in a separate specification. On January 28, the State Internet Information Office issued the Provisions on the Administration of In-Depth Synthesis of Internet Information Services (Draft for Solicitation of Comments) (hereinafter referred to as the "Draft for Solicitation of Comments") and solicited public comments.
The so-called deep synthesis technology refers to the use of deep learning, virtual reality as the representative of the generation of synthesis algorithms to produce text, images, audio, video, virtual scenes and other information technology, including speech synthesis, face change, virtual reality and other scenarios. In recent years, deep synthesis technology has been frequently used for pornography production and scams.
The Draft for Comments proposes that when deep synthesis service providers provide deep synthesis services, they should use a conspicuous method to identify the content of deep synthesis information. If you provide significant editing functions for biometric information such as faces and voices, you must also obtain the separate consent of the personal data subject being edited.
The content of the deep synthesis information should be prominently identified
Nandu· The AI outpost found that before the promulgation of the draft for comments, the Civil Code regulated the in-depth synthesis in the relevant articles of portrait rights: no organization or individual may infringe on the portrait rights of others by means of scandalization, defacement, or falsification by using information technology means. Since then, the Provisions on the Governance of the Network Information Content Ecology, the Provisions on the Recommendation and Management of Internet Information Service Algorithms, and the Provisions on the Administration of Online Audio and Video Information Services have also put forward relevant requirements.
Zhang Linghan, associate professor of the School of Grammar and Law of the University of Science and Technology Beijing, believes that the introduction of the draft for comments is an important measure to establish a hierarchical classification management system for algorithms. The Civil Code is stipulated from the perspective of protecting citizens' civil rights such as portrait rights and reputation rights, while the Provisions on the Administration of Recommendation of Internet Information Service Algorithms establish general obligation rules for five types of algorithms (including generative synthesis), while the Draft for Comments imposes stricter regulatory measures on high-risk deep synthesis algorithms, further implementing and refining the relevant system of algorithm classification in the Guiding Opinions on Strengthening the Comprehensive Governance of Internet Information Service Algorithms.
It is worth noting that the Provisions on the Recommendation and Administration of Internet Information Service Algorithms proposes that where it is found that an algorithm that has not made a significant identification generates synthetic information, it should be significantly identified before it can continue to be transmitted. The "Provisions on the Administration of Online Audio and Video Information Services" also require that where online audio and video information service providers and users of network audio and video information services use new technologies and new applications based on deep learning, virtual reality, etc. to produce, publish, or disseminate non-authentic audio and video information, they shall be identified in a conspicuous manner.
In fact, in the draft for comments, the deep synthesis information content identification management system is also one of the important contents, and further embodied in text, sound, image, video and other scenarios, requiring deep synthesis service providers to use effective technical measures to add their own identification, traceability, and do not affect the user's use of the logo in the depth synthesis of information content, and for the synthesis of deep synthetic content such as synthetic voices, imitation sounds, face replacement and other deep synthesis content to be marked in a significant way, to effectively prompt the synthesis of information content to the public. If it is not conspicuously identified, the transmission of this information shall be stopped immediately.
Different identification methods and standards require deep synthesis service providers to build a set of algorithmic security systems, which require a set of display identification production, identification, inspection and filtering schemes. Zhang Linghan pointed out that enterprises themselves need to customize the protocol standards and the implementation of the logo, identify the senior data, and this identification scheme will be tested and evaluated by the regulatory authorities.
In addition, the Draft for Comments follows the principle of informed consent in the handling of personal information. Article 12: Where deep-depth synthesis service providers provide conspicuous editing functions for biometric information such as faces and voices, they shall prompt deep synthesis service users to lawfully inform and obtain the separate consent of the personal data subject being edited.
That is to say, if the Provisions are implemented, no matter what purpose the "AI face-changing" video is used for, it is necessary to inform the user that its content is technically synthesized rather than true, and obtain the separate consent of the personal information subject being edited, otherwise it will constitute a violation.
When introducing the necessity of formulating the draft for comments, the Cyberspace Administration of China said that on the basis of the regulatory requirements of the rules and normative documents of relevant departments that have been issued, it is necessary to further clarify and refine the application scenarios of deep synthesis technology, clarify the information security obligations of deep synthesis service providers and users, and introduce special management regulations with systematic, targeted and operable nature.
Deep synthesis service providers should authenticate users
The Cyberspace Administration of China pointed out that while serving user needs and improving user experience, deep synthesis is also used by some lawbreakers to produce, copy, publish, and disseminate illegal information, slander and defame the reputation and honor of others, and imitate the identity of others to carry out fraud and other illegal acts.
Public information shows that there have been many fraud cases in the fields of face generation and speech synthesis.
According to the case released by the Cyber Security Bureau of the Ministry of Public Security, in December 2020, a company's financial Xiaowang received a call from a leader asking him to immediately transfer 20,000 yuan to the supplier. Because the boss's accent is very realistic, Xiao Wang believes it to be true, and the transfer is completed within 1 hour, and then found that he was deceived. In the clean net 2021 special operation, Nantong police have cracked an AI face recognition fraud case: criminal gangs use AI software to forge face recognition authentication to unbind user accounts, and then sell accounts to downstream fraudsters at low prices, thereby illegally making profits.
With the popularization of "face-changing" technology, stars have also become the hardest hit areas. In 2020, some people used image and sound synthesis technology on short video software to "incarnate" actor Jin Dong, so that many middle-aged and elderly women were addicted to it and deceived. That same year, according to foreign media reports, actress Emma Watson's 30-second face-changing video was played more than 23 million times on sites, and the face-changing videos of several female celebrities were viewed millions of times, such as Taylor Swift, Natalie Portman, Billy Elish, etc.
In view of the above chaos, the draft for comments clarifies the main responsibility of deep synthesis service providers. For example, deep synthesis service providers shall conduct real identity information authentication for deep synthesis service users in accordance with law, otherwise they must not provide information release services; employ technical or manual methods to review the input data and synthesis results of deep synthesis service users, stop transmitting unidentified content, establish and complete feature libraries for identifying illegal and undesirable deep synthesis information content, and so forth.
In terms of legal liability, the Draft for Comments proposes that where provisions are violated, the Internet information department shall give a warning, order corrections within a time limit, and suspend relevant business before making corrections; if the correction is refused or the circumstances are serious, it shall be ordered to suspend information updates and fined between 10,000 and 100,000 yuan. If damage is caused to others or if it constitutes a crime, civil liability or criminal liability will also be pursued in accordance with the law.
Wu Shenkuo, assistant dean of the Internet Development Research Institute of Beijing Normal University, doctoral supervisor, and deputy director of the Research Center of the Internet Society of China, believes that the introduction of the draft for comments has very important ecological significance for the use of deeply forged technologies to implement criminal governance such as pornography, violent terrorism, and false information. "(Draft for Comments) can improve the level of traceability, the accuracy of determining the subject of the crime, and improve the ability of all parties to respond to situations of governance and disposal of deep synthetic abuse." It can also improve the efficiency and level of handling illegal and criminal activities that have occurred. ”
Written by: Nandu reporter Li Yaning trainee reporter Hu Gengshuo