Core Tip
On June 20, 2023, the National Internet Information Office released the first batch of records of deep fusion algorithms in China, and many leading companies and AIGC pioneers successfully made the list. This is not only a key step in China’s risk governance of algorithms and artificial intelligence, but also a significant breakthrough in the compliance of the AIGC industry.
Today, the Sa team will explain to you in detail what is the record of deep fusion algorithms, and when the AIGC industry needs to be recorded, how to record and other issues.
1. What is the record of deep fusion algorithms?
- Multiple Wall Street institutions are expanding their cryptocurrency businesses, and Bitcoin has surpassed $29,000 for the first time since the end of May.
- 6 Narratives That Could Take Off in the Next Bull Market
- Understanding Global Stablecoins: An In-Depth Analysis of Their Current Status and Regulations in 2023
So far, China has two specialized legislations for AIGC tools, both of which are formulated and released by the National Internet Information Office:
1. “Internet Information Service Deep Fusion Management Regulations” (hereinafter referred to as “Deep Fusion Regulations”);
2. “Management Measures for Generative Artificial Intelligence Services (Draft for Soliciting Opinions)” (hereinafter referred to as “Management Measures Draft”)
Due to the strong guidance or inducement ability of AIGC itself, and the fast dissemination speed and wide dissemination of information network technology, AIGC content is likely to achieve viral spread in a short period of time. For example, a self-media platform recently used ChatGPT to create fake news and widely spread it to attract traffic and guide public opinion in some regions. This case has clearly shown us the risk of AIGC tools themselves as a tool for guiding public opinion or fraud. Based on this characteristic, the regulatory authorities in our country have given AIGC companies the obligation to file their own algorithms based on the long-term harmony and stability of social public order.
So first of all, we need to understand what the so-called deep fusion technology refers to. According to Article 23 of the “Deep Fusion Regulations”, deep fusion technology refers to the technology of using deep learning, virtual reality and other generative synthesis algorithms to produce network information such as text, images, audio, video, virtual scenes, including but not limited to:
(1) Technology for generating or editing text content such as passage generation, text style conversion, and question and answer dialogue;
(2) Technologies for generating or editing voice content, voice conversion, and voice attribute editing;
(3) Technologies for generating or editing non-voice content, such as music generation and scene sound editing;
(4) Technologies for generating or editing biological features in image or video content, such as face generation, face replacement, character attribute editing, face manipulation, and pose manipulation;
(5) Technologies for generating or editing non-biological features in image or video content, such as image generation, image enhancement, and image restoration;
(6) Technologies for generating or editing digital characters and virtual scenes, such as 3D reconstruction and digital simulation.
In addition to the six types of applications listed above, according to Article 6 of the draft regulations: before providing services to the public using generative artificial intelligence products, safety assessments should be submitted to the National Cyberspace Administration in accordance with the “Regulations on the Security Assessment of Internet Information Services with Public Opinion Attributes or Social Mobilization Capabilities,” and algorithm filing and change, cancellation filing procedures should be carried out in accordance with the “Regulations on the Management of Internet Information Services Algorithm Recommendation.”
Therefore, after the “Regulations” take effect, if this provision does not undergo significant changes, we should also include “deep synthesis technology” in the category of “technologies that generate text, images, sounds, videos, code and other content using algorithms, models, and rules.” according to Article 2 of the management regulations.
2. Which deep synthesis technology algorithms need to be filed?
The Sajie team reminds you that not all deep synthesis technologies require their service providers to register for filing. Under the guidance of the principle of technological neutrality, only deep synthesis technology service providers that meet specific conditions need to register their algorithms.
Specifically, the AIGC industry that carries out deep synthesis algorithm filing mainly falls within the scope of “deep synthesis service providers with public opinion attributes or social mobilization capabilities” as stipulated in Article 19 of the “Deep Synthesis Regulations.” However, in practice, the standard for defining what is “with public opinion attributes or social mobilization capabilities” is relatively vague.
According to Article 2 of the “Regulations on the Security Assessment of Internet Information Services with Public Opinion Attributes or Social Mobilization Capabilities,” internet information services that have public opinion attributes or social mobilization capabilities include the following situations:
(1) Launch forums, blogs, microblogs, chat rooms, communication groups, public accounts, short videos, live broadcasts, information sharing, mini programs and other information services or set up corresponding functions; ;
(2) Launch other Internet information services that provide public opinion expression channels or have the ability to mobilize the public to engage in specific activities.
Strictly speaking, AIGC does not belong to any of the services provided by the service providers in the first paragraph above, and most AIGCs currently do not regard themselves only as a social media platform. In our opinion, from the perspective of substantive supervision, based on the risk of widespread dissemination of AIGC itself on social media platforms, it most likely conforms to the fallback provision of “public opinion expression channels or other Internet information services that have the ability to mobilize the public to engage in specific activities”.
Therefore, from this perspective, all service providers in the basic AIGC industry may be involved in deep synthesis algorithm filing, including but not limited to text generation, image generation, video generation, audio generation, etc. The most important judgment criteria is the openness of the user group. If the service provider provides services to the general public, it is basically within the scope of the “Deep Synthesis Regulation”.
(3) How to carry out the filing of deep synthesis algorithm
First, the filing method is online filing. According to the requirements of the “Internet Information Service Security Evaluation Regulations for Information Services with Public Opinion Attributes or Mobilization Ability of Society”, service providers that meet the filing requirements shall file on the security management service platform (address; http://www.beian.gov.cn/).
In the filing process, the most important material affecting the success rate of the filing is the “Evaluation Report” submitted by the deep synthesis service provider. If the deep synthesis service provider and the technical supporter provide models, templates and other tools with two specific functions, they must provide a self-evaluation or entrust a professional organization to conduct a security evaluation and issue an “Evaluation Report”:
1. Generate or edit biological identification information such as faces and voices;
2. Generate or edit special objects, scenes and other non-biological identification information that may involve national security, national image, national interests and social public interests.
Because the two types of services mentioned above involve sensitive areas and information, the evaluation of their specific content should be based on the basic evaluation dimensions and add evaluation standards that are consistent with the “Personal Information Protection Law”, “Cybersecurity Law”, and “Data Security Law”.
Generally speaking, the evaluation standards include seven items:
(1) The establishment of security management personnel, information auditors and security management agencies;
(2) Measures to verify the real identity of users and retain registration information;
(3) Measures to retain log information such as user account, operation time, operation type, network source address and target address, network source port, client hardware characteristics, and user-posted information records;
(4) Measures to prevent and dispose of illegal and harmful information in user account and communication group names, nicknames, introductions, remarks, and identification, information publishing, forwarding, commenting, and communication group functions, and related record-keeping measures;
(5) Technical measures for personal information protection and prevention of the spread and diffusion of illegal and harmful information and the risk of loss of control of social mobilization functions;
(6) Establish a complaint and reporting system, publish information such as complaint and reporting methods, timely accept and handle relevant complaints and reports;
(7) Establish a work mechanism to provide technical and data support and assistance to regulatory and law enforcement agencies in performing their duties in accordance with the law.
Final Thoughts
The deep synthesis filing system is an important initiative and new breakthrough in China’s AI and algorithm risk governance. The Sa Jie team suggests that AI and GC industries must conduct algorithm security assessments and deep synthesis algorithm filings under the guidance of professional compliance personnel. At present, the Sa Jie team believes that this algorithm filing is an important and universal compliance obligation for the AI and GC industry. Unless AIGC service providers with small influence and extremely closed need not perform this obligation, other AIGC service providers must fulfill this obligation to prevent the breeding of administrative and even criminal risks.
Like what you're reading? Subscribe to our top stories.
We will continue to update Gambling Chain; if you have any questions or suggestions, please contact us!
