CCTV News: The "Measures for the Generation of Synthetic Content in Artificial Intelligence" (hereinafter referred to as the "Measures for the Generation of Synthetic Content") will be officially implemented from September 1 this year. The "Measures for the Generation of AI-generated text, pictures, videos and other contents must be forced to add explicit and implicit identification, which also means that all content generated by AI must "disclose the identity" and can no longer "disguise the real with falsehood". So, how should we add the content identifier generated by AI? Can publishers design logos according to their preferences? Regarding these issues that netizens are more concerned about, let’s hear what the experts say.

AI generated content should be identified according to the corresponding specifications

The "Identification Measures" clearly states that the synthetic content identification generated by artificial intelligence includes explicit identification and implicit identification. Explicit identification refers to an identification added in the generated synthetic content or interactive scene interface, presented in text, sound, graphics, etc. and can be clearly perceived by users. The "Logination Measures" clarify the standards for explicit identification in six categories of situations generated and synthesized by artificial intelligence.

Zhao Jingwu, associate professor of the School of Law of Beijing University of Aeronautics and Astronautics: In the above six categories of situations, the content, location, and length of explicitly identified are standardized. For example, if I want to use text identification, the text must also contain keyword information such as "artificial intelligence" or "AI", "generate", or "synthesis", and the font requirements must be clearly visible. In other words, the AI ​​content producer and publisher's identification of the work should at least meet the basic condition of "significant and obvious".

In addition to explicit identification, there is also an implicit identification, which refers to an identification added in the data of the generated synthetic content file and is not easily visible to users. The "Identification Measures" clarify the elements of implicit identification, and also proposes that Internet application distribution platforms need to review whether the application complies with the identification.

Zhao Jingwu, associate professor at the School of Law of Beijing University of Aeronautics and Astronautics: Implicit identification is not complicated. This type of identification mainly includes generating synthetic content attribute information, service provider name or encoding, content number and other production element information. You only need to add the above information to the metadata of the file that generates the synthetic content. Common forms such as digital watermarks belong to the implicit logos mentioned in this Measures.

Experts: AI-generated content must be "identified"

Experts introduce that when individual users use AI to create text, if there is no infringement on the interests of individuals, enterprises, society and other subjects, users do not need to make special marks. However, if the text content generated by AI involves imitating someone's discourse, imitating the media for news releases, or imitating government departments to issue relevant announcements, it must be explicitly marked.

The existence of the logo will help reduce the audit operation cost

Experts introduced that under the requirements of the "Identification Measures", the existence of the logo will also intercept the behavior of large-scale transfer of artificial intelligence to directly generate content, giving the content platform clearer compliance requirements and management procedures, which is conducive to reducing the overall audit operation cost.

Zhang Jiyu, Executive Director of the Institute of Future Rule of Law at Renmin University of China: It will be more conducive to the authenticity of the entire cyberspace and digital space content, which also enables our self-media to use artificial intelligence technology to make their videos more outstanding, but don’t use them to attract attention by creating some false news. Therefore, it is to guide the majority of self-media to use artificial intelligence technology more actively. The implementation of the "Login Method" is conducive to fair competition and effective management. Experts introduced that the implementation of the "Login Method" is conducive to fair competition and effective management. It is understood that as the first combination of "methods + strong standards" in the field of artificial intelligence in China, the mandatory national standard "Cybersecurity Technology Artificial Intelligence Generation Synthetic Content Identification Method" will be implemented simultaneously with the "Artificial Intelligence Generation Synthetic Content Identification Measures" on September 1.