The Interim Measures for the Management of Generative Artificial Intelligence Services clearly does not apply to scenarios.

Original title: Interpretation of Interim Measures for Service Management of Generative Artificial Intelligence

Clarify the "not applicable" scenario and fully "loosen" AI development

Reporter Shendyapin

Since the beginning of this year, there has been a "group model war" in the field of artificial intelligence in China, and the industry cannot leave the "big model" in three words;

By the end of May this year, 79 large-scale models with parameters above 1 billion in China have been released.

"Talking" with people, covering songs, writing scripts and writing contracts, Generative Artificial Intelligence (AIGC) constantly refreshes human cognition and shines into real life …

Generative artificial intelligence technology refers to models and related technologies with the ability to generate text, pictures, audio, video and other content. At the same time of development, the application of generative artificial intelligence is accompanied by many chaos such as difficult to distinguish between true and false, face-changing fraud and so on.

Recently, in order to promote the healthy development and standardized application of generative artificial intelligence, safeguard national security and social public interests, and protect the legitimate rights and interests of citizens, legal persons and other organizations, the National Network Information Office and other seven departments jointly promulgated the Interim Measures for the Management of Generative Artificial Intelligence Services (hereinafter referred to as the "Measures"), which will take effect on August 15, 2023.

How to make good use of the "rudder" of the rule of law and the "paddle" of innovation to better release the innovation vitality of generative artificial intelligence and guide the development of artificial intelligence industry to be stable and far-reaching? In this regard, People’s Daily reporter interviewed relevant industry experts.

Put forward "not applicable" scenarios to leave enough room for innovation and development.

Compared with the Administrative Measures for Generative Artificial Intelligence Services (Draft for Comment) issued in April this year (hereinafter referred to as the Draft for Comment), the Measures have made adjustments in the regulatory objects.

Article 2 of the Measures clearly puts forward the "not applicable" scenario, that is, "if industry organizations, enterprises, educational and scientific research institutions, public cultural institutions and relevant professional institutions develop and apply generative artificial intelligence technology and fail to provide generative artificial intelligence services to the domestic public, the provisions of these Measures shall not apply."

"This formulation can be said to be a major breakthrough." Guo Junbo, a researcher at the National Key Laboratory of Communication Content Cognition, pointed out that this shows that the purpose of the Measures is to give consideration to the development and security of generative artificial intelligence and leave enough room for the research and development of generative artificial intelligence.

At the same time, he pointed out that compared with the Draft for Comment, the Law on the Progress of Science and Technology in People’s Republic of China (PRC) was added to the Law on the Basis of the Measures, which further highlighted that its purpose was to promote the healthy development and standardized application of generative artificial intelligence.

Xiao Sa, a senior partner of Beijing Dacheng Law Firm, also admitted that the "Measures" put forward the "not applicable" scenario, which made the target of regulation clearer and more targeted, reflecting China’s support and encouragement for the research and application of generative artificial intelligence.

In addition, the "Measures" are also relatively "soft" in terms of penalties. Article 21 of the Measures proposes that if a provider violates the provisions of these Measures, it shall be punished by the relevant competent department in accordance with the provisions of the Cyber Security Law of the People’s Republic of China and other laws and administrative regulations; If there are no provisions in laws and administrative regulations, the relevant competent departments shall give a warning and informed criticism according to their duties and order them to make corrections within a time limit; Refuses to correct or if the circumstances are serious, it shall be ordered to suspend the provision of related services.

"Article 21 adopts gradually increasing punishment measures, allowing service providers to constantly improve and correct, rather than simply’ killing with a stick’." Guo Junbo believes that the application range of generative artificial intelligence is extremely wide, and fully "loosening" will help promote the rapid technological innovation of various organizations in the field of artificial intelligence, and explore the running-in and evolution of technology in various application scenarios, thus promoting the prosperity and development of generative artificial intelligence industry.

Draw a "red line" and define a "logo" to ensure the safety of personal information.

Some time ago, the police in a certain place in China announced that some fraudsters used artificial intelligence technology to make video calls with the victims through face-changing videos, and defrauded the victims of millions of yuan in more than ten minutes. With the increasingly close combination of artificial intelligence and daily life scenes, problems such as privacy leakage and rumor fraud have become increasingly prominent. In this regard, the Measures put the protection of users’ personal information security in an important position.

Article 9 of the Measures proposes that the service provider of generative artificial intelligence shall bear the responsibility of the producer of network information content and fulfill the obligation of network information security according to law. "This clearly brings the service providers of generative artificial intelligence into the scope of regulation of the responsibility of the network platform, so that there is no gray area for violators to hide." Xiao Wei said.

At the same time, Article 11 of the Measures requires service providers not to collect unnecessary personal information, not to illegally keep input information and usage records that can identify users, and not to illegally provide users with input information and usage records. Article 14 of the Measures proposes that if a provider discovers illegal content, it should take timely measures such as stopping generation, stopping transmission and eliminating, and take measures such as model optimization training for rectification.

However, in view of the current problems that the content of generative artificial intelligence is difficult to distinguish, and even the technology is abused and misused, the Measures clarify that service providers should identify relevant generated content in accordance with the Regulations on the Management of Deep Integration of Internet Information Services.

Guo Junbo reminded, "The role of the logo is to warn and remind users that the content is generated by AI, but it cannot guarantee the authenticity of the content." Xiao Sa said, "At present, identifying artificially generated and deeply synthesized content is the lowest cost and is expected to fundamentally eliminate the above-mentioned related problems."

Expanding public training data and guiding the development of the industry to be stable and far-reaching

With the iteration of global generative artificial intelligence technology, more "smart" big models will surely penetrate into more industrial fields and life scenes. At present, how to further seize the opportunity of industrial transformation and promote the stable and far-reaching development of the industry has attracted much attention from all walks of life.

The relevant person in charge of the National Network Information Office said that the Measures clearly encourage the innovative application of generative artificial intelligence technology in various industries and fields, generate positive and healthy high-quality content, explore and optimize application scenarios, and build an application ecosystem.

"At present, China’s generative artificial intelligence technology is in the catch-up stage, and non-development is the biggest insecurity." Guo Junbo pointed out that the "Measures" proposed "inclusive and prudent supervision of generative artificial intelligence services", which will accelerate the landing of domestic generative artificial intelligence technology. Relying on the huge number of users in China, domestic generative artificial intelligence is expected to quickly accumulate real data, drive the iteration and optimization of technology, and promote the innovation and development of artificial intelligence.

It is worth noting that in promoting industrial development, the Measures specifically proposed that "independent innovation of basic technologies such as algorithms, frameworks, chips and supporting software platforms of generative artificial intelligence should be encouraged, international exchanges and cooperation should be carried out on an equal and mutually beneficial basis, and international rules related to generative artificial intelligence should be formulated." At the same time, the "Measures" also proposed to "promote the orderly and open classification of public data and expand high-quality public training data resources."

"This shows that the country is optimistic about the development of generative artificial intelligence." In Xiao Sa’s view, data barriers and huge public data idleness are two key issues that hinder the development of generative artificial intelligence. After the implementation of the Measures, it is expected to greatly expand the artificial intelligence training database and greatly promote the depth and breadth of artificial intelligence application in China.

Guo Junbo admits that in the big model era, training a model with hundreds of billions of parameters requires huge computing resources. In addition, the number of high-quality data sets opened in Chinese has lagged behind that in English for a long time, which limits the further improvement of artificial intelligence technology based on Chinese. The promulgation of the "Measures" will help promote the open sharing, quantity improvement and quality improvement of the underlying basic resources of artificial intelligence such as computing power and data.

"It is expected that in the next step, the regulatory authorities will also regulate the generative artificial intelligence in certain fields, such as medical care and finance, according to industry regulations." Guo Junbo said. (Intern Han Xinning also contributed to this article)