Mr. Richard WANG, Partner, Patent Attorney, Panawell & Partners

 

In May 2020, OpenAI announced the beta version of the GPT-3 model, and then the official release of ChatGPT-3.5 in June 2022, thereafter, generative artificial intelligence (AI) quickly garnered widespread attention, turning a quite specialized technological domain of AI field into a topic of public discussion. On July 13, 2023, the Cyberspace Administration of China (CAC) released, jointly with six other Chinese government agencies, the Interim Measures for the Administration of Generative Artificial Intelligence Services (the Interim Measures).

 

This article will be generally interpreting and analyzing the Interim Measures, with an aim to explore its background, main contents, and potential impacts on the AI and Internet industries, and make suggestions for the reference of practitioners in the generative AI field.

 

1. Background

 

Generative Artificial Intelligence (also known as Generative AI or GAI technology), like large-scale pre-trained models based on deep learning (e.g. GPT), has made significant progress in recent years. These models have been developed to generate text, images, and audio, offering limitless potential in various applications ranging from text summarization to creative writing, and from artistic creation to automated customer services. However, the widespread application of this technology has brought about a host of legal and ethical issues, such as ethical dilemmas in content creation, intellectual property disputes, misinformation circulation, and privacy issues, thus causing widespread concerns. It takes merely three months from the CAC release of the Measures for the Administration of Generative Artificial Intelligence Services (Draft for comments) on April 11, 2023, to the official implementation on July 13, 2023, which fully shows the importance the Chinese government has attached to the generative AI technology. It is rare also that up to seven State Council agencies of China collectively issue legal regulations on a very specific emerging technological application of AI technology, demonstrating the Chinese government’s stance towards the industry’s development amidst this wave of AI evolution, while establishing the basic norms for generative AI services to follow the requirements set forth in, among others, the Cybersecurity Law, Data Security Law, Personal Information Protection Law, and Science and Technology Progress Law of China.

 

The Interim Measures, a crucial legislation concerning generative AI in China, constitute, together with the Provisions on the Administration of Algorithm Recommendations of Internet Information Service (the Algorithm Recommendation Provisions) and the Administrative Provisions on the Administration of Deep Synthesis of Internet Information Service (the Deep Synthesis Provisions), the key regulatory provisions in the AI and algorithm fields.

 

2. Main Contents

 

1) Scope of Application

 

The Interim Measures cover various aspects of generative AI services, including, but not limited to, natural language generation, image generation, text summarization, and cross-border service scenarios, aiming to include multiple application domains of generative AI service provisions from news media to advertising, from healthcare to creative industries, and other possible fields in the future.

 

The Interim Measures specify that services providing generated text, images, audio, or video to the public within mainland China using generative AI technology are subject to these regulations. Unlike its draft version, the Measures clearly exclude generative AI technology research and applications which do not provide services to the public in China. Therefore, for an AI service provider, if its generative AI technology is used only for internal research or applications, its compliance burdens are somewhat eased in the model training phase and internal operations, provided that cybersecurity, data security, and personal information protection laws and regulations are followed, which shows the Chinese government intention to encourage generative AI innovation and development.

 

2) Service Providers’ Obligations

 

The lifecycle of generative AI is broadly divided into three stages: model training, application, and optimization. Those involved may include data collectors, data providers, model developers, service providers, and users. In practice, it is possible for data collector, data provider, model developer, and service provider to be one, or different entities cooperating with one another.

 

Probably for the legislative convenience, however, the Interim Measures only specify two types of entities, namely the generative AI service providers (providers) and users of generative AI services (users), with the scope of providers clarified, i.e., organizations or individuals providing generative AI services using generative AI technology (including those providing generative AI services through programmable interfaces, i.e. API). It's noteworthy that unlike the Deep Synthesis Provisions which distinguish between deep synthesis technology supporters, service providers, and users, the Interim Measures mainly address the characteristics of generative AI, centering on the providers’ service provisions, and narrowing down their responsibilities to a certain extent, so as to better support the development of generative AI services. The rights and obligations existing among data providers, model developers, and service providers are likely to be regulated under the future regulations or their commercial contracts.

 

Furthermore, as the Interim Measures are sector or industrial regulations, some of the provisions have already been set forth in the higher-level laws or relevant regulations. For example, obligations regarding compliance with laws, administrative regulations, respecting social morality and ethics, collection, use, and protection of users personal data, data security, and protection of others' intellectual property rights, have all been laid out in, among others, the Cybersecurity Law, Algorithm Provisions, and Deep Synthesis Provisions.

 

The Interim Measures clearly state that in content management, providers, as online information content producers, should be responsible for network information security; when personal information is involved, providers shall also be responsible as personal information processors, obliged to protect personal information. In respect of training data, providers have the obligation to ensure legality, meet quality requirements, and do content labeling and algorithm debiasing to enhance authenticity, accuracy, objectivity, and training data diversity. In the rights and obligations in relation to users, providers are obliged to formulate service agreements, construct reasonable use and addiction prevention mechanisms, ensure service stability, and establish complaints and reports handling mechanisms to ensure users’ effective use of generative AI services.

 

It is worth noting that, under the Interim Measures, if the generative AI service provision from outside to mainland China does not comply with legal and regulatory requirements, the CAC would notify the relevant authorities to take technological, and any other necessary, measures to deal with the matter. Therefore, the Interim Measures do not prohibit generative AI service provision from overseas to China; however, in cases where such services violate Chinese laws and regulations, the Chinese government could take necessary measures to block the service provision, which means that the Interim Measures have a certain degree of regulatory effects on foreign service providers.

 

3) Punitive Measures

 

The Interim Measures lay out punitive measures for unlawful activities to ensure effective enforcement of the regulations. The penalties, primarily in line with the laws, such as the Cybersecurity Law, Data Security Law, Personal Information Protection Law, Science and Technology Progress Law of the People’s Republic of China, include warnings or fines, suspension or revocation of relevant licenses, or criminal liabilities. However, it is important to note that for actions not specified under the above laws or administrative regulations, the relevant supervisory agencies still have the authority, based on their duties, to issue warnings and publicized criticism, or order corrections within a period of time for violations of the provisions of the Interim Measures. If a correction is refused or the circumstances are severe, the related service provision could be ordered to be suspended. For this special provision might have been made with account taken of the fact of the fast pace of technological advancement versus the relatively slower legislation procedure, giving law enforcement officials with relatively larger flexibility in enforcement for them to respond emerging violations swiftly. However, this kind of penalty lacks explicit legal basis, and the interpretative authority of the law enforcement officials might be large or arbitrary, which might cause AI practitioners to encounter uncertainties with regard to their action or operation expectation. However, this relatively flexible and broad stipulation is not unique to the Interim Measures, as similar provisions are found in the Algorithm Provisions. For this matter, the AI practitioners should pay attention to them intensely.

 

3. Possible Impacts on AI Technology Development

 

The Interim Measures prescribe a set of obligations for participants in generative AI technology, encompassing not only model developers and data providers, but also service providers utilizing the models and the end users or consumers. Besides requiring providers to take measures, such as warning, function restricting, suspending, or terminating services when users are found engaging in unlawful activities using generative AI services. Providers are also required to report to the relevant supervisory agencies. Users, upon finding that the generative AI services do not comply with laws or administrative regulations, have rights to complain or report to the relevant supervisory agencies about the upstream model developers or service providers.

 

This mutual supervisory and restrictive relationship will compel generative AI technology providers to pay more attention to ethical and legal compliance to ensure the morality and legality of the technology. However, it may also cause generative AI service providers to restrict users’ behavior when providing services to the public in order to avoid punishment on account of user abuse or misuse of their technology which leads to non-compliant content generation, and use technical measures to limit content generation or usage, or take further technological or manual means to monitor or filter generated contents, which not only increases the cost for service providers or makes it harder for them to develop their generative AI applications, but also impacts the users’ experience in their use of the service.

 

Regarding the Interim Measures provision that for generative AI services with public opinion attributes or social mobilization capabilities, security assessments should be conduct under regulations, and formalities for the filing, modification, or cancelation of filing on algorithm performed under the Algorithm Provisions, this registration requirement and registration system are extremely worthy of attention by AI practitioners. The scope of public opinion attributes or social mobilization capabilities is explicitly defined in the Regulations on the Security Assessment of Internet Information Services with Public Opinion Attributes or Social Mobilization Capabilities issued by the CAC, including information service, such as the forums, blogs, microblogs, chatrooms, communication groups, public accounts, short videos, online live broadcasts, information sharing, mini-programs and other information services or attached corresponding functions; and other internet-based information services providing public opinion channels or being able to mobilize the public to engage in specific activities. Therefore, if a service provider’s generative AI technology involves the aforementioned services, not only should security assessments be conducted in accordance with the national regulations, but algorithm registration procedures also be fulfilled. It can be predicted that both domestic and overseas service providers offering generative AI technology services in China will basically and inevitably encounter issues of security assessments and registration. Since these regulations have just been introduced, the impacts on the development of generative AI technology in China will take some time to emerge.

 

Within a short time of one and a half years, three significant sector regulations concerning AI technology, namely the Algorithm Provisions (effective as of March 1, 2022), the Deep Synthesis Provisions (effective as of January 10, 2023), and the Interim Measures have been formulated by the relevant supervisory agencies and entered into force, which demonstrates the Chinese government's intention to achieve technological innovation and economic growth in the AI field, while watching out its social, ethical, and security impacts, and making response in an effort to achieve balance between AI technology promotion and related risks control by establishing a clear regulatory framework to oversee the application and development of AI technology. This represents a cautious attitude of the Chinese government towards the new AI technology to address the evolving technological and ethical challenges. It can also be anticipated that with the continuous advancement of AI technology, the Chinese government will formulate more rules or regulations to address new issues that would be triggered by the technological progress.

 

Author:

Mr. Richard Yong WANG

Mr. Wang received his bachelor's degree in 1991 from the computer science department of East China Normal University, his master's degree from the Institute of Computing Technology of the Chinese Academy of Sciences in 1994, and also the degree of master of laws from Renmin University of China in 2005. Mr. Wang joined Panawell in 2007. In the past years, Mr. Wang has handled thousands of patent applications for both domestic and foreign clients, and he has extensive experiences in patent application drafting, prosecution, reexamination, invalidation, administrative litigation, infringement litigation, software registration, and layout designs of integrated circuit.