China's Draft Measures for Generative AI Governance

Table of Contents

1. Introduction
2. The Draft Measures for Generative Artificial Intelligence Services
   1. Background
   2. Scope and Definitions
   3. Governance Principles
   4. Content Moderation and Algorithm Transparency
   5. Security Assessment and Management
   6. Self-Assessment and Algorithm Disclosure
   7. User Protection and Complaint Mechanism
   8. Proper Use and Marketing Guidelines
   9. Data Acquisition and Annotation
   10. Penalties and Enforcement
3. Conclusion

The Draft Measures for Generative Artificial Intelligence Services

Generative artificial intelligence (AI) has been a topic of discussion and exploration in recent years. As the use of AI technology becomes more prevalent, governments around the world are grappling with the need to regulate this emerging field. In China, the State Administration of China has released draft administrative measures for Generative AI services, which are currently open for public consultation. These measures aim to address various issues related to the governance of generative AI, including data protection, discrimination, bias, and algorithmic transparency.

Background

The Chinese government is keen on establishing its own governance model for new technologies like generative AI. The draft measures apply to both research and development activities as well as the public use of generative AI within the territory of China. Generative AI is defined as technology that generates content, including text, pictures, audio, video, and code, based on algorithms, models, and rules. It is yet to be clarified whether the draft measures apply only to generative AI services offered to consumers in China or if they also encompass services targeting enterprises.

Scope and Definitions

The draft measures define providers of generative AI as entities or individuals that utilize generative AI products to offer services such as chat text, picture, and audio generation. This definition includes service providers that enable others to generate content through APIs or other means. The measures do not distinguish between providers of generative AI offering backend technologies and those building services at the application level. However, regardless of the level of service, providers are required to protect personal information in accordance with China's Personal Information Protection Law.

Governance Principles

The premise of generating value with generative AI must adhere to several principles outlined in the draft measures. Firstly, the generated content must be consistent with social order, societal morals, and national security. Providers are also required to adopt measures that avoid discrimination or biased algorithms in training data. Additionally, the content created by generative AI should be true, accurate, free of fraudulent information, and respect intellectual property rights and other applicable laws and regulations.

Content Moderation and Algorithm Transparency

Providers of generative AI face the challenge of filtering out inappropriate content and optimizing algorithms to prevent the generation of such content. The draft measures mandate that providers should implement mechanisms to identify content produced by generative AI and enable the use of tagging mechanisms. This is to ensure compliance with provisions and management standards related to internet information services.

Security Assessment and Management

Before offering generative AI services to the public, providers must undergo a security assessment conducted by the Cyberspace Administration of China (CSC). This assessment is aimed at assessing the technological features and societal mobilization potential of the services. The assessment provisions, which were introduced in 2018, cover various internet information services, including public forums and live streaming platforms.

Self-Assessment and Algorithm Disclosure

Providers of generative AI are required to disclose information about their algorithms to the CSC. This includes details such as the service provider's name, algorithm type, and self-assessment report. Furthermore, providers must ensure the use of real identity information by end-users and adopt measures to prevent addiction to generative AI services. It is also crucial for providers to disclose information that may impact users' choices, such as the source, quality, and training data used.

User Protection and Complaint Mechanism

Providers of generative AI are obligated to protect user privacy and activity logs. User profiling and sharing of users' preferred preferences are prohibited. Furthermore, providers must establish a mechanism to receive and review complaints from end-users. It is paramount for providers to guide users in the proper utilization of generative AI and prevent misuse that may damage the rights and interests of others or engage in improper marketing.

Data Acquisition and Annotation

Providers of generative AI must adhere to clear rules for data annotation and train employees involved in such tasks. Discriminatory content generation based on characteristics such as race, nationality, or gender is strictly prohibited.

Penalties and Enforcement

Penalties for violating the draft measures are outlined in Article 20. Providers who violate the agreement may face penalties in accordance with the Personal Information Protection Law, Cybersecurity Law, or other Relevant regulations. This could include warnings, corrective actions, service suspension, fines, or even criminal liability, depending on the severity of the violation.

Conclusion

The release of the draft measures for generative AI services by the State Administration of China reflects the government's objective to establish its own governance model for emerging technologies. These measures address various aspects of generative AI, including content moderation, algorithm transparency, security assessment, user protection, and data management. They aim to strike a balance between promoting innovation and protecting the rights and interests of individuals and society. As generative AI continues to advance, it is crucial for governments to collaborate and develop comprehensive regulations to address ethical and societal concerns effectively.