On 17 January 2024, the PRC Ministry of Industry and Information Technology (MIIT) issued the draft “Guidelines for the Development of a Comprehensive System of National Industrial Standards for Artificial Intelligence” (the Draft Guidelines) for public comments by 31 January 2024.
The development of these Guidelines is designed to fully exploit the directive role of standards in advancing the high-quality growth of the artificial intelligence (AI) industry. This initiative, poised to profoundly influence the framework of AI development and deployment in China, sets a goal to establish more than 50 new national and industry-specific standards by 2026.
The Draft Guidelines list six major categories of standards that will be central to this development, encompassing a wide range of standards that address various aspects of AI, as outlined below.
I. Fundamental and Common Standards
The Fundamental and Common Standards primarily standardize the content of AI terminology, reference architecture, testing and evaluation, management, and sustainability.
II. Basic Support Standards
The Basic Support Standards mainly include parts such as basic data services, intelligent chips, intelligent sensors, computing devices, computing centers, system software, development frameworks, and hardware-software synergy.
III. Key Technology Standards
i) Machine Learning Standards: These standards regulate the training data, data preprocessing, model expression and format, and model effectiveness evaluation in machine learning. They include standards for self-supervised learning, unsupervised learning, semi-supervised learning, deep learning, and reinforcement learning.
ii) Knowledge Graph Standards: These standards outline the description, construction, operation, sharing, management, and application of knowledge graphs. This includes knowledge representation and modeling, knowledge acquisition and storage, knowledge fusion and visualization, knowledge computation and management, quality evaluation and interconnectivity of knowledge graphs, delivery and application of knowledge graphs, and system architecture and performance requirements.
iii) Large Model Standards: These standards regulate the technical requirements for training, inference, and deployment of large models, including general technical requirements for large models, evaluation metrics and methods, service capability maturity assessment, and content generation evaluation.
iv) Natural Language Processing Standards: These standards set forth the technical requirements and evaluation methods for language information extraction, text processing, and semantic processing in natural language processing. This includes standards for syntactic analysis, semantic understanding, semantic expression, machine translation, automatic summarization, automatic question answering, and large language models.
v) Intelligent Voice Standards: These standards regulate the technical requirements and evaluation methods for front-end processing, voice processing, voice interfaces, and data resources. This includes standards for deep synthesis authenticity detection, full-duplex interaction, and general voice large models.
vi) Computer Vision Standards: These standards set criteria for image acquisition, image/video processing, image content analysis, 3D computer vision, computational photography, and cross-media fusion, including functionality, performance, and maintainability.
vii) Biometric Recognition Standards: These standards regulate the technical requirements for biometric sample processing, biometric data protocols, devices, or systems. This includes standards for biometric data exchange formats and interface protocols.
viii) Human-Machine Hybrid Augmented Intelligence Standards: These standards regulate multi-channel, multi-modal, and multi-dimensional interaction pathways, modes, methods, and technical requirements. This includes standards for brain-computer interfaces, online knowledge evolution, dynamic adaptation, dynamic recognition, human-machine collaborative perception, decision-making and control.
ix) Agent Standards: These standards regulate instances of agents centered around general large models and basic functions and application architectures of agents, including agent-based learning, multi-task decomposition, reasoning, prompt engineering, agent data interfaces and parameter ranges, human-machine collaboration, autonomous agent operation, multi-agent distributed consistency, and more.
x) Swarm Intelligence Standards: These standards define the control, formation, perception, planning, decision-making, and communication of swarm intelligence algorithms. They include standards for autonomous control, collaborative control, task planning, path planning, collaborative decision-making, and networking communication.
xi) Cross-Media Intelligence Standards: These standards regulate the technical requirements for processing, transformation analysis, and fusion applications of multimodal data such as text, images, videos, and audio. This includes standards for data acquisition and processing, modality conversion, modality alignment, fusion and collaboration, and application expansion.
xii) Embodied Intelligence Standards: These standards specify the norms for multimodal active interaction, autonomous behavioral learning, simulation, knowledge reasoning, embodied navigation, and group embodied intelligence.
IV. Intelligent Products and Services Standards
i) Intelligent Robot Standards: These standards specify the technical requirements for the application of artificial intelligence in the field of robotics, including the development of technical application standards for intelligent cognition and decision-making in robots.
ii) Intelligent Transport Vehicle Standards: These standards regulate the technical requirements for perception, recognition and prediction, collaboration and game theory, decision-making and control, and evaluation in intelligent transport vehicles. This includes standards for environmental fusion perception, intelligent recognition and prediction, intelligent decision-making and control, and multi-modal testing and evaluation.
iii) Intelligent Mobile Terminal Standards: These standards define the technical requirements for the application of artificial intelligence in mobile terminals, including image recognition, facial recognition, intelligent voice interaction, as well as standards related to information accessibility and aging-friendly features in intelligent mobile terminals.
iv) Digital Human Standards: These standards regulate the appearance, motion generation, voice recognition and synthesis, and natural language interaction of digital humans. This includes standards for basic capability assessment of digital humans, multimedia synthesis rendering, basic data collection methods, identification, and recognition methods.
v) Intelligent Service Standards: These standards regulate services provided based on AI technologies like large models, natural language processing, intelligent voice, and computer vision. This includes standards for testing and evaluation of intelligent programming, intelligent design, and intelligent anti-counterfeiting.
V. Industry Application Standards
The key areas of AI applications include intelligent manufacturing, smart home, smart cities, and scientific intelligent computing.
VI. Safety/Governance Standards
i) Safety Standards: These standards regulate the safety requirements for the entire lifecycle of AI technology, products, systems, applications, and services. This includes standards for basic safety, data, algorithm, and model safety, network technology and system safety, safety management and services, safety testing and assessment, safety labeling, content identification, product and application safety.
ii) Governance Standards: Tailored to the actual needs of AI governance, these standards regulate the technical research and operational service requirements of AI. This includes technical requirements and evaluation methods for AI robustness, reliability, and traceability, as well as AI governance support technologies. They also standardize the ethical governance requirements throughout the AI lifecycle, including AI ethical risk assessment, fairness and interpretability in AI, ethical governance technical requirements and evaluation methods, and AI ethical review standards.
The Draft Guidelines issued by the MIIT cover everything from fundamental and common standards to safety/governance standards and are expected to have a profound impact on China’s AI industry. As per the plan, these standards will be introduced gradually within the next few years. Based on China’s practices in standard-setting, national standards are typically written/compiled by research institutes, researchers, and experts organized by the State Administration for Market Regulation (SAIC) and the Standardization Administration of China (SAC). Industry standards, on the other hand, are formulated and issued by the competent departments of the respective industries. Regarding AI, multiple departments are involved, mainly including the MIIT, the Ministry of Science and Technology (MOST), and the National Development and Reform Commission (NDRC). We will continue to closely monitor the issuance and implementation of these standards to stay updated with the latest developments.