China has issued a guideline for the development of an AI metrology system and related capacity building, targeting industrial bottlenecks including algorithmic "black boxes" and data shortage.

The State Administration for Market Regulation (SAMR), which released the document with the National Development and Reform Commission, said it marks a pivotal shift for China's AI sector from expanding computing power and industrial scale to improving quality and consolidating fundamental capabilities. It is of great significance for advancing in-depth integration of AI technologies with the real economy and accelerating the growth of new quality productive forces.

To bridge the gap between laboratory innovation and industrial application, the guideline centers on six major areas — foundational support, general technologies, core technology, metrology technical standards, industrial metrology industry and AI-enabled metrology.

To address inaccurate measurement and enhance the credibility of AI, it calls for research on key technologies such as monitoring and characterization of internal states of AI systems, in response to pain points including algorithmic black boxes and poor interpretability of decision-making. It promotes reliable, secure and credible metrology standards for AI, enabling the performance of AI technologies to be measurable, comparable and traceable.

The outline of China's 15th Five-Year Plan (2026-2030) calls for breakthroughs in new-type metrology and calibration instruments, such as quantum metrology and in-situ metrology. The guideline supports the establishment of national-level R&D and application centers for metrology technologies, and the development of AI metrology standard devices with independent intellectual property rights.

China will accelerate building full-chain metrology capabilities covering algorithm models, computing power efficiency and data quality. These efforts will create unified measurement benchmarks for AI products.

To tackle data shortage, the guideline mandates developing datasets with top-tier metrological characteristics, standard reference datasets and test datasets. It also aims to put in place a mechanism to share basic resources to break down industrial data silos and make data sharing secure, delivering high-quality data support for training and evaluating AI algorithms.

It promotes in-depth application of metrology technologies across 14 key sectors including smart manufacturing, smart healthcare and smart transportation. Research will be conducted on key indicators such as the reliability of AI diagnostic algorithms to solve quality assessment problems amid industrial digital transformation, and strengthen people's sense of security and satisfaction with AI applications. The focus is on empowering all industries to deliver greater benefits to the public via the smart economy.

The SAMR said it will build a number of R&D and application centers for AI metrology technologies in the next step. Pilot projects will be launched in priority areas such as smart supervision and smart healthcare to develop replicable and scalable application scenarios of "AI + metrology."