Why do Chinese AI models perform worse than US ones?

Gabriele Iuvinale
30 apr
Tempo di lettura: 6 min

The quantity and quality of training data are critical to the performance of artificial intelligence (AI) models. However, the production of such data in China is hampered by problems such as insufficient quantity, low quality, and fragmented distribution, compounded by limitations arising from commercial ecosystems, regulatory frameworks, and limited public data development and use. But Xi insists on calling for self-sufficiency in AI

Chinese artificial intelligence (AI) has made significant progress.

Indeed, Beijing stands as a major player in the development of AI globally. The country's AI growth path has been characterized by strategic investments, ambitious government plans, and strong public-private synergy.

The recent DeepSeek “case” to me is a demonstration of this.

In addition, the government has set AI as a national priority.

With the 2017 New Generation Artificial Intelligence Development Plan, China seeks to pursue the goal of becoming a world leader in AI by 2030. This plan has received significant political attention and is supported by the highest levels of the Chinese leadership, primarily Xi Jinping, General Secretary of the Communist Party of China (CPC).

Recently, then, there has been a strong integration of AI in healthcare, security, manufacturing, smart cities, and the military.

China also ranks first in the world in STEM graduates and second in research and development (R&D) spending.

Its geographically concentrated technology centers have revolutionized supply chain and manufacturing integration. Government technology policies, such as the Made in China 2025 and Digital Silk Road programs, demonstrate Beijing's focus on a long-term technology strategy.

Chinese investment in AI is estimated to reach a substantial $26.7 billion by 2026, accounting for about 8.9 percent of global AI investment. This will position China as the world's second largest destination for investment in the sector. These figures indicate China's strong commitment and growing influence in the field of AI, both domestically and on a global scale.

However, despite all this, in addition to China's huge population and advanced digital infrastructure that generate huge amounts of data critical for training AI algorithms, the overall performance of Chinese AI is still below the highest levels in the world, especially compared to U.S. models.

In addition to computing power, talent, capital and other factors, the quality of training data delivery and the efficiency of flow and circulation have become key factors limiting model performance and industry innovation.

In China, however, training data face problems of “low quality, low quantity and dispersed distribution,” says a paper titled “Study on Constraints and Policy Responses for the Production and Circulation of AI Training Data in China,” published in the Bulletin of the Chinese Academy of Sciences (Chinese version: Vol. 40: Issue 4 , Article 11).

Specifically, “the insufficient accumulation of high-quality Chinese corpuses, the poor flow of public data, the high cost of data labeling in vertical sectors, and regulatory barriers, such as copyright and personal information protection, contribute to the lack of quality and distribution.”

The cost of acquiring, using and sharing training data also remains high due to regulatory barriers.

All these limitations, the paper argues, make it “difficult to fully alleviate the structural shortage of high-quality data.”

What to do, then?

Improving data quality and promoting data circulation are key to winning the artificial intelligence competition in China, the paper adds.

“Only by fully considering the basic characteristics of data elements and formulating targeted policies and measures can Beijing create a favorable environment for artificial intelligence. Policy measures can lay a solid foundation for the development of artificial intelligence and promote the healthy and sustainable development of the industry.”

These, specifically, are the proposed recommendations:

incentivizing research institutions to generate open source datasets
promotion of AI application scenarios,
adopting a “loosein, focus-out” regulatory approach,
introduction of exemption provisions for intellectual property,
refinement of guidelines on the protection of personal information
accelerating the adoption of a “loosein, focus-out” regulatory approach
accelerating the creation of a unified national platform for public data

China's Xi calls for self sufficiency in AI

Chinese President Xi Jinping has promised “self-sufficiency and self-reliance” to develop AI in China, state media reported Saturday.

Speaking at a Politburo study session Friday, Xi said China should take advantage of its “comprehensive new national system” to advance AI development.

"We must recognise the gaps and redouble our efforts to comprehensively advance technological innovation, industrial development, and AI-empowered applications," said Xi, according to the official Xinhua news agency. Xi noted policy support would be provided in areas such as government procurement, intellectual property rights, research and cultivating talent.

"We must continue to strengthen basic research, concentrate our efforts on mastering core technologies such as high-end chips and basic software, and build an independent, controllable, and collaborative artificial intelligence basic software and hardware system," Xi said.

He added that AI regulations and laws should be speeded up to build a "risk warning and emergency response system, to ensure that artificial intelligence is safe, reliable, and controllable."

Xi said last year that AI shouldn't be a "game of rich countries and the wealthy," while calling for more international governance and cooperation on AI.

Massive Public and Private Investment

The Chinese state has mobilized billion-dollar funds to support research and development in the AI sector.

Chinese tech giants (e.g., Baidu, Alibaba, Tencent, Huawei) are heavily involved and funding startups and innovative projects.

Pilot zones (e.g., Beijing, Shanghai, Shenzhen) serve as hubs for experimentation and development.

Education and Academic Research

Chinese universities, such as Tsinghua and Peking University, are at the forefront of AI research.

Numerous research institutes (e.g., CAICT, CASIA) collaborate with industry to accelerate innovation.

Beijing has also focused heavily on increasing talent training with specific university programs on AI (a subject recently included in university curricula for 28 types of degrees).

The regulatory framework for artificial intelligence

In three decades, Beijing has adopted at least a dozen state laws and regulations on scientific development, namely:

The Law on the Scientific and Technological Progress of the People's Republic of China (revised 2021) Dec. 24, 2021;
The Regulations of the National Science and Technology Award (Third Amendment of the State Council Ordinance No. 731) October 7, 2020;
The Regulations of the People's Republic of China on the Administration of Human Genetic Resources dated May 28, 2019;
The Law of the People's Republic of China on Promoting the Transformation of Scientific and Technological Achievements (revised 2015) of August 29, 2015;
The Regulations of the National Science and Technology Award (Second Revision of State Council Order No. 638) of July 18, 2013;
The Decision of the State Council on the Repeal and Amendment of Certain Administrative Regulations (Excerpts) dated July 18, 2013;
The Decision of the State Council on the Repeal of Certain Administrative Regulations (Excerpts) dated January 15, 2008;
The Law on the Scientific and Technological Progress of the People's Republic of China (revised 2007) dated December 29, 2007;
The Law on Scientific and Technological Popularization of the People's Republic of China of June 29, 2002;
The Decision of the State Council on the Repeal of Certain Administrative Regulations Issued Before the End of 2000 (Excerpts) dated October 6, 2001;
The Law of the People's Republic of China on Promoting the Transformation of Scientific and Technological Achievements (1996) dated May 15, 1996;
The Law on the Scientific and Technological Progress of the People's Republic of China (1993) of July 2, 1993.

AI standardization guidelines

In recent years, as global competition in the field of AI has become increasingly fierce and the impact of the gap of its development has become deeper, the Chinese government has always maintained a positive attitude toward promoting and supporting the development of AI technology.

Therefore, the Guidelines for the Construction of the Comprehensive Standardization System of the National Artificial Intelligence Industry were released in July 2024. They propose that by 2026, more than 50 national and industry standards will be formulated to standardize technical requirements for the comprehensive intelligentization of manufacturing industry processes and intelligent upgrading of key industries using AI technology.

In essence, with them, China aims to strike a balance between promoting AI advances and ensuring responsible and ethical AI practices within the country.

The Guidelines for Building the Comprehensive Standardization System of the National Artificial Intelligence Industry work in tandem with existing regulatory frameworks that provide development guidance for the AI industry, such as:

the Cybersecurity Act;
the Data Protection Act;
the Data Security Act;
the Cryptography Act;
the Personal Information Protection Law (PIPL);
the provisions on deep synthesis;
the administrative provisions on recommender algorithms in Internet-based information services.