The New Phase of Compliant Development in the AI Industry: Optimizing the Path to 'Overtaking on the Curve'
-
Compared to other countries, China possesses a vast foundation in the real economy and is accelerating the construction of a modern industrial system. This creates a larger, more urgent, and more valuable practical demand for the deep integration of AI technology with industry applications, providing a broader space for innovation and practice in AI technological advancements. This is also the opportunity for China's domestic industry in the field of AI during the era of large models.
With the rapid development of domestic generative AI, relevant regulatory policies are gradually being implemented. The "Interim Measures for the Management of Generative Artificial Intelligence Services" (hereinafter referred to as the "Measures"), jointly issued by seven departments including the Cyberspace Administration of China, officially came into effect on August 15. As China's first normative regulatory document targeting generative AI, the "Measures" aim to "promote the healthy development and standardized application of generative AI," emphasizing the principles of balancing development and security, fostering innovation alongside lawful governance, and implementing inclusive, prudent, and categorized regulatory approaches for generative AI services.
Under the increasingly refined regulatory framework, China's AI industry is entering a new phase of compliant development. As experts have noted, "the greatest insecurity lies in not developing."
Leveraging Regulatory Technology to Activate the AI Market
"The advent of ChatGPT has sparked a new wave of AI revolution, bringing about extensive and profound changes in the relationships between humans and machines, technology and industry, and the virtual and real worlds. Technological innovation also poses challenges to the civilized order of human society," said Yu Wei, Dean of the Compliance Technology Research Institute at Southern Finance Omnimedia Group, during a recent seminar titled "China's Legislative Approach to AI in the Context of Global Governance Discourse Competition," hosted by the Institute of Law at the Chinese Academy of Social Sciences.
Currently, a global intellectual discourse competition is underway, driving a new wave of AI regulation. Yu Wei explained that Europe is seeking dominance in AI regulation, having long prioritized legislative efforts. In 2019, the European Commission released the "Ethics Guidelines for Trustworthy AI," proposing seven criteria for evaluating trustworthy AI. In 2020, the EU issued the "White Paper on Artificial Intelligence," offering various policy options for AI regulation. On June 14 this year, the European Parliament overwhelmingly voted to pass the draft EU AI Act, which is expected to be formally approved after final negotiations.
Wang Jun, Chief Researcher at the Southern Finance Compliance Technology Research Institute, suggested that China could learn from Europe's regulatory experience in areas such as establishing scenario-based systems, clarifying the government's leading role while coordinating multi-stakeholder governance, and scientifically applying technical tools.
"We have yet to achieve a comprehensive categorization and scenario-based differentiation of risks due to their complexity, which creates a contradiction between singular governance techniques and diverse governance scenarios. Moving forward, we can refine regulation based on scenarios, implementing differentiated oversight for different technical approaches, application models, and responsible entities across various risk points," Wang Jun said. He added that a tiered governance model could be adopted, allowing room for trial and development in medium- and low-risk areas while actively employing regulatory technology to select suitable models and invigorate the AI market.
Gu Haiyan, General Manager of the Legal Department at Sina Group, similarly advocated for more differentiated regulatory measures. For example, the draft EU AI Act adopts a risk-based regulatory approach, categorizing AI into unacceptable risk, high risk, low risk, and minimal risk, with corresponding legal obligations for each.
"We must address the risks accompanying the AI boom," Yu Wei pointed out. For instance, at the data level, how to establish high-quality language databases and strengthen end-to-end data compliance management in response to the massive data demands of generative AI. Legally, whether the outputs of generative AI can constitute works as defined by copyright law remains contentious, and further clarification is needed on copyright ownership. Additionally, risks such as discrimination, bias, and the spread of misinformation are amplified by the large-scale data training methods of big models. Teaching AI ethical principles to ensure precise corrections while balancing fairness and efficiency requires further research.
Building a Computing Power Ecosystem to Support the Development of the AI Industry
Data shows that in the first half of this year, over 100 large-scale models were released domestically. According to incomplete statistics, there are currently about 80 large-scale models in China with parameters exceeding 1 billion. The proposed measures include promoting the construction of generative AI infrastructure and public training data resource platforms, facilitating the collaborative sharing of computing resources, and improving their utilization efficiency. Additionally, efforts are being made to classify and open public data in an orderly manner to expand high-quality public training data resources. The use of secure and trustworthy chips, software, tools, computing power, and data resources is also encouraged.
Computing power serves as the foundation of the digital era and the engine for AI development. According to the latest data from the Ministry of Industry and Information Technology, as of the end of June, the total scale of data center racks in operation nationwide exceeded 7.6 million standard racks, with a total computing power of 197 exaflops (197EFLOPS). Over the past five years, the annual growth rate of computing power has been nearly 30%, and the total storage capacity has surpassed 1,080 exabytes (EB).
Liu Yunjie, an academician of the Chinese Academy of Engineering, stated that China's computing power industry has broad prospects for development, given the country's status as a manufacturing powerhouse and the substantial demand for computing power in both the real economy and consumer sectors like gaming, AR, and VR. "With policy support and technological advancements, the vision of on-demand computing power is within reach. In the future, we will be able to use computing power as easily as water and electricity," he said.
However, he also emphasized that for China's computing power network to meet the demands of large-scale models, coordinated development across all aspects is necessary. For instance, training data is essential for building general or industry-specific large-scale models, which requires the proper protection, utilization, and management of industry data.
Leveraging Technological Trends to Advance Industry-Specific Large-Scale Models
Generative AI's capabilities extend far beyond providing information services; it can serve as a "technological foundation" empowering industries such as finance, healthcare, and autonomous driving. In the future, it is expected to become a "technological infrastructure" for society. The proposed measures explicitly encourage the innovative application of generative AI technology across various industries and fields to produce positive, healthy, and high-quality content, explore optimized application scenarios, and build an ecosystem of applications.
"Compared to general-purpose large-scale models like ChatGPT, our shortcomings are quite evident," Liu Yunjie remarked at the 2023 China Computing Conference. "China's opportunity lies in industry-specific large-scale models."
General-purpose large-scale models typically refer to large deep learning models widely applicable across multiple domains, while industry-specific large-scale models are designed for particular vertical industries. These models are trained on datasets specific to those industries to improve accuracy and efficiency. Examples include risk control models in the financial sector.
Liu Yunjie explained that building on the foundational capabilities of general-purpose large-scale models, the development of industry-specific large-scale models has become an inevitable trend in technological advancement. On one hand, industry-specific knowledge and experience can be incorporated into the models to enhance their quality and accuracy. On the other hand, these models can continuously learn and iterate, helping businesses better understand industry trends and make more informed decisions.
Currently, China has gradually established a systematic R&D capability covering theoretical methods and hardware-software technologies. A number of influential pre-trained large-scale models have flourished, forming a technology cluster that keeps pace with global advancements. For example, Huawei Cloud's Pangu large-scale model has been applied in industries such as mining, pharmaceuticals, power, meteorology, and oceanography, with over 1,000 innovative projects implemented. By providing advanced algorithms and solutions, it drives full-stack independent innovation in large-scale models and accelerates the localization of computing power.
Meanwhile, compared to other countries, China possesses a massive industrial foundation and is accelerating the construction of a modern industrial system. This creates larger, more urgent, and more valuable practical demands for the deep integration of AI technology with industrial applications, providing broader innovation practice space for AI technological advancements. This represents the opportunity for China's domestic industries in the field of artificial intelligence during the era of large models.