China Legal Science 2025年第2期 | 人工智能法治的现实路径

原创姜伟中国法学 2025年03月21日 08:47

REALISTIC PATH OF THE RULE OF LAW ON ARTIFICIAL INTELLIGENCE

Jiang Wei

TABLE OF CONTENTS

I. INTERVENTION TIMING OF LEGAL GOVERNANCE

II. BALANCE BETWEEN DEVELOPMENT AND SAFETY

III. LEGISLATIVE MODEL OF AI REGULATION

The year 2024 marks the 30th anniversary of China’s full access to the Internet. After long-term unremitting efforts, China has formulated and promulgated more than 150 laws on cyberspace, and the cyberspace rule of law system has been gradually improved from scratch. A path for governing the cyberspace in accordance with the law has been explored, which not only conforms to international common practices but also features Chinese characteristics. In the era of intelligence, the cyberspace environment faced by the world is diﬀerent from that of 30 years ago, or even 10 years ago. The risks of digital technology represented by artificial intelligence (AI) have spilled over from cyberspace and become a common challenge and issue faced by all countries in the world. At present, large AI models are growing explosively, their risks have directly threatened the real society, and the European Union, China, and the United States have put AI legislation on the agenda. Generative AI has impacted the existing rules of order, and will inevitably give rise to AI law that is diﬀerent from both traditional and cyber laws.

In terms of AI governance, countries around the world have both cooperation and competition, and both consensus and disagreement. They are faced with confusion in both theoretical research and system construction.

I. INTERVENTION TIMING OF LEGAL GOVERNANCE

In the development of human society, every major technological revolution will bring new challenges to state governance. The biggest problem facing AI legislation is that the panorama of AI technology development has not yet been fully revealed, and the unknown is far greater than the known. How are the AI characteristics reflected in specific rules? Lawmakers need a process of understanding development and governance laws of AI technology. According to the Collingridge Dilemma proposed by a British scholar, if the law intervenes AI too early or regulates AI too late, it is not conducive to human welfare. Therefore, it is understandable to be cautious about the enactment of AI legislation. However, laws should be formulated to regulate risks that are actually dangerous in a timely manner. The research, development and application of AI must not harm human rights and dignity, which is the greatest consensus of all countries in the world, and AI ethics and security have been widely valued by the international community. China has successively released documents such as the Global Initiative on Data Security and the Global AI Governance Initiative, which have provided valuable references for relevant international discussions and rulemaking. In 2023, 28 countries, including China, the United States, and the United Kingdom, as well as the European Union, signed the world-ﬁrst AI safety document, the Bletchley Declaration, calling on all countries to fully understand the necessity and urgency of AI risk regulation. Therefore, it is necessary to introduce regulatory legislation in a timely manner to address the risk of AI endangering humans.

II. BALANCE BETWEEN DEVELOPMENT AND SAFETY

Encouraging innovation and regulating risk are fundamental principles of AI governance. Lawmakers who have introduced AI acts claim to embody the principle of equal emphasis on development and safety, but the AI plans introduced by each country are highly controversial, and various parties have diﬀerent feelings, especially regulators and technology companies often find it difficult to reach a consensus on whether the act is conducive to innovation and development. Dozens of large European companies, including Airbus and Siemens, have taken collective action to publicly oppose the adoption of the EU AI Act on the grounds that it may damage European competitiveness. The author believes that in the development of digital technology, there can be no absolute safety, only relative safety. The best legislation policy is the development-oriented protection for digital technology, not the prohibition-oriented protection. The widespread consensus in the international community is that there can be no sustainable development without safety, but the absence of development is the greatest unsafety. China has always adhered to the principle of ‘attaching equal importance to development and safety, and promoting innovation and law-based governance’ in AI governance. On September 22, 2024, the United Nations Summit of the Future oﬃcially adopted the Global Digital Compact, emphasizing development, inclusiveness, innovation, protection of cultural diversity in the digital space, and creating an open, fair, inclusive and non-discriminatory environment for digital development. Therefore, to balance the relationship between development and safety, we should emphasize that development is the main line and safety is the bottom line, and clarify development as a priority, so as to ensure safety through development, promote development with safety, prevent risks in development, and control chaos in innovation. Lawmakers should position AI as an emerging technology and a new quality productive force, give priority to the enactment of the AI Promotion Law, encourage innovation, advocate development, provide guarantees, fully respect the autonomy of science and technology, and pay attention to the ethics of science and technology. For the risk regulation of AI research, development and application, we should uphold the concept of ‘prudence and moderation’, and promote inclusive and circumspect supervision, such as adopting fault-tolerant supervision, supportive supervision, and punitive supervision, so as to improve the level of normalized supervision, especially to enhance the predictability of supervision. For situations that have already led to harmful social consequences, sanctionable legal norms can be supplemented for penalty.

III. LEGISLATIVE MODEL OF AI REGULATION

Legislation on AI is new, and all countries are in the exploratory stage. Whether AI law intervenes in the ﬁeld of AI in an all-round way, or selectively intervenes in different fields and stages, countries have chosen different legislative paths based on their specific national conditions, development realities and value orientations. The European Union has adopted a harmonized specific law, and the AI Act is of a codified nature, with the expectation of resolving all issues on development and risk. In the United States, local legislation has been promoted to prevent AI risks through state autonomy. At present, more than 20 states in the United States have proposed AI acts, but the content of the acts varies greatly from state to state. In 2024, the Utah AI Policy Act and the Colorado Act Concerning Consumer Protections in Interactions with AI Systems in the United States came into effect. China has adopted interim methods to carry out experimental regulation, such as the Interim Measures for the Management of Generative AI Services. These legislative models have their own advantages and disadvantages. So far, all the explanations and theories about AI are transitional and still need to be constantly tested by practice. AI legislation also requires a process of gradual deepening of understanding, and requires the development of a long-term governance mindset. It is not advisable to seek quick successes and instant benefits, let alone adopt a radical approach to solving the issue. The author believes that the appropriate way of legal regulationis to adopt composite legislation in specific fields for the current outstanding issues, select the diﬀerent areas and stages of AI to carry out experimental regulation, and release a uniﬁed and standardized ‘comprehensive AI law’ in a timely manner after it has been proven to be eﬀective in practice. Interim law is convenient to ﬁll the gap in the system through short-term policies, and if the interim law does not meet the needs of practice, it is also easy to be quickly corrected and improved in time. Of course, the interim law also needs to coordinate all legislative activities to ensure the steady advancement of the rule of law on AI.

继续滑动看下一个