# The Path of Formulating the Basic Law of Artificial Intelligence in China — Analysis of the Desirability of the EU Artificial Intelligence Act - Type: Literature - Source: Studies in Law and Justice - Date: 2023-09-01 - Original: https://doi.org/10.56397/slj.2023.09.09 - Canonical: https://overview.legal/posts/132567 - Topics: Artificial Intelligence, AI Governance Framework, AI Risk Assessment ## Summary The European Commission released the proposed Regulation on Artificial Intelligence (the EU AI Act) on 21 April 2021, which reflects the EU’s leadership orientation in establishing norms and standards in emerging fields, and also reflects the urgent need for legal unity of the EU as a unified market entity. The Act sets out harmonized rules for the development, placing on the market, and use of AI in the European Union. The ideas of a risk-based approach and experimental governance are of great ## Full text 68 The Path of Formulating the Basic Law of Artificial Intelligence in China — Analysis of the Desirability of the EU Artificial Intelligence Act Xiaotong Bing 1 1 Independent Researcher, China Correspondence: Xiaotong Bing , Independent Researcher, China . doi:10.56397/SLJ.2023.09.0 9 Abstract The European Commission released the proposed Regulation on Artificial Intelligence (the EU AI Act) on 21 April 2021 , which reflects the EU ’ s leadership orientation in establishing norms and standards in emerging fields, and also reflects the urgent need for legal unity of the E U as a unified market entity. The Act sets out harmoni z ed rules for the development, placing on the market, and use of AI in the European Union . The ideas of a risk - based approach and experimental governance are of great significance for reference. However , with the advent of ChatGPT, the Art ificial Intelligence Act has raised questions about the legal applicability of “ human - computer interactive ” generative AI. China ’ s AI governance adopts scene - by - scene and field - by - fie ld legislation, with both “ hard law ” such as laws and regulations, and “ s oft law ” such as industry norms, generally presenting a vertical governance path and lacking a unified basic law guideline. With the gradual shortening of the scientific and technolog ical innovation cycle in the field o f artificial intelligence, the establ ishment of a unified basic law on artificial intelligence should be put on the agenda. Setting up a dual - risk categorization regulatory framework, and categorizing from both macro and micro perspectives may be a good at tempt. It adopts a full chain regulat ory mechanism with a full process and multiple subjects, and clarifies the rights and obligations of all legal subjects in the whole cycle, to jointly assist the development of AI “ fo r the better ” . Keywords: ChatGPT , ar tificial intelligence basic law , risk - based classification , experimental treatment , dual - risk categorization supervision 1. Introduc tion The emergence of ChatGPT can be described as a breakthrough in the field of ar tificial intelligence, especially GP T - 4 has been greatly improved in all asp ects, showing a trend of specialization. Some scholars claim that it will shorten the implementation time of the meta - universe by at least ten years. Compared with the traditional artificial intelligence technology l imited to a certain field, its audience is wider and can be spread to all walks of life, showing good human - machine interaction. And through the continuous accumulation and optimization of massive data, it presents good professionalism and is highly sought after by people from all over the worl d, achieving the Studies in Law and Justice I SSN 2958 - 0382 www.pioneerpublisher.com/slj V ol ume 2 N umber 3 September 202 3 Studies in Law and Justice 69 miracle of consumer growth rate. However, the risks behind the rapid development of new technologies cannot be underestimated. Take Western countries as an example, the discriminatory information and false information contained in ChatGPT disturb the social atmosphere and there are also problems such as infringement of personal privacy and business secrets. As a result, China does not allow the application of ChatGPT in the country at present . In addit ion to the above set of issues, there are also issues affecting digital sovereignty and security, such as the penetration of ideology . It is undeniable that the emergence of generative artificial intelligence has brought scientific and technological progre ss, but also brought huge security ri sks . A t this time, the law needs to play a good guiding and regulating role. To this end, the EU took the lead in issuing the AIA to establish EU - wide norms and standards, and to be a t the forefront of artificial intell igence legislation , which also pr otect s the rights of EU citizens through the principle of “ long - arm jurisdiction ” . I n the meantime , China has also made new progress in artificial intelligence legislation and issued the Interim Measures for the Management of Generative Artificial Intellig ence Services based on fully soliciting opinions, which put forward clear service specifications for providers and required relevant departments to strictly follow the classification - base d principle to supervise or guide. T his is consistent with China ’ s legislative approach to the field of artificial intelligence , which is based on scenarios. For example, the Data Security Law, Personal Information Protection Law, Internet Information serv ice algorithm recommendation managem ent provisions, etc. The absenc e of a legal framework for AI also lacks systematicity, in addition to which the multi - sectoral regulatory landscape may create regulatory overlap and waste regulatory costs. At the same ti me, rapid legislative responses to n ew technologies reflect the eff iciency of legislation, which can solve problems to a certain extent, but cannot play a fundamental role in guiding the law, reflecting the lag of the law. In addition, there are inevitably problems of overlap and illogicalit y in several provisions, which increase the cost of implementation. To sum up, in the future, artificial intelligence products like ChatGPT will emerge in large numbers, and China ’ s artificial intelligence companies will also launch artificial intelligence products in line with socialis m with Chinese characteristics. Until then, if a basic law on artificial intelligence can be enacted, legal risks can be significantly reduced and legal protection will be better provided. 2. A nalysis of the Desirability of t he Artificial Intelligence Ac t 2 .1 Risk - Based Approach The principle of risk - based classification is the core concept of the AIA, which divides artificial intelligence systems into unacceptable risk, high risk, limited r isk , and minimal risk, and decreases from prohibit ion to non - su pervision according to the degree of risk. At the same time, common application scenarios are listed into different risk levels, which are of reference significance. For example, applications t hat threaten people ’ s safety, liveli hoods , and rights as unacce ptable risks are strictly prohibited and subject to severe fines for violators. This kind of risk regulation covers the whole process when the product is put on the market, and the risk type sh ould be assessed quickly, and the co rresponding measures should be adjusted in time when the degree and type of risk change, which shows strong flexibility and pertinen ce . In general, the risk - based approach can satisfy today ’ s AI field and its products an d can also meet the general criteria of the AI f undamental l aw, which is informative. However, after the birth of ChatGPT, the principle of risk - based classification can not be comprehensively and accurately assessed . Unlike previous AI technologies , ChatG PT creates both internal and externa l risks. It is not a techno logical “ black box ” in the traditional sense, i.e., the technical principles are known only to some people, but not to regulators and the public. Generative AI has reasoning capabilities that e ven the R&D team cannot decipher, so internal risks are even mor e unpredictable. In addition, ChatGPT embodies strong human - machine interaction, even if the provider takes preventive measures to control the user s ’ input behavior s, which can not stop them f rom making malicious input. ChatGPT collects and learns every in teractive content, if the malicious content exists for a long time, the adverse consequences will increase significantly. Therefore, after the advent of ChatGPT, the criteria for risk classifi cation are difficult to meet the nee d for universality, and furt her changes maybe needed. Studies in Law and Justice 70 2.2 Supervision Sandbox Mechanism The EC aims to prevent the rules from stifling innovation and hindering the creation of a flourishing AI ecosystem in Europe, by int roducing legal sandboxes that afford breathing room to AI develo pers . The essence of the “ regulatory sandbox ” is an exemption mechanism , which gives a certain fault - tolerant space to promote better development. When enterprises test innovative products, th ey do not need to worry about the co nflict between innovation an d regulatory principles. The specific operation is to establish a temporary framework for action in advance in an uncertain environment and constantly revise it through implementation. The Art ificial Intelligence Act explicitly proposes to support the esta blishment of “ regulatory sandbox ” measures for SMEs and start - ups, reducing the compliance burden. When artificial intelligence products are widely put into use, if they can be fully tested i n a specific region, it is conducive to identify ing risks and p ropos ing targeted countermeasures. It is safer to have mechanisms in place to deal with risks that are difficult to predict. In summary, the principle of the risk - based approach classifies new developments in artificial intellige nce into existing classific ation standards and determines regulatory methods by estimating risks, but it is difficult to achieve in the innovative application of deep synthesis algorithms, and there are certain limitatio ns. As a supervision method, the reg ulatory sandbox mechanism c an not only help the regulated to continuously reduce risks but also provide more effective supervision methods for regulators, which is of great significance. 3. The Institutional Conception o f China ’ s Artificial Intelligence Ba sic Law A good institutional d esign of the basic law of artificial intelligence should not only consider the algorithm governance of a single scene, which is already relatively mature but also consider the complex govern ance of generative artificial intell igence at present and the more intelligent products and technologies that may appear in the future. As the basic law should seek the commonness of different types of technology, the design of norms for the commonness can further reduce the lag of laws. The idea of the Basic Law mainly includes four parts: basic principles, classification standards, supervision mechanism, and responsibility. 3.1 Basic Principles First, a dhere to the overall approach to national security. I n the new era, all aspects of the co untry ’ s work should pay attent ion to both development issues and security issues, and maintaining data security is the meaning of the concept. The extraterritorial application of artificial intelligence such as ChatGPT i s built on the basis of Western valu es and thinking orientation, w hich may lead to ideological penetration in China, especially affecting the value formation of minors . It may also affect China ’ s digital sovereignty and security. Therefore, when formulatin g the basic l aw of artificial intell igence, we must adhere to the overall national security concept and try to build an active defense system with passive data exit, and especially build and strengthen the network attack monitoring platform to focus on pro tecting national data. Second, atta ch equal importance to develop ment and security. Legislation is to guide the goodness of artificial intelligence, and the fundamental purpose is to achieve high - quality development in the field of artificial intelligence . It is based on this principle that Article 3 of the newly issued Interim Measures for the Management of Generative Artificial Intelligence Services stipulates that inclusive, prudent , and classified regulatory methods should be implemented. In particular , technology developers should be ap propriately reduce d the lega l ly binding force and give sufficient space for technological innovation development under the rule of law. Third, a dhere to ethical principles and shall not violate the mandatory provisions o f state and administrative regulatio ns and public order and good customs. A I applications must conform to human ethics before they can be launched, especially in line with Chinese values and international consensus. At present, there are existing platforms to describe human moral cognition a nd behavior in different sce narios, and then form ethical and moral evaluations covering human and artificial intelligence ethical and moral performance. The novelty lies in the fact that in the course of attempting to s urpass the advances of the other sta tes, each state pushes forwa rd towards less human control, reaching potentially, the leve l of - almost - zer o human interference in lethal weapons ’ functions. The endgame may be the complete collapse of human centrism and th e humanization of the Studies in Law and Justice 71 international order . Therefore, the b asic l aw should require that ethical compliance obligations must be done before the release of artificial intelligence products, and do a good job of prohibitions. 3.2 Classification Standards At p resent, China ’ s legislation in the f ield of artificial intellige nce mainly carries out vertical governance according to different scenarios. But there are also common horizontal regulations, such as algorithm security assessment, algorithm filing system , a nd so on. The AIA provides targeted supervision according to the degree of risk, but with the emergence of generative artificial intelligence and the unclear definition of risk, there are difficulties in the specific application of the Act. If you want to develop a general method of artifici al intelligence, th is paper believes that the risk standard is a very worthy idea in the horizontal governance path, but it should not be divided into the degree of risk by enumerating . Th is paper believes that whether t here are internal risks can be used as a criterion to judge trad itional or newly developing artificial intelligence. Traditional artificial intelligence products or services are unidirectional in the way they are provided and cannot achieve human - computer interaction, and the design risks in this process are mainly ext ernal risks such as personal information leakage. At present, China has formed a governance pattern of both “ soft law ” and “ hard law ” for external risks, which can effectively deal with extern al risks. However, with the emergenc e of generative artificial i ntelligence, the risk has gradually changed from external risk to both internal and external risk, and the internal has changed from a simple technical “ black box ” to “ human common ignorance i n the face of strong artificial inte lligence. ” Not only internal risks are becoming more difficult to control but also external risks are more uncontrollable due to the differentiation of human - machine interaction. In the future, with the emergence of more and more strong artificial intellig ence technologies, artificia l intelligence will have higher reasoning ability, and the internal and external risks generated will gradually increase. Therefore, the simple level of risk can not cover all artificial intel ligence technology . This paper belie ves that we can divide tradi tional and emerging artificial intelligence technology according to the degree of internal and external risk and carry out relevant system design to cover artificial intelligence technology mo re comprehensively , which play s a pr eventive and normative role in future artificial intelligence technology. 3.3 Supervision Mechanism The legal governance of artificial intelligence especially new technologies should be development - oriented, but how to a chieve high - quality development is m ore important. Most of the d omestic academic supervision of artificial intelligence technology is in three ways: sub - subject supervision, whole - chain supervision , and sub - model supervision. Regarding sub - subject supervis ion, China has put forward the respo nsibility of compacting the subject in terms of data security and information content security, especially strengthening the responsibility of network service providers such as platforms. At the level of main responsibil ity, it shows the general direction of emphasizing service provi ders over technology developers and users , which effectively promotes the standardized development of the platform, but a single main responsibility cannot meet the increasing development of a rtificial intelligence, especially w hen technical personnel ass ume s an increasingly important role. As a scientific supervision method, full - chain supervision can cover the whole process of artificial intelligence research , development, production , and app lication, which is conducive to the safe development of artifici al intelligence, and should be used as a basic way of supervision. The sub - model regulation is specifically proposed for generative artificial intelligence, and its underlying logic is that ge nerative artificial intelligence is a three - in - one technology fo rm of technical support, service provision , and content, that is, technology developers may also play the role of service providers, so it is impossible to find an appropriate legal status by simply dividing responsibilities thr ough the subject. It is more reasonable to divide it by the basic model, professional model , and service application. Concerning the above regulatory approaches, th is paper believes that a multi - body, full - chain , and dua l - risk categorization regulatory mec hanism can be formed. The es tablishment of a regulatory framework for dual risk categorization is done through a macro and micro perspective. Firstly, from a macro perspective , it is categorized into traditional AI Studies in Law and Justice 72 and e merging AI according to whether it h as internal risk and differen t regulatory principles are set for it. For traditional AI, the principle of safety is adopted to ensure that external risks are continuously reduced. For emerging AI, it adopts the principle of giving equal importance to devel opment and safety, encouragin g innovation, and focusing regulation on after - the - fact risk contingency. Second ly, the other criterion is to focus on the internal of traditional or emerging AI and classify different regula tory standards according to the degr ee of risk. Here, we can lear n from the EU ’ s categorization criteria, which strictly prohibits hazard requirements as unacceptable risks. For high - risk AI systems, the traditional AI sector will be strictly regulated thr oughout the entire process, includin g rigorous assessment beforeh and as well as full - cycle monitoring. In the specific implementation process, all parties should adhere to a macro and micro - consistent regulatory approach. Specifically, first of all, for tr aditional artificial intelligence in the pre - risk management stag e, the regulatory authorities should strictly perform their regulatory duties and take risk control measures, such as risk assessment and filing systems to enhance the transparency and interp retability of the algorithm, which c an control risks from the sou rce. The enterprises are required to conduct regular audits and carry out security vulnerability investigation work r egularly . T echnology developers are required to abide by ethical rules and accept relevant ethical reviews con sciously . They must not viola te not only the public order and good customs but also mandatory provisions of laws and administrative regulations. At this stage, security should be the first value proposition. For emerging artificial intelligence technology, it should not be too strict, and supervision should focus on the risk emergency stage. Since China is in the early stage of generative artificial intelligence technology research and development, the top - level system de sign should leave enough space and t ime for it. A regulatory sand box mechanism can be used to designate specific test areas and specify test times. Secondly, in the risk emergency stage, traditional artificial intelligence should respond quickly to risks: service providers should quickly est ablish rumor - refuting and rep orting mechanisms, take restrictive measures to stop transmission of harmful products, recall defective products in time , or take compulsory destruction . At this time, strengthen ing the respo nsibility of service providers is of significance. For emerging a rtificial intelligence technologies, there is a lack of experience in risk so classification should be adopted. When the risk comes from the client, the service provider should actively perfo rm the obligation of emergency remed y. When the risk comes from a non - client, it should be traced back to the upper level to further identify the source of the problem. Finally, in the post - prevention stage, technical developers should summarize their expe rience in time and modify technical loopholes in time mak ing it c lear that developers should fulfill their product follow - up observation obligations. With the gradual popularization of artificial intelligence, users as the audience must improve artificial intelligence literacy and strictly a bide by regulations . They sha ll not violate public order and good customs and they are supposed to enhance security awareness and pay attention to the protection of their own personal sensitive data and business secrets. 3. 4 R esponsibility In addition to c larifying the legal obligati ons of all parties, the formulation of the basic law of artificial intelligence needs to design relief channels to give victims adequate means of legal relief. The damage caused by artificial intelligence products mainly include s three situations: First, t he damage caused by product defects. The second is the damage caused by the use of products. Third, the damage is caused by its accurate operation following the preset procedure, and there is no fault of others or intermediate l inks. Given the first situat ion, if there are defects in the design and manufacturing of the product, the relevant personnel can be required to bear the responsibility according to the product liability, the designer, th e producer , and the seller bear the responsibility first. After that the party who bears the responsibility first has the right to recover from the person who is at fault. In the second case, if the accident of the artificial intelligence is caused by the person who has the responsibility fo r the management and contr ol of the artificial intelligence product, it needs to be held liable to the extent that it fails to fulfill the obligation of good management. When the third situation occurs, since the party r esponsible for supervision is not at fault and Studies in Law and Justice 73 there are no de fects in the intermediate link, risk management methods can be considered at this time. Look at which parties in this situation can minimize risk and deal with negative impacts. If such party fa ils to fulfill the corresponding ris k management obligations, it shall be liable for damages. 4. Conclusion At present, the rules and systems for artificial intelligence risks are too scattered, and the legal level is low, which is not conducive to enterpr ises to fulfill their obligations an d it may cause difficulty in the supervision of regulatory authorities. Therefore, the development of an artificial intelligence basic law should be put on the agenda. Under the background of the fourth industrial revolu tion, all countries are reserving su fficient legal space for t he development of new technologies as much as possible . T o promote the creation of technology continuously, they choose to “ let the bullets fly a little longer ” . But through the legislation to s olve this stormy interdisciplinary p roblems need to be courag eous and creative, after all, excessive free dom of development is not real progress . T he rule of the l egal track to achieve high - quality development of artificial intelligence is the goal we shou ld pursue. References Liu Xinyu. (20 23). From the meta - universe to Law 3.0: A genealogy of artificial intelligence law . Journal of Sha nghai University , (4), 18 - 28. Liu Yanhong. (2023). Three security risks and legal regulations of generative Artificial Int elligence: A case study of ChatGPT. Oriental Law , 30 - 43. Mauritz Kop. (2021). EU Artificial Intelligence Act: The Euro pean Approach to AI. Stanford - Vienna Transatlantic Technology Law Forum , (2) , 1 - 11. Wang Jianwen. (2023). Cyber and Artificial Intelligen ce Law . Beijing: Law Press. Zeng Xio ng, Liang Zheng & Zhang Hui. (2022). The new development of algorithm governance practice in Europe and America and the construction of comprehensive algorithm governance framework in China . E - Government , (7) , 67 - 75. Zen g Xiong, Liang Zheng & Zhang Hui. (2 022). The regulation path of artificial intelligence in EU and its enlightenment to China. E - Government , (9), 63 - 72. Zhang Linghan & Yu Lin. (2023). From Tra ditional governance to Agile governance: Changing governance pa radigms for Generative Artificial In telligence . E - Government , 136 - 145. Zhang Linghan . ( 2023). The Legal Position and Hierarchical Governance of Generative AI. M odern Law Science , (4) , 126 - 141. Zhang Lu . (2023) . A preliminary study on general artificial int elligence risk governance and superv ision . E - Government , 107 - 117. Zhang Xuebo & Wang Hanrui. ( 2023 ). Legal regulation of generative artificial intelligence . Shanghai Legal Studie, (6) , 246 - 254 . --- Generated by overview.legal · https://overview.legal/posts/132567 · 2026-07-21