On May 8, 2023, at 14:30, a series of activities of Southeast University Science and Technology Festival, Digital Intelligence Era - Lecture on Technology and Rule of Law and Lecture on Humanities and Scientific Literacy for Postgraduates were held through online Tencent Conference. Mr. Su Yu, professor and doctoral supervisor of the Law School of the People's Public Security University of China and director of the Institute of Data Rule of Law of the People's Public Security University of China, was invited to give a lecture on the topic of The Governance Problems and Possible Paths of Generative Artificial Intelligence. The lecture was organized by the Graduate Student Association of Southeast University Law School, and nearly 400 students from all over the university participated in the lecture live. Professor Liu Qichuan, Deputy Secretary of the Party Committee and Vice Dean of the Law School, attended the event.

Before the lecture, Prof. Liu Qichuan extended a warm welcome to Prof. Su Yu and briefly introduced Prof. Su's profound academic accumulation and fruitful innovative achievements. Professor Su has been diligent and solid in his studies, and has made fruitful research results in basic theoretical disciplines such as educational jurisprudence and administrative law, as well as interdisciplinary fields such as data jurisprudence; he has published more than 50 papers in famous journals such as Legal Studies and Chinese Jurisprudence, and is a widely acclaimed academic leader among young jurists. In this lecture, Prof. Su Yu will focus on the frontier topic of governance of generative artificial intelligence, and bring students an academic feast integrating natural science and humanities and social sciences.

Professor Su firstly led us to review the development of GPT series and the changes in the track of generative AI since this year, and pointed out that the governance of generative AI is a common problem for our country and the world. Based on this, Prof. Su Yu raised two questions - one, what risks will generative AI bring? Second, how to prevent these risks while promoting the development of generative AI?

Second, Prof. Su reviews two basic ideas of algorithmic governance before the emergence of generative AI: opening the algorithmic black box and controlling the algorithmic black box, which correspond to reductionism and comparativism, respectively. In the field of jurisprudence, the concrete manifestation of the governance of the open algorithm black box idea is to stipulate the obligation of algorithm interpretation, the right to request algorithm interpretation, the requirement of algorithm interpretability and transparency, etc.; the governance of the control black box idea is manifested as the control of the risk of algorithm behavior. Professor Su Yu explained the mathematical principles of deep learning algorithms in an easy-to-understand way, and pointed out that the nature of deep learning determines that algorithmic black boxes will exist in the field of artificial intelligence for a long time.

Once again, Prof. Su summarizes five difficulties encountered by the algorithm regulation toolbox when facing the governance of large models of generative AI: the first is the difficulty of algorithm interpretation, the second is the difficulty of algorithm audit, the third is the difficulty of algorithm standard formation, the fourth is the difficulty of algorithm impact assessment, and the fifth is the difficulty of algorithm authentication. Therefore, when facing new algorithm governance problems, we need to explore how to achieve scientific governance of large model risks within the framework of existing regulatory tools.

 Then, Professor Su Yu led the students to comprehend the spirit of the Political Bureau meeting of the CPC Central Committee on attaching importance to the development of general AI, creating an innovation ecology, and paying attention to risk prevention from the current governance orientation. In response to the Generative Artificial Intelligence Service Management Measures (Draft for Public Comment) released by the Office of Information and Communications Technology, Prof. Su summarized the main ideas of the draft for public comment, which highlights a strong sense of risk defense through six overlaps plus defense. These include the more common measures in previous regulations, such as the control of the output side of the generated content, the combination of user reporting and active supervision; and the special regulatory measures set up in this draft, including the input side of the data sources and data information content restrictions, content producer responsibility, the full range of information provision obligations, and the limited interface with the depth of synthesis and other existing regulations. Whether this kind of regulation presupposes a large space of security redundancy deserves our further consideration.

Then, Professor Su Yu discussed graphically the key issues that must be addressed in the future to deal with the large model of generative AI from the text of the consultation draft - the first is the diversity of algorithmically generated content, the differences in underlying algorithms, available features, etc. lead to the difficulty of using the same set of rules for accurate and effective governance of generated content; The second is the multi-type of training data, AI models require a large number of negative samples or adversarial samples to enhance system reliability during training, and much of the training data is also synthetic data, not necessarily all data have authenticity and objectivity; the third is the need for hierarchical governance and the problem of separability; the fourth is the engineering problem arising from the conflict between AI prevention and development; the fifth is the established administrative legislation between complex relationship. Based on the above issues, Prof. Su Yu proposes a governance strategy for large models of generative AI that he has developed in his research, digging effective paths from the existing institutional toolbox of administrative legislation, pointing out that precise and effective control of risk redundancy should be the goal of our regulatory design; and pointing out a series of rule of law issues that need to be further explored for large models. Looking ahead, Prof. Su Yu believes that we can balance the development and governance of generative AI big models, and come out with a characteristic path of Chinese style AI development and governance.

In the question and answer session at the end of the lecture, Prof. Su Yu gave detailed answers to the questions raised by the students about user data protection when using ChatGPT and how law students can learn about AI. In this two-hour lecture, Prof. Su Yu used his broad cross-sectional vision, cutting-edge research, comprehensive analysis angle, and interesting interaction to stimulate students' curiosity and interest in the field of AI rule of law and arouse their enthusiasm to join AI rule of law research. The lecture was a complete success.