On June 26th, Huarui FinTech Salon (No. 10) and BIMSA Digital Finance Lecture (No. 3), co-organized by Shanghai HuaRui Bank, Yanqi Lake Beijing Institute of Mathematical Sciences and Applications (BIMSA), the International Monetary Institute (IMI) of Renmin University of China (RUC) and RUC FinTech Institute, was held online. The salon cordially invites experts and scholars from various fields, including government, industry, academia, and research, to conduct discussions and research around "New Risks Arising from FinTech Innovation". During the salon, Zhu Shihu, Deputy General Manager of the Information Technology Department and General Manager of the Data Center at Everbright Trust, delivered a keynote speech. Fellow participants of the salon included Han Liyan, Research Fellow of the Yanqi Lake Beijing Institute of Mathematical Sciences and Applications and Professor of the School of Economics and Management of Beihang University, Liu Qingfu, Professor of the School of Economics at Fudan University and Professor of the Digital Economy Laboratory of Yanqi Lake Beijing Institute of Mathematical Sciences and Applications, Luo Yu, Deputy Secretary of the Party Committee and Professor of the School of Finance, RUC and Shen Zhihua, Vice President of Prolos Financial Group and former Retail Risk Technology Director and General Manager of Big Data Decision Management Department of Ping An Bank. The salon was moderated by Long Fei, Director and Research Fellow of the Digital Economy Laboratory of Yanqi Lake Beijing Institute of Mathematical Sciences and Applications.

Zhu Shihu, Deputy General Manager of the Information Technology Department and General Manager of the Data Center at Everbright Trust, delivered a keynote speech on "New Risks Arising from FinTech Innovation", mainly focusing on the following five aspects:

First, he guided everyone to reexamine the definitions of data, labels, and features. He believed that data is an objective existence of facts. Labels are descriptions processed from data, and features are attributes processed for different purposes. These elements, when transferred to the human brain, form our information, or to say, our impressions. He provided detailed explanations and examples to illustrate the distinctions and connections between data, labels, and features, as well as concepts such as data value, data criteria, "data decentralization and centralized management."

Second, he elaborated on the connotations and extensions of FinTech from three aspects: technology, finance, and risk. "FinTech starts with data, thrives with technology, stabilizes with systems, and culminates with values." This statement not only reveals the laws of technological development but also highlights the relationship between productivity and production relations. He emphasized that FinTech is just a tool which can only change the form and expression of risk and finance, not the nature of them.

Third, regarding the risks brought about by new businesses, taking Internet loans as an example, he analyzed the process and reasons behind the development of data-driven businesses. He argued that Internet loans are not a business innovation but rather a dimension reduction strike of account-based credit cards on card-based credit cards. He distinguished two types of Internet loans: shallow (risk complexity) scenario and deep (risk complexity) scenario, and introduced the concept of a three-tier risk pyramid in deep scenario, including customer-level risk, in-scenario risk, and out-scenario risk. He emphasized the importance of identifying repayment willingness and the counter-intuitive phenomena that new businesses may bring. The risks of Internet loans mainly manifest on three levels: customer-level risk, in-scenario risk, and out-scenario risk. Customer-level risk mainly involves the credit status and repayment capacity of borrowers; in-scenario risk pertains to problems that may arise during the design and operation of loan products; out-scenario risk includes the impact that factors such as the macroeconomic environment, policies and regulations may have on the loan business.

Fourth, he discussed the risks brought about by new technologies. He believed that models are the core of technology, and their development has gone through four generations, namely rule-based models, rule+data models, big data models (deep learning models), and foundation models. He compared the advantages, disadvantages of these four generations of models and their applicable scenarios, and pointed out the issues of non-interpretability, black-box nature, and instability associated with foundation models. He advised financial institutions to have clear objectives, appropriate methods, comprehensive validation, and effective monitoring when using new technologies, to avoid blindly pursuing high numerical precision while neglecting actual risk control.

Fifth, unknown risks. First, there are unknown risks brought about by ChatGPT. If a technology is controlled too early for fear of adverse consequences, its potential for significant breakthroughs might be hindered. However, if it is controlled too late, then it may go out of control, and may become more difficult to control in the future. Second, the philosophical boundary between humans and AI remains uncertain. Third, the “subject-object” relationship between humans and AI is subtly evolving, for example, whether ChatGPT is augmenting human intelligence or using human to display its own intelligence.

Translated by Xu Jiayin