Perspective ChatGPT Chats on Computational Experiments: From Interactive Intelligence to Imaginative Intelligence for Design of Artificial Societies and Optimization of Foundational Models By Xiao Xue , Xiangning Yu , and Fei-Yue Wang , Fellow, IEEE    P OWERED by the rapid development of Internet, the pene- tration of the Internet of Things, the emergence of big data, and the rise of social media, more and more complex sys- tems are exhibiting the characteristics of social, physical, and information fusion. These systems are known as cyber-physical- social systems (CPSS) [1], [2]. These CPSS face unprecedented chal- lenges in design, analysis, management, control and integration due to their involvement with human and social factors [3], [4]. To cope with this challenge, there are two main approaches to CPSS research: 1) Data driven analysis method. Regard complex systems as black boxes, focus on the relationship between inputs and outputs, without modeling and analyzing the complex processes within the system. In the practical application, complex systems tend to be replaced by sta- tistical models based on data and intelligent algorithms, such as con- volutional neural network (CNN), recurrent neural networks (RNN), foundation models, etc. The latest ChatGPT (chat generative pre- trained transformer) is a typical example of this approach. 2) Knowledge driven analysis method. According to the principle of “simple consistency”, the complex system in practice can be rec- ognized, understood and analyzed by designing and restoring the structure and function of each system component. The computa- tional experiments method is a representative method [5]. Starting from the micro-scale, it cultivates an “artificial society” of the real system in the cyber world. Then, a variety of experiments can be conducted to identify the causal relationship between intervention variables and system emergence to realize the interpretation, under- standing, guidance and regulation of macro phenomena. Both the two methods have advantages and disadvantages when analyzing complex systems. The knowledge modeling method can effectively capture the essential characteristics and principal contra- dictions of the system, obtaining an effective model structure. How- ever, due to the limited cognitive ability at the time, it can be chal- lenging to accurately describe the operation and evolution mecha- nism of complex systems. In contrast, data modeling method has advanced by leaps and bounds over the years. The advantage of data modeling is that it can automatically acquire the information and knowledge hidden in the data. But, it heavily relies on the quantity and quality of data samples, and conducting an in-depth analysis and interpretation of the system mechanism can be difficult. As shown in Fig. 1, the difference of the two methods can be visually represented by the “cognitive gap” [6]. Integrating different research methods may provide a solution to bridge the “cognitive gap”. Knowledge-driven analysis Data-driven analysis Model error Big data Artificial society models Simulation model Small data Mathematical model Domain data Merton’s law Newton’s law Physical system Cyber-physical system Cyber-physical social system Computational experiments Foundational model Cognitive gap Fig. 1. The “cognitive gap” between data-driven analysis and knowledge- driven analysis. This paper will address two main issues: Firstly, it will explore how ChatGPT can be utilized to improve computational experiments, particularly in the construction of artificial society models. Currently, the manual design of such models poses a significant challenge to the Corresponding author: Xiao Xue. Citation: X. Xue, X. N. Yu, and F.-Y. Wang, “ChatGPT chats on computational experiments: From interactive intelligence to imaginative intelligence for design of artificial societies and optimization of foundational models,” IEEE/CAA J. Autom. Sinica, vol. 10, no. 6, pp. 1357–1360, Jun. 2023. X. Xue and X. N. Yu are with the College of Intelligence and Computing, Tianjin University, Tianjin 300350, China (e-mail: jzxuexiao@tju.edu.cn; yxn9191@tju.edu.cn). F.-Y. Wang is with the State Key Laboratory for Management and Control of Complex Systems, the Institute of Automdtion of Chinese Academg of Sciences, Beijing 100190, China (e-mail: feiyue.wang@ia.ac.cn). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JAS.2023.123585 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 10, NO. 6, JUNE 2023 1357