News Center



Home News Center News

How General AI Enables Experimental Economic Research

Jan 18-2025   



*The 4th Peking University-Tsinghua University Seminar on China's Economy was held at the National School of Development at Peking University. This article is based on a speech by Associate Professor Liu Xiao of Tsinghua University.

 

Artificial General Intelligence (AGI) is influencing and changing our work and life at an unprecedented speed, and AI technology, represented by GPT, brings new opportunities and challenges to experimental economics research. First, we will explore how to make full use of AGI tools in the conception of experimental design, the operation of the execution process, and the interpretation of data analysis. Second, we will explore how the experimental economics research paradigm can be applied to the study of machine behaviour and human-machine interaction. Finally, we aim to refine and construct a new research paradigm of “digital experimental economics” and its core content, and to provide new ideas and directions for the development of the field of experimental economics.

 

Experimental economics starts from micro theory and uses experimental interventions to understand how people make decisions, including measures of rationality, coordination and strategic thinking. The development of AGI can strengthen experimental economic research. In the design of experiments, AGI can help researchers formulate research questions, conduct literature reviews, design hypothesis tests and experiments, and also assist with code, text and programming; in the conduct of experiments, AGI can help monitor subject group participation, cheating behaviour, etc.; and in the analysis of experiments, AGI can help clean up data, simplify data analysis and interpretation, etc.

 

In addition to the help of AGI in experimental economics, a deeper issue is how to use experimental economics to understand AGI, how to measure the rationality of large models, and how to compare the rationality of each model, which is the frontier of experimental economics in the study of AGI today. Traditional experimental economics represents the rationality of individual people based on the generalised display preference axiom, and provides a number of metrics to measure individual rationality, such as the Critical Cost Efficiency Index (CCEI), the Houtman and Maks Index (HMI), the Money Pump Index (MPI), the Minimal Cost Index (MCI), and so on.

 

We conducted experiments in 96 supermarkets in 12 prefecture-level cities in Hunan Province, and the results showed that the Big Model is more rational and homogeneous than humans in all preference environments. The Big Model is more risk-seeking, altruistic and patient than humans. Analyses based on model characteristics, on the other hand, show that model size, whether it is aligned or not, and text length all significantly affect the rationality of big models.

 

Looking ahead, we are optimistic that AGI will enhance experimental economic research. The use of AGI for experimental support can improve the consistency of experiments and promote uniformity and standardisation, and automated experimental operations also provide new ideas for the reproducibility and scalability of experiments.

How General AI Enables Experimental Economic Research

Jan 18-2025   



*The 4th Peking University-Tsinghua University Seminar on China's Economy was held at the National School of Development at Peking University. This article is based on a speech by Associate Professor Liu Xiao of Tsinghua University.

 

Artificial General Intelligence (AGI) is influencing and changing our work and life at an unprecedented speed, and AI technology, represented by GPT, brings new opportunities and challenges to experimental economics research. First, we will explore how to make full use of AGI tools in the conception of experimental design, the operation of the execution process, and the interpretation of data analysis. Second, we will explore how the experimental economics research paradigm can be applied to the study of machine behaviour and human-machine interaction. Finally, we aim to refine and construct a new research paradigm of “digital experimental economics” and its core content, and to provide new ideas and directions for the development of the field of experimental economics.

 

Experimental economics starts from micro theory and uses experimental interventions to understand how people make decisions, including measures of rationality, coordination and strategic thinking. The development of AGI can strengthen experimental economic research. In the design of experiments, AGI can help researchers formulate research questions, conduct literature reviews, design hypothesis tests and experiments, and also assist with code, text and programming; in the conduct of experiments, AGI can help monitor subject group participation, cheating behaviour, etc.; and in the analysis of experiments, AGI can help clean up data, simplify data analysis and interpretation, etc.

 

In addition to the help of AGI in experimental economics, a deeper issue is how to use experimental economics to understand AGI, how to measure the rationality of large models, and how to compare the rationality of each model, which is the frontier of experimental economics in the study of AGI today. Traditional experimental economics represents the rationality of individual people based on the generalised display preference axiom, and provides a number of metrics to measure individual rationality, such as the Critical Cost Efficiency Index (CCEI), the Houtman and Maks Index (HMI), the Money Pump Index (MPI), the Minimal Cost Index (MCI), and so on.

 

We conducted experiments in 96 supermarkets in 12 prefecture-level cities in Hunan Province, and the results showed that the Big Model is more rational and homogeneous than humans in all preference environments. The Big Model is more risk-seeking, altruistic and patient than humans. Analyses based on model characteristics, on the other hand, show that model size, whether it is aligned or not, and text length all significantly affect the rationality of big models.

 

Looking ahead, we are optimistic that AGI will enhance experimental economic research. The use of AGI for experimental support can improve the consistency of experiments and promote uniformity and standardisation, and automated experimental operations also provide new ideas for the reproducibility and scalability of experiments.