Simulating Economic Experiments Using Large
Language Models: Design and Development of a
Computational Tool
Name
kar-skar-meng-eecs-2023-thesis.pdf
Description
Thesis PDF
Size
1.49 MB
Format
Adobe PDF
Checksum (MD5)
78b175696968f00b0e99f0cbbd5df17f
Author(s)
Kar, Sohini
Advisor(s)
Horton, John
Date Issued
June 2023
Publisher
Massachusetts Institute of Technology
Abstract
Large language models, known for capturing the syntax, semantics, and broader representation of human behavior, can be used to simulate humans in AI-based experimentation. Because these models provide responses consistent with humans, they may be used to pilot studies or glean insights into social and economic scenarios. This research presents the development of homo silicus, a Python-based library for simulating economic experiments and social scenarios using AI subjects. Using the library, users can design, test, and analyze results of experiments conducted in silico. We replicate various classical economic research to better understand how well AI subject responses align with observed human behavior, and we test the impact of parameters such as temperature and prompt engineering to determine their influence on results. The homo silicus library provides researchers with a cost-effective method to iterate on projects without extensive resources or specific participant recruitment. This research contributes to advancing the field of AI-powered social simulation models and their applications in economic experimentation.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
