Date Thesis Awarded
4-2024
Access Type
Honors Thesis -- Access Restricted On-Campus Only
Degree Name
Bachelors of Arts (BA)
Department
Economics
Advisor
Robert Hicks
Committee Members
Lisa Anderson
Daniel Vasiliu
Abstract
Our research explores how knowledge of an AI opponent impacts investment behavior in a public goods game, with the aim of shedding light on the dynamics of human-AI trust. We had three treatments of human subjects play the same VCM with varying levels of information about their opponents. While we did not find varying levels of knowledge to significantly impact investment, consistent with existing studies, we found ChatGPT to exhibit greater prosocial behavior than its human counterparts. Regardless of the opponent, human participants were likely to exploit players who exhibited high levels of prosocial behavior, leading to the paradoxical outcome of humans freeriding off AI. We also found the AI to be sensitive to prompting and system role content, suggesting that the desired level of cooperation can be fostered through parameterization. This paper provides a commentary on the current state of human-AI trust through a strategic game’s framework.
Recommended Citation
Smith, Sydney, "Beyond Human Players: Investigating AI Participation in Public Goods Games" (2024). Undergraduate Honors Theses. William & Mary. Paper 2089.
https://scholarworks.wm.edu/honorstheses/2089
