Cognitive Uncertainty, GPT, and Contribution in Public Goods Game

Abstract: This paper establishes a connection between cognitive noise (Enke and Graeber, 2023) and the level of contribution in the public goods game. Our experimental results demonstrate that a cooperative advice can assist individual in either gaining a better understanding of their true social preference, or translating their true social preferences into contribution actions that maximize their utility as the game repeats. Further, we argue that cognitive noise complements, rather than replaces, taste-based social preference to explain the contribution decision. Our correlational data supports the notion that cognitive uncertainty is positively correlated with contribution in the public goods game at the aggregate level, or cognitive uncertainty lead people to behave as if they are more cooperative. However, there is heterogeneity, where cognitive noise is negatively correlated with the contribution level of some participants at an economically significant extent. These findings suggest the significance of only considering contribution decisions that exceed a certain cognitive certainty threshold in a public goods game if they are to be taken at face value. We also find that advice from the Generative Pre-trained Transformer (hereafter referred to as “GPT”) reduces cognitive uncertainty for all participants, though the impact of the advice does not seem to depend on whether or not the participants are informed the advice was made by GPT.

Keywords: Cognitive Noise Experiments, Public Goods Experiments, GPT