Games of Diplomacy: Eliciting Spite Against Automated Opponents Through Structured Game Environments

Abstract

Social games require a nuanced understanding of human behavior and are uniquely difficult to solve computationally, even with access to large behavioral datasets. Examining specific aspects of social gameplay, however, can simultaneously yield insight into human behavior and inform the design of more accurate, human-like artificial agents. Understanding and modeling human social behaviors is therefore central to the development of intelligent virtual agents, whether the objective is player profiling, game design, agent modeling, or robustness to diverse social behaviors. Previous research in social games and behavioral game theory has explored player types and factors such as social value orientation, but has often overlooked social emotions and behaviors like spite that deviate from classical rationality and yet are pervasive in competitive human interaction. This paper specifically examines spiteful behavior, a common but underexplored social phenomenon in the agent design literature. We use a variation of the board game Diplomacy as a controlled experimental environment to manipulate parameters such as risk and cost associated with spiteful actions. Our study successfully elicited spiteful behavior in human participants playing against an automated opponent compared to controls across all experimental conditions. Moreover, our findings suggest that the decision to act spitefully may not be strongly influenced by associated costs and risks, contrary to initial hypotheses. We discuss implications for the design of intelligent agents that must recognize, model, or reproduce spiteful behavior in service of more realistic and human-like social interaction.