Psychologically-Valid Generative Agents: A Novel Approach to Agent-Based Modeling in Social Sciences

• DOI: 10.1609/aaaiss.v2i1.27698
• Publication Date: 2024-01-23T00:51:51Z
• Authors: Konstantinos Mits...; Ritwik Bose; Brodie Mather; Archna Bhatia; Kevin Gluck; Bonnie Dorr; Christian Lebiere; Peter Pirolli
• Keywords: Agent-based social simulation; Initialization; Social Dynamics

Incorporating dynamic realistic human behaviors in population-scale computational models has been challenging. While some efforts have leveraged behavioral theories from social science, validated specifically applicable to Agent-based modeling remain limited. Existing approaches lack a comprehensive framework model the situated, adaptive nature of cognition and choice. To address these challenges, this paper proposes novel framework, Psychologically-Valid Generative Agents. These agents consist Cognitive Architecture that provides data-driven cognitively-constrained decision-making functionality, Large Language Model generates human-like linguistic data. In addition, our benefits Stance Detection, Natural Processing technique, allows highly personalize initialization agents, based on real-world data, within simulations. This combination flexible yet structured approach endogenously represent how people perceive, deliberate, respond or other types complex dynamics. Previous work demonstrated promising results by using subset components proposed architecture. Our potential exhibit highly-realistic behavior can be used across variety domains (e.g., public health, group dynamics, psychological sciences, financial markets).