AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society (2502.08691v1)

Abstract: Understanding human behavior and society is a central focus in social sciences, with the rise of generative social science marking a significant paradigmatic shift. By leveraging bottom-up simulations, it replaces costly and logistically challenging traditional experiments with scalable, replicable, and systematic computational approaches for studying complex social dynamics. Recent advances in LLMs have further transformed this research paradigm, enabling the creation of human-like generative social agents and realistic simulacra of society. In this paper, we propose AgentSociety, a large-scale social simulator that integrates LLM-driven agents, a realistic societal environment, and a powerful large-scale simulation engine. Based on the proposed simulator, we generate social lives for over 10k agents, simulating their 5 million interactions both among agents and between agents and their environment. Furthermore, we explore the potential of AgentSociety as a testbed for computational social experiments, focusing on four key social issues: polarization, the spread of inflammatory messages, the effects of universal basic income policies, and the impact of external shocks such as hurricanes. These four issues serve as valuable cases for assessing AgentSociety's support for typical research methods -- such as surveys, interviews, and interventions -- as well as for investigating the patterns, causes, and underlying mechanisms of social issues. The alignment between AgentSociety's outcomes and real-world experimental results not only demonstrates its ability to capture human behaviors and their underlying mechanisms, but also underscores its potential as an important platform for social scientists and policymakers.

An Overview of "AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society"

The paper "AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society" proposes a novel framework for simulating social dynamics through the integration of LLM-driven generative agents in a simulated societal environment. This innovation highlights a shift from traditional methodologies, requiring costly experiments, to a computational approach that facilitates the paper of complex social dynamics through scalable, replicable simulations.

"AgentSociety" fundamentally enhances the capabilities of agent-based modeling by utilizing cutting-edge LLMs to endow agents with human-like cognitive and behavioral characteristics. This system simulates societal phenomena by populating a virtual environment with over 10,000 autonomous agents, facilitating up to 5 million simulated interactions among agents and their environment. The utility of this large-scale simulation is demonstrated through explorations of significant social issues, such as polarization, the dissemination of inflammatory content, the implementation of universal basic income policies, and the impacts of external shocks like natural disasters.

Key Components of AgentSociety

1. LLM-Driven Generative Agents: The simulation platform leverages LLMs to create agents that simulate human-like cognition, encompassing emotions, needs, motivations, and environmental cognition. These agents can engage in dynamic social behaviors, facilitated by their internal psychological states and interactions with a realistically designed environment.
2. Societal Environment Modeling: AgentSociety incorporates a sophisticated environmental framework, including urban spaces with realistic mobility options, a social interaction platform, and economic systems mirroring real-world dynamics. This environment serves as a substrate for the agents' actions and interactions, providing essential feedback and constraints that shape their behavior.
3. Large-scale Simulation Engine: The platform utilizes a high-performance simulation engine to manage the interactions of up to 10,000 agents concurrently. The engine supports a range of traditional social science methodologies like surveys, interviews, and simulated interventions, offering a versatile testbed for research.

Experimental Evaluation and Results

The paper demonstrates the practical applicability of AgentSociety through several experiments aligned with real-world social phenomena:

• Polarization: The system successfully models how polarization emerges and evolves, validating its simulation accuracy against observed social patterns.
• Spread of Inflammatory Messages: Experimental setups assess the dynamics of message propagation and the effectiveness of intervention strategies in controlling the dissemination of inflammatory content.
• Universal Basic Income (UBI): The impact of economic policies on individual and aggregate behaviors is evaluated, with findings aligned with known outcomes from existing UBI implementations.
• External Shocks: Simulations of scenarios such as hurricanes reveal the adaptability and response patterns of agents under stress, highlighting the environment's role in shaping behavior.

Implications and Future Directions

AgentSociety stands out as a promising tool for both theoretical investigations and practical applications. It opens new avenues in computational social science, offering a high-fidelity platform for the examination of complex social processes. The implications of this research extend to policymakers seeking cost-effective means for policy testing and forecasting societal reactions to interventions.

Moreover, as LLMs continue to improve, the realism and applicability of such simulations are expected to increase, potentially transforming societal planning and crisis management. Future developments could expand the scale and scope of AgentSociety, integrating additional variables or domains, enabling more comprehensive studies of human behavior interactions at societal scales.

In conclusion, AgentSociety represents a significant advancement in agent-based modeling, combining the cognitive capabilities of LLMs with robust simulation infrastructures. This integration provides a robust framework for simulating social dynamics effectively, positioning it as a foundational tool for understanding and managing the complexities of human societies.