From Individual to Society: A Survey on Social Simulation Driven by Large Language Model-based Agents

TL;DR

This survey addresses the limitations of traditional sociological research by detailing how large language model based agents enable scalable social simulations. It presents a three tier framework—individual, scenario, and society simulations—and dissects architectures, construction methods, objectives, and evaluation for each tier. The paper also consolidates datasets and benchmarks, outlines evolving trends from coarse to multi modal simulations, and discusses practical implications for policy and decision making. Overall, it offers a comprehensive blueprint for advancing LLM driven social simulations and their cross disciplinary impact.

Abstract

Traditional sociological research often relies on human participation, which, though effective, is expensive, challenging to scale, and with ethical concerns. Recent advancements in large language models (LLMs) highlight their potential to simulate human behavior, enabling the replication of individual responses and facilitating studies on many interdisciplinary studies. In this paper, we conduct a comprehensive survey of this field, illustrating the recent progress in simulation driven by LLM-empowered agents. We categorize the simulations into three types: (1) Individual Simulation, which mimics specific individuals or demographic groups; (2) Scenario Simulation, where multiple agents collaborate to achieve goals within specific contexts; and (3) Society Simulation, which models interactions within agent societies to reflect the complexity and variety of real-world dynamics. These simulations follow a progression, ranging from detailed individual modeling to large-scale societal phenomena. We provide a detailed discussion of each simulation type, including the architecture or key components of the simulation, the classification of objectives or scenarios and the evaluation method. Afterward, we summarize commonly used datasets and benchmarks. Finally, we discuss the trends across these three types of simulation. A repository for the related sources is at {\url{https://github.com/FudanDISC/SocialAgent}}.

Figures (7)

• Figure 1: Illustration of simulations empowered by LLM-driven agents. We categorize the simulations into individual simulation, scenario simulation and society simulation. From left to right, the diversity and scale of individual modeling generally increase. Conversely, from right to left, the granularity of individual modeling becomes more refined.
• Figure 2: Illustration of individual simulation blueprint. An individual agent is typically composed of an architecture with modules involving profile, memory, planning, and action through construction method, prompting or training, to simulate specific objectives like characters or demographics . Individual simulation can be evaluated statically and interactively with different dimensions being observed.
• Figure 3: Illustration of scenario simulations. Given a specific scenario, building a multi-agent system involves modeling environment, roles, organization, and communication with detailed modules or mechanisms adjusted to the targeted scenario being supported. After simulating the scenario, the desired output, typically the result of a task or problem, is obtained and evaluated using different levels and strategies.
• Figure 4: Illustration of society simulations. To construct society simulations, the corresponding society's construction elements, i.e., composition, network, social influence and outcomes need to be carefully designed. Building on this, various scenarios can be simulated. The performance of individuals and the overall performance of the system are evaluated.
• Figure 5: Illustration of individual simulation trend, which goes through coarse simulation, more nuanced simulation, and situation-oriented simulation.