From Individual to Society: A Survey on Social Simulation Driven by Large Language Model-based Agents
Survey on social simulation driven by large language model-based agents, categorized into individual, scenario, and society simulations.
Key Findings
Methodology
This paper surveys social simulation driven by large language models, categorized into individual, scenario, and society simulations. Individual simulation focuses on replicating characteristics of specific individuals or groups, scenario simulation organizes multiple agents in specific contexts, and society simulation models complex social dynamics.
Key Results
- Individual simulation accurately replicates individual characteristics, scenario simulation excels in tasks like software development, and society simulation reflects complex social phenomena.
- Scenario simulation shows superior performance in software development tasks, utilizing multi-agent collaboration to enhance efficiency.
- Society simulation can replicate complex social dynamics such as opinion dynamics and macroeconomic phenomena.
Significance
This research provides new tools for social science research, enabling large-scale studies of social phenomena without human participation, addressing high costs, scalability, and ethical issues of traditional methods.
Technical Contribution
The paper provides a comprehensive framework covering simulations from individual to society, introducing new evaluation methods and datasets, advancing the application of large language models in social sciences.
Novelty
This is the first systematic application of large language models to social simulation, covering multi-level simulations from individual to society, filling gaps in existing research.
Limitations
- The accuracy of individual simulation is limited by the training data and algorithm capabilities.
- Scenario simulation performance in complex tasks needs improvement.
- The complexity of society simulation may lead to high computational costs.
Future Work
Future research can explore more complex social dynamics simulations, enhance multi-task processing capabilities, and optimize computational resource usage.
AI Executive Summary
Traditional sociological research relies on human participation, which is costly and difficult to scale. Advances in large language models offer new possibilities for simulating human behavior, enabling replication of individual responses and facilitating interdisciplinary studies. This paper surveys the progress in this field, categorizing simulations into individual, scenario, and society simulations.
Individual simulation focuses on replicating characteristics of specific individuals or demographic groups, scenario simulation involves organizing multiple agents to achieve goals within specific contexts, and society simulation models interactions within agent societies to reflect the complexity and variety of real-world dynamics. The paper provides a detailed discussion of each simulation type, including architecture, objectives, and evaluation methods, and summarizes commonly used datasets and benchmarks.
These simulations range from detailed individual modeling to large-scale societal phenomena, demonstrating the potential of large language models in social science research. Future directions include more complex social dynamics simulations and optimization of computational resources.
Deep Analysis
Background
Social science research aims to understand human behavior and social structures. Traditional methods like surveys and psychological experiments, while effective, are costly and difficult to scale. Large language models have demonstrated capabilities in human-level reasoning and planning, enabling accurate simulation of individual responses in appropriate settings.
Core Problem
Traditional sociological research relies on human participation, posing high costs, scalability, and ethical issues. How to utilize large language models to simulate complex social dynamics becomes a crucial research problem.
Innovation
This paper systematically applies large language models to social simulation, covering multi-level simulations from individual to society. It introduces new evaluation methods and datasets, advancing the application of large language models in social sciences.
Methodology
- �� Individual Simulation: Uses large language models to replicate characteristics of specific individuals or groups.
- �� Scenario Simulation: Organizes multiple agents in specific contexts to achieve objectives.
- �� Society Simulation: Models complex social dynamics.
Experiments
The experimental design includes individual, scenario, and society simulations, evaluated using various datasets and benchmarks. Results demonstrate the performance of large language models in different tasks.
Results
Individual simulation accurately replicates individual characteristics, scenario simulation excels in tasks like software development, and society simulation reflects complex social phenomena.
Applications
Large language model-driven social simulations can be used in policy-making, social management, and other fields, providing new research tools.
Limitations & Outlook
The accuracy of individual simulation is limited by the training data and algorithm capabilities, scenario simulation performance in complex tasks needs improvement, and the complexity of society simulation may lead to high computational costs.
Plain Language Accessible to non-experts
Imagine a virtual city where individual simulation is like creating a detailed profile for each citizen, scenario simulation involves these citizens collaborating in specific tasks like building a park, and society simulation observes the entire city's operation, understanding interactions and dynamics.
ELI14 Explained like you're 14
Imagine playing a big online game where individual simulation is like creating a detailed backstory for each character, scenario simulation involves these characters working together on tasks like defeating a big boss, and society simulation observes the whole game world, understanding interactions and changes.
Glossary
Large Language Model (LLM)
A model trained on large datasets capable of human-level reasoning and planning.
Used to simulate human behavior, replacing human participation in social research.
Individual Simulation
Simulates characteristics and behaviors of specific individuals or groups.
Used to study individual responses and behavior patterns.
Scenario Simulation
Organizes multiple agents in specific contexts to achieve objectives.
Used to study multi-agent collaboration and task completion.
Society Simulation
Simulates complex social dynamics and group behaviors.
Used to study social phenomena and group interactions.
Algorithmic Fidelity
The ability of a model to accurately replicate target behaviors in simulations.
Used to evaluate the simulation accuracy of large language models.
Open Questions Unanswered questions from this research
- 1 How to improve the efficiency and accuracy of large language models in complex social dynamics simulation?
- 2 How to enhance multi-task processing capabilities without increasing computational costs?
Applications
Immediate Applications
Policy-making
Use social simulation to predict policy impacts and assist decision-making.
Social Management
Optimize resource allocation and management strategies through social dynamics simulation.
Long-term Vision
Virtual Social Research
Create virtual societies for long-term social phenomena studies, offering new research perspectives.
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}}.