LLMs and generative agent-based models for complex systems research.

Area of Science:

• Computational Social Science
• Artificial Intelligence
• Complex Systems Modeling

Background:

• Large Language Models (LLMs) are emerging as powerful tools for scientific research.
• Generative Agent-Based Models (GABMs) integrate LLMs to simulate human behavior and complex interactions.
• LLMs are disrupting fields like network science, evolutionary game theory, social dynamics, and epidemic modeling.

Purpose of the Study:

• To review the disruptive role of LLMs in various scientific fields.
• To assess advancements in using LLMs for social behavior prediction, cooperation enhancement, and disease modeling.
• To identify challenges and future research directions for integrating LLMs in decision-making.

Main Methods:

• Review of recent advancements in LLM applications in scientific modeling.
• Assessment of LLM capabilities in reproducing human-like behaviors (fairness, cooperation).
• Analysis of LLM advantages (cost, scalability) and limitations (prompt sensitivity, hallucinations).

Main Results:

• LLMs can simulate human-like behaviors such as fairness and cooperation.
• LLMs offer advantages like cost-efficiency and scalability in modeling.
• Inconsistencies in LLM behavior due to prompt sensitivity and hallucinations present control challenges.

Conclusions:

• LLMs show significant potential for transforming scientific research and decision-making.
• Addressing biases, prompt design, and human-machine interaction dynamics is crucial for effective LLM integration.
• Future research should focus on refining LLMs, standardizing methodologies, and exploring emergent cooperative behaviors.