Empowering Economic Simulation for Massively Multiplayer Online Games through Generative Agent-Based Modeling (2506.04699v1)

Abstract: Within the domain of Massively Multiplayer Online (MMO) economy research, Agent-Based Modeling (ABM) has emerged as a robust tool for analyzing game economics, evolving from rule-based agents to decision-making agents enhanced by reinforcement learning. Nevertheless, existing works encounter significant challenges when attempting to emulate human-like economic activities among agents, particularly regarding agent reliability, sociability, and interpretability. In this study, we take a preliminary step in introducing a novel approach using LLMs in MMO economy simulation. Leveraging LLMs' role-playing proficiency, generative capacity, and reasoning aptitude, we design LLM-driven agents with human-like decision-making and adaptability. These agents are equipped with the abilities of role-playing, perception, memory, and reasoning, addressing the aforementioned challenges effectively. Simulation experiments focusing on in-game economic activities demonstrate that LLM-empowered agents can promote emergent phenomena like role specialization and price fluctuations in line with market rules.

Empowering Economic Simulation in MMOs with Generative Agent-Based Modeling

This paper presents a pioneering approach in Massively Multiplayer Online (MMO) economy simulations through the use of LLMs integrated with Generative Agent-Based Modeling (GABM). The authors identify and address several limitations in traditional Agent-Based Modeling (ABM), specifically regarding its ability to emulate human-like economic activities involving agent reliability, sociability, and interpretability. Leveraging the capabilities of LLMs, the research introduces LLM-driven agents with superior role-playing proficiency, generative capacity, and reasoning aptitude to overcome these limitations.

Key Contributions and Framework

The paper’s main contribution is the development of an LLM-empowered MMOAgent within a sophisticated virtual environment designed for MMO economics. The key components of the MMOAgent framework include:

• Profile Design: The framework generates player profiles using realistic game data, employing k-means clustering and utilizing GPT-4 for profile synthesis. This data-driven approach facilitates the creation of agents that more accurately reflect player behavior patterns.
• Perception and Action: Agents interpret their environment through a parsing module that converts raw game observations into comprehendible text. Structured actions with explicit semantics are mapped to predefined game functions, enhancing decision-making efficiency.
• Reasoning with Feedback: The paper innovates feedback-enhanced reasoning by incorporating Chain-of-Thought (CoT) prompting, enabling agents to adjust actions based on execution feedback, thereby improving their decision-making process.
• Memory Modules: The design includes Short-Term Memory (STM) and Numeric-aware Long-Term Memory (LTM) modules, capturing recent trajectories and emphasizing significant game experiences, respectively. This design emulates cognitive processes akin to human decision-making.

Experimental Results

The paper evaluates the performance of MMOAgent against several baselines, including rule-based and RL-based models, across different resource scenarios (Rich, Moderate, Scarce). Key findings include:

• Superior Performance: MMOAgent significantly outperforms baselines in both capability and activity diversity metrics, demonstrating better comprehension of the game environment.
• Human-like Consistency: The agent’s decision-making aligns closely with its assigned profiles, indicating the framework's capacity to simulate realistic human-like player behaviors.
• Economic Phenomena Simulation: The system simulates macroeconomic principles, such as supply and demand dynamics and the equality-profitability trade-off, reinforcing its utility as a research tool for MMO economies.

Implications and Future Developments

The integration of LLMs with GABM for MMO economic simulations holds significant implications for both academic research and gaming industry applications. The framework enhances our understanding of virtual economy dynamics, acting as a testing ground for economic strategies and policies in complex, interactive environments.

Future developments could explore the expansion of such LLM-empowered models beyond MMO contexts, adapting them for broader economic simulations consistent with real-world applications. However, challenges related to the inherent hallucination of LLMs and limitations in interpretability remain to be addressed.

Overall, this paper extends the frontiers of what can be achieved with agent-based simulations, adding a robust layer of generative and adaptive behavior through the integration of cutting-edge LLMs.