OASIS: Open Agent Social Interaction Simulations with One Million Agents (2411.11581v5)

Abstract: There has been a growing interest in enhancing rule-based agent-based models (ABMs) for social media platforms (i.e., X, Reddit) with more realistic LLM agents, thereby allowing for a more nuanced study of complex systems. As a result, several LLM-based ABMs have been proposed in the past year. While they hold promise, each simulator is specifically designed to study a particular scenario, making it time-consuming and resource-intensive to explore other phenomena using the same ABM. Additionally, these models simulate only a limited number of agents, whereas real-world social media platforms involve millions of users. To this end, we propose OASIS, a generalizable and scalable social media simulator. OASIS is designed based on real-world social media platforms, incorporating dynamically updated environments (i.e., dynamic social networks and post information), diverse action spaces (i.e., following, commenting), and recommendation systems (i.e., interest-based and hot-score-based). Additionally, OASIS supports large-scale user simulations, capable of modeling up to one million users. With these features, OASIS can be easily extended to different social media platforms to study large-scale group phenomena and behaviors. We replicate various social phenomena, including information spreading, group polarization, and herd effects across X and Reddit platforms. Moreover, we provide observations of social phenomena at different agent group scales. We observe that the larger agent group scale leads to more enhanced group dynamics and more diverse and helpful agents' opinions. These findings demonstrate OASIS's potential as a powerful tool for studying complex systems in digital environments.

• The paper demonstrates a novel simulation platform that leverages large language models to replicate social media dynamics with up to one million agents.
• It employs a modular architecture—including an Environment Server, RecSys, and Agent Module—to create realistic and scalable social interaction scenarios.
• Experimental findings reveal the platform’s ability to model phenomena such as information propagation, group polarization, and herd effects on platforms like X and Reddit.

OASIS: A Comprehensive Simulator for Social Media Phenomena

The paper presents a sophisticated platform, OASIS (Open Agent Social Interaction Simulations), designed to simulate social media interactions with unprecedented scale and generalizability. The authors introduce OASIS as a novel tool to paper complex social dynamics involving up to one million agents powered by LLMs. The platform seeks to overcome the limitations of existing agent-based models (ABMs) by providing a flexible environment capable of simulating various social media behaviors across different platforms like X (formerly Twitter) and Reddit.

Methodology and Architectural Overview

OASIS is structured around five core components: the Environment Server, RecSys, Agent Module, Time Engine, and Scalable Inferencer. These components collaboratively recreate realistic social media environments. The Environment Server maintains dynamic data such as user profiles and interaction histories, facilitated through a relational database ensuring efficient data management.

The RecSys component provides a sophisticated recommendation system tailored to mimic real-world platforms, incorporating algorithms that account for both in-network and out-of-network content dissemination, leveraging models like TwHIN-BERT for semantic similarity analysis.

Agents in OASIS are guided by LLMs, mimicking human-like interactions through extensive action spaces and detailed character settings, including memory modules for storing interaction histories and Chain-of-Thought reasoning to enhance decision-making processes. This modularity allows OASIS to adapt to different social media settings with ease, providing a flexible platform for simulating diverse interaction types.

Scalability and Performance

Scalability is a cornerstone of OASIS, as the platform supports simulations involving up to one million agents. The scalable design is facilitated by an asynchronous, distributed system architecture, enabling efficient resource utilization across GPU clusters. This allows for large-scale experiments without performance bottlenecks, highlighting the system's engineering prowess.

Moreover, user generation algorithms ensure realistic social network modeling by combining real user data with synthetic profiles, maintaining the scale-free nature of social networks. This approach accommodates complex, large-scale interactions while preserving diversity and authenticity in simulated environments.

Experimental Scenarios and Key Findings

The versatility and effectiveness of OASIS are demonstrated through its application in replicating several real-world social dynamics. Notably, OASIS successfully simulates information propagation and group polarization on X, and herd effects on Reddit.

1. Information Propagation: The platform accurately models the scale and breadth of information spread similar to real-world trends observed on X, though with noted discrepancies in depth, attributed to simplified RecSys design.
2. Group Polarization: Experiments reveal an emergent escalation toward extremism among agent opinions during interactions, particularly pronounced with uncensored models, demonstrating OASIS's ability to reproduce complex sociopsychological phenomena.
3. Herd Effects: In Reddit-like scenarios, the agents exhibit a stronger inclination toward herd behavior than humans, suggesting a tendency for LLM-driven agents to conform reflexively to observed majority behaviors, particularly when content is initially marked as unpopular.

Furthermore, the platform's adaptability to agent scale reveals that larger groups exhibit more diverse and useful interactions, emphasizing the significance of scale in studying collective behaviors.

Future Directions and Ethical Considerations

OASIS stands as a versatile tool for investigating emergent social phenomena, offering valuable insights into the behavior of digital societies. However, future developments could enhance RecSys complexity and incorporate multimodal content for a more authentic simulation. The ethical landscape of deploying such a powerful simulator requires careful attention to ensure bias prevention and privacy protection, alongside sturdy safeguarding against potential misuse, especially in misinformation or manipulation contexts.

In summary, OASIS presents a robust infrastructure for simulating large-scale social media dynamics, demonstrating the potential to significantly advance research in computational social sciences by providing a scalable, adaptable, and detailed virtual environment.