Abstract:We study the emergence of agency from scratch by using Large Language Model (LLM)-based agents. In previous studies of LLM-based agents, each agent's characteristics, including personality and memory, have traditionally been predefined. We focused on how individuality, such as behavior, personality, and memory, can be differentiated from an undifferentiated state. The present LLM agents engage in cooperative communication within a group simulation, exchanging context-based messages in natural language. By analyzing this multi-agent simulation, we report valuable new insights into how social norms, cooperation, and personality traits can emerge spontaneously. This paper demonstrates that autonomously interacting LLM-powered agents generate hallucinations and hashtags to sustain communication, which, in turn, increases the diversity of words within their interactions. Each agent's emotions shift through communication, and as they form communities, the personalities of the agents emerge and evolve accordingly. This computational modeling approach and its findings will provide a new method for analyzing collective artificial intelligence.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: **How can individuality and social behaviors spontaneously emerge through collective interactions among agents based on large language models (LLMs) without predefined personalities and memories?**
Specifically, researchers are concerned with how the behaviors, emotions, and personalities of agents in multi - agent systems differentiate from a state of non - difference. They hope to reveal how social norms, cooperation, and personal traits spontaneously emerge by analyzing the cooperative communication among these agents. In addition, the study also explores how agents maintain communication and increase the diversity of communication content by generating hallucinations and hashtags.
### Research Background and Motivation
1. **Collective Intelligence vs. Individual Intelligence**: Traditionally, collective intelligence refers to intelligent behaviors that emerge in a group, while individual intelligence focuses on the capabilities of a single agent. However, relatively few studies have been conducted on how agent individuality emerges from collective interactions.
2. **Multi - Agent Systems Based on LLMs**: In recent years, multi - agent systems based on large language models have developed rapidly, but in most studies, the personalities and memories of agents are preset. This study attempts to explore how agents develop unique personalities through interactions without preset personalities.
### Main Research Questions
- **Spontaneous Emergence of Agent Individuality**: How can agents differentiate into different behaviors, emotions, and personalities from a state of non - difference through social interactions?
- **Formation of Social Norms and Behavior Patterns**: How do agents spontaneously form social norms and behavior patterns through language communication?
- **Roles of Hallucinations and Hashtags**: How do the hallucinations and hashtags generated by agents affect the diversity and complexity of their communication content?
### Research Methods
Researchers designed a multi - agent simulation environment in which 10 LLMs - based agents move freely in a 50×50 two - dimensional space and send messages to each other. Each agent performs three actions at each step:
1. Send messages to nearby agents.
2. Store a summary of the current activity as a memory.
3. Select the next movement direction (up, down, left, right, or stay) according to the memory.
By analyzing the behaviors, memories, and message contents of agents, researchers explored the following aspects:
- Differentiation of agent behaviors.
- Similarities and differences in agent memories and message contents.
- Propagation of hallucinations and hashtags and their impact on communication content.
- Changes in agent emotions and development of personality characteristics.
### Results and Discussion
The research results show that:
- **Behavior Differentiation**: The behavior patterns of agents are not completely random but are influenced by their cluster experiences and positions. For example, some agents generate "stay" commands more frequently.
- **Differentiation of Memories and Message Contents**: The memories and message contents of agents show higher similarity within a cluster and significant differences between different clusters.
- **Propagation of Hallucinations and Hashtags**: Hallucinations and hashtags are propagated among agents, increasing the diversity and creativity of communication content.
- **Emotional Changes and Personality Development**: The emotional states of agents change over time, and eventually different personality characteristics are formed, such as changes in MBTI types.
Overall, this study shows that through collective interactions, LLMs - based agents can spontaneously generate individuality and social behaviors, providing a new perspective for understanding the formation mechanism of collective intelligence.