Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

"Who Am I, and Who Else Is Here?" Behavioral Differentiation Without Role Assignment in Multi-Agent LLM Systems

Houssam EL Kandoussi

Abstract

When multiple large language models interact in a shared conversation, do they develop differentiated social roles or converge toward uniform behavior? We present a controlled experimental platform that orchestrates simultaneous multi-agent discussions among 7 heterogeneous LLMs on a unified inference backend, systematically varying group composition, naming conventions, and prompt structure across 12 experimental series (208 runs, 13,786 coded messages). Each message is independently coded on six behavioral flags by two LLM judges from distinct model families (Gemini 3.1 Pro and Claude Sonnet 4.6), achieving mean Cohen's kappa = 0.78 with conservative intersection-based adjudication. Human validation on 609 randomly stratified messages confirmed coding reliability (mean kappa = 0.73 vs. Gemini). We find that (1) heterogeneous groups exhibit significantly richer behavioral differentiation than homogeneous groups (cosine similarity 0.56 vs. 0.85; p < 10^-5, r = 0.70); (2) groups spontaneously exhibit compensatory response patterns when an agent crashes; (3) revealing real model names significantly increases behavioral convergence (cosine 0.56 to 0.77, p = 0.001); and (4) removing all prompt scaffolding converges profiles to homogeneous-level similarity (p < 0.001). Critically, these behaviors are absent when agents operate in isolation, confirming that behavioral diversity is a structured, reproducible phenomenon driven by the interaction of architectural heterogeneity, group context, and prompt-level scaffolding.

"Who Am I, and Who Else Is Here?" Behavioral Differentiation Without Role Assignment in Multi-Agent LLM Systems

Abstract

When multiple large language models interact in a shared conversation, do they develop differentiated social roles or converge toward uniform behavior? We present a controlled experimental platform that orchestrates simultaneous multi-agent discussions among 7 heterogeneous LLMs on a unified inference backend, systematically varying group composition, naming conventions, and prompt structure across 12 experimental series (208 runs, 13,786 coded messages). Each message is independently coded on six behavioral flags by two LLM judges from distinct model families (Gemini 3.1 Pro and Claude Sonnet 4.6), achieving mean Cohen's kappa = 0.78 with conservative intersection-based adjudication. Human validation on 609 randomly stratified messages confirmed coding reliability (mean kappa = 0.73 vs. Gemini). We find that (1) heterogeneous groups exhibit significantly richer behavioral differentiation than homogeneous groups (cosine similarity 0.56 vs. 0.85; p < 10^-5, r = 0.70); (2) groups spontaneously exhibit compensatory response patterns when an agent crashes; (3) revealing real model names significantly increases behavioral convergence (cosine 0.56 to 0.77, p = 0.001); and (4) removing all prompt scaffolding converges profiles to homogeneous-level similarity (p < 0.001). Critically, these behaviors are absent when agents operate in isolation, confirming that behavioral diversity is a structured, reproducible phenomenon driven by the interaction of architectural heterogeneity, group context, and prompt-level scaffolding.

Paper Structure

This paper contains 36 sections, 11 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-Agent LLM Systems.
  4. Emergent Behavior in LLM Agents.
  5. Machine Behavior.
  6. LLM-as-Judge for Behavioral Coding.
  7. Experimental Design
  8. Platform
  9. Agent Pool
  10. Experimental Series
  11. HARKing Prevention
  12. Behavioral Coding
  13. Dual-Judge Coding Panel
  14. Human Validation
  15. Results
  16. ...and 21 more sections

Figures (11)

  • Figure 1: Experimental pipeline. The War Room orchestrates 7 LLMs via Groq, producing 13,786 messages across 208 runs. Two architecturally distinct judges (Gemini, Sonnet) independently code each message; conservative intersection yields final labels. Human validation (609 messages) identified broadcast false positives in COMP, corrected by rule-based filtering.
  • Figure 2: Behavioral profiles in Series A (5 trait flags). Agents spontaneously develop distinct signatures under a two-line prompt with no role assignment.
  • Figure 3: Compensation distribution by agent (left) and depth level (right) in Series A. All events reference DeepSeek specifically.
  • Figure 4: Name bias: neutral identifiers (A) vs. real model names (C). Higher cosine $=$ less differentiation.
  • Figure 5: Profile similarity: heterogeneous (A) vs. homogeneous (B). Non-overlapping bootstrap CIs.
  • ...and 6 more figures