Modeling Epistemic Uncertainty in Social Perception via Rashomon Set Agents
arXiv:2603.20750v1 Announce Type: new Abstract: We present an LLM-driven multi-agent probabilistic modeling framework that demonstrates how differences in students' subjective social perceptions arise and evolve in real-world classroom settings, under constraints from an observed social network and limited questionnaire data. When social information is incomplete and the accuracy of perception differs between students, they can form different views of the same group structure from local cues they can access. Repeated peer communication and belief updates can gradually change these views and, over time, lead to stable group-level differences. To avoid assuming a global "god's-eye view," we assign each student an individualized subjective graph that shows which social ties they can perceive and how far information is reachable from their perspective. All judgments and interactions are restricted to this subjective graph: agents use retrieval-augmented generation (RAG) to access only loc
arXiv:2603.20750v1 Announce Type: new Abstract: We present an LLM-driven multi-agent probabilistic modeling framework that demonstrates how differences in students' subjective social perceptions arise and evolve in real-world classroom settings, under constraints from an observed social network and limited questionnaire data. When social information is incomplete and the accuracy of perception differs between students, they can form different views of the same group structure from local cues they can access. Repeated peer communication and belief updates can gradually change these views and, over time, lead to stable group-level differences. To avoid assuming a global "god's-eye view," we assign each student an individualized subjective graph that shows which social ties they can perceive and how far information is reachable from their perspective. All judgments and interactions are restricted to this subjective graph: agents use retrieval-augmented generation (RAG) to access only local information and then form evaluations of peers' competence and social standing. We also add structural perturbations related to social-anxiety to represent consistent individual differences in the accuracy of social perception. During peer exchanges, agents share narrative assessments of classmates' academic performance and social position with uncertainty tags, and update beliefs probabilistically using LLM-based trust scores. Using the time series of six real exam scores as an exogenous reference, we run multi-step simulations to examine how epistemic uncertainty spreads through local interactions. Experiments show that, without relying on global information, the framework reproduces several collective dynamics consistent with real-world educational settings. The code is released at https://anonymous.4open.science/r/Rashomonomon-0126.
Executive Summary
This article introduces a multi-agent probabilistic modeling framework that uses Large Language Models (LLMs) to explore how differences in social perception arise and evolve in real-world classroom settings. The framework, named Rashomon Set Agents, takes into account the limitations of social information and the accuracy of perception among students. By assigning each student a subjective graph representing their perspective, the framework simulates how epistemic uncertainty spreads through local interactions. The results demonstrate the framework's ability to reproduce collective dynamics consistent with real-world educational settings, highlighting its potential for understanding social perception and group dynamics. The open-source code and framework provide a valuable resource for researchers and educators seeking to better understand and navigate complex social situations.
Key Points
- ▸ The Rashomon Set Agents framework uses LLMs to model social perception in classroom settings.
- ▸ The framework takes into account the limitations of social information and the accuracy of perception among students.
- ▸ The results demonstrate the framework's ability to reproduce collective dynamics consistent with real-world educational settings.
Merits
Strength in Novelty
The Rashomon Set Agents framework offers a novel approach to understanding social perception and group dynamics, leveraging LLMs to simulate complex social interactions.
Strength in Flexibility
The framework's ability to accommodate individual differences in social perception and accuracy of perception makes it a flexible tool for analyzing diverse social situations.
Demerits
Limitation in Generalizability
The framework's performance may be limited by its reliance on questionnaire data and observed social networks, which may not be representative of all social situations.
Limitation in Computational Resource Intensity
The use of LLMs and probabilistic modeling may require significant computational resources, potentially limiting the framework's applicability in resource-constrained settings.
Expert Commentary
The Rashomon Set Agents framework represents a significant advancement in understanding social perception and group dynamics. By leveraging LLMs and probabilistic modeling, the framework provides a nuanced and dynamic representation of complex social interactions. However, its reliance on questionnaire data and observed social networks limits its generalizability. To fully leverage the framework's potential, researchers and educators must consider its limitations and work to develop more comprehensive and representative datasets. Additionally, the framework's computational resource intensity may require innovative solutions to make it more accessible in resource-constrained settings.
Recommendations
- ✓ Recommendation 1: Further research should focus on developing more comprehensive and representative datasets to improve the framework's generalizability.
- ✓ Recommendation 2: Researchers should explore innovative solutions to address the framework's computational resource intensity and make it more accessible in resource-constrained settings.
Sources
Original: arXiv - cs.AI
