Beyond Self-Interest: Simulating Human-like Social Behavior via Modeling Social Value Orientation in Agent Motivation

Jingzhe Lin, Ceyao Zhang, Yaodong Yang, Yizhou Wang, Song-Chun Zhu, Fangwei Zhong

Keywords: Social value orientation, Agent-based social simulation, Cooperation and competition dynamics

TL;DR: Simulating Human-like Social Behavior via Modeling Personal Desire and Social Value Orientation

Abstract: Large language models (LLMs) have recently shown promise in simulating complex social behavior through autonomous agents. However, existing systems often overlook how agents’ internal psychological structures evolve through prolonged interaction. This paper presents a novel method, ASVO (Autonomous Social Value-Oriented agents), which integrates Social Value Orientation (SVO) with value-driven LLM agents. Our approach enables generalizable, high-fidelity generation of social behaviors by combining structured motivational profiles with dynamic SVO adaptation. Each agent is equipped with a structured set of social desires. The intensities fluctuate over time and are updated using an LLM-based reflective mechanism conditioned on social context. Agents continuously adapt their expectations, monitor personal and others' satisfaction, and update their SVO accordingly. We conduct systematic simulations across varying social environments to examine the emergence of cooperation, competition, and personality drift. Results show that our agents exhibit more human-like behavioral adaptations, with SVO shifts aligning with observed social dynamics. Our findings suggest that integrating structured desire systems and adaptive SVO drift enables more realistic and interpretable multi-agent social simulations.

Area: Modelling and Simluation of Societies (SIM)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 1720