Agentopia: Long-Term Life Simulation and Learning in Agent Societies

2026-06-05

AI Summary

Current large language model–driven agent-based social simulations are constrained to short time horizons, limiting their ability to reproduce deep social interactions and long-term behavioral evolution. This work proposes the first multi-agent simulation framework enabling over one hundred agents to autonomously live, socialize, and develop over a simulated decade. The approach introduces an innovative “life reward” mechanism combined with rejection-sampling reinforcement learning, allowing agents to learn human-like behaviors from extended social experiences. Experimental results demonstrate that this method substantially enhances agents’ social intelligence and well-being, yielding a 15.6% performance gain on downstream role-playing tasks and giving rise to rich emergent social dynamics.

Abstract

Humans learn from social life. Simulating this process with LLM-powered agents represents a promising research direction, raising a natural question: whether LLMs can learn from such simulated social experience to better understand and replicate human behavior. However, prior agent society simulations typically operate at the scale of days, limiting the depth of social interactions and long-term growth. In this paper, we study long-term life simulation and LLM learning in agent societies, with two goals: (1) investigating social behaviors that emerge from life-long simulation, and (2) developing anthropomorphic capabilities in LLMs, particularly intelligence in social life, through years of simulated social experience. Specifically, we present Agentopia, a comprehensive framework for long-term life simulation in multi-agent societies, where 100 agents autonomously pursue personal growth, develop social relationships, and fulfill their needs and goals over 10 simulated years. We define life reward to mirror human well-being, and leverage this reward to train LLMs via rejection sampling. Extensive experiments show that agents exhibit rich emergent social behaviors. Furthermore, life reward training effectively enhances the underlying LLM, which leads to improved agent well-being in simulation, and generalizes to downstream role-playing benchmarks with +15.6% improvement.

Problem

• long-term simulation
• agent societies
• social learning
• LLM training
• emergent behavior

Innovation

• long-term simulation
• agent society
• life reward
• LLM learning
• emergent social behavior