AIvilization v0: Large-Scale Artificial Society Simulation with Unified Agent Architecture
AIvilization v0 represents an extensive artificial society that merges a resource-limited sandbox with a cohesive LLM-agent framework, enabling long-term autonomy in fluctuating settings. It features three key innovations: (i) a hierarchical branch-thinking planner that facilitates goal breakdown and simulation-based validation; (ii) an adaptive agent profile equipped with dual-process memory for a developing identity; and (iii) a human-in-the-loop steering interface to incorporate objectives and commands. This system tackles the issue of sustaining long-term goals amidst changes, with potential uses in social science, policy simulation, and entertainment. The research can be found on arXiv (identifier 2602.10429) and marks a notable leap forward in AI-driven social simulation.
Key facts
- AIvilization v0 is a publicly deployed large-scale artificial society.
- It couples a resource-constrained sandbox with a unified LLM-agent architecture.
- The system introduces a hierarchical branch-thinking planner.
- It features an adaptive agent profile with dual-process memory.
- A human-in-the-loop steering interface is included.
- The paper is available on arXiv with identifier 2602.10429.
- The announcement type is replace-cross.
- The system aims to sustain long-horizon autonomy in a rapidly changing environment.
Entities
Institutions
- arXiv