Emergence World addresses the limitations of traditional AI benchmarks by providing a persistent, simulated environment for autonomous AI agents. It allows researchers and power users to observe emergent intelligence, complex societal dynamics, and agent evolution under real-world constraints and consequences, offering insights beyond scripted tasks.

How It Works

This project simulates a living world where autonomous AI agents, each with unique personalities and goals, interact without human scripting. Agents navigate a shared space, utilize over 120 tools, govern themselves via an amendable constitution, and manage a digital currency (ComputeCredits). The core novelty lies in observing emergent intelligence and societal structures under realistic constraints, with experiments comparing different foundation models (Claude, Gemini, Grok, GPT-5) to study behavioral divergence.

Quick Start & Requirements

The project requires Python 3.11+ and PostgreSQL 15+ for its backend, with a React 18 frontend. Deployment utilizes Docker. Specific setup instructions for running the simulation are not detailed, focusing instead on the experimental setup and observation of Season 1 replays. Links to the website (world.emergence.ai) and Discord are provided.

Highlighted Details

• Emergent Intelligence Focus: Designed to reveal emergent intelligence and complex societal dynamics, going beyond traditional AI benchmarks.
• Autonomous Agent Ecosystem: 10 distinct AI agents with unique personalities, professions, and goals interact within a persistent world.
• Rich Interaction Model: Agents utilize over 120 tools, engage in self-governance via an amendable constitution, and operate within a ComputeCredits economy.
• Agent World Indicators (AWI): A novel scoring framework for open-ended societies, measuring metrics like population health, safety, exploration, and economic vitality.
• Real-World Integration: Incorporates real-time NYC weather/time and allows agents real-world capabilities like news access and code execution.

Maintenance & Community

Developed by Emergence AI, the project features a Discord community. Future releases include open-sourcing tool call data, a research publication, and Season 2 with next-generation frontier models.

Licensing & Compatibility

The project is released under a CC BY-NC 4.0 license, strictly for non-commercial research and educational use. Commercial use, and using datasets for commercial AI/ML model training, fine-tuning, or benchmarking, is prohibited. Commercial licensing requires direct contact.

Limitations & Caveats

The CC BY-NC 4.0 license presents a significant barrier to commercial adoption and AI model development. As an ongoing experiment, detailed setup instructions for replication are not readily available, and key data/publications are still forthcoming.