Overview

We present a method for automated discovery of system-level dynamics in Flow-Lenia—a continuous cellular automaton with mass conservation and parameter localization—using an AI scientist system based on Intrinsically Motivated Goal Exploration Processes (IMGEP)A family of diversity search algorithms that autonomously set goals and explore parameter spaces to discover diverse behaviors. Exploring full ecosystemic simulations is crucial for understanding the emergence of complex collective behaviors that individual pattern analysis cannot capture, such as competition, cooperation, and environmental adaptation. While previous applications of IMGEPs, a family of diversity search algorithms, focused on automated exploration of self-organized individual patterns or behavior, we extend this methodology to explore system-level dynamics. Our implementation evaluates systems using simulation-wide metrics (e.g. evolutionary activityA measure of how much genetic change occurs in a system over time, indicating the potential for open-ended evolution, compression-based complexityA complexity measure based on how much data can be compressed, reflecting the amount of structure and patterns in the system, and multi-scale entropyAn entropy measure that captures the predictability of a system across different temporal and spatial scales) to guide exploration toward diverse evolutionary regimes. Results show IMGEPs discover significantly more diverse dynamics than random search, revealing complex self-organized ecological interactions in Flow-Lenia. We complement automated discovery with an interactive exploration tool, creating an effective human-AI collaborative workflow for scientific investigation. Though demonstrated specifically with Flow-Lenia, this methodology provides a framework potentially applicable to other parameterizable complex systems where understanding emergent collective properties is of interest.

Key Contributions

  • Extension of IMGEP as a systematic discovery tool for ecosystem-level emergent properties in complex systems, applied here to uncover diverse evolutionary dynamics in Flow-Lenia
  • Multi-metric approach to evaluating evolutionary dynamics, combining evolutionary activity, complexity measures, and multi-scale entropy analysis
  • Discovery of diverse evolutionary phenomena suggesting ecological interactions

Interactive Visualization Tool

Explore Flow-Lenia's evolutionary dynamics with our interactive tool. Examine parameter configurations, emergent behaviors, and evolutionary metrics discovered by IMGEP.

Launch Visualization Tool

Video Demonstrations

These videos showcase the some interesting dynamics discovered through our experiments:

A mostly static arrangement featuring large uniform chunks of matter. A set of mutations propagate near the end, results in a formation of membranes.

Different types of matter form into a large, complex pattern.

A diverse array of patterns, some made up of tightly bound chunks of matter that resemble colonies.

Various kinds of vibrating masses, some containing a more static nuclei, swimming in a sea of simple mass chunks

Chaotic dynamics with rapid matter movement and constant mutations.

Larger pattern "ingests" multiple smaller chunks, reminiscent of feeding.

Cohesive pattern moving diagonally.

Slightly dissipative pattern attracted to the corner.

Smaller patterns efficiently manouvering through narrow passages.

Initial matter propagates into neighbouring rooms, afterwhich localized mutations result in speciation.

Movement behaviour of a large pattern changes multiple times after mutations. From directed motion, to stillness, to chaotic motion.

Initial matter splits into two very fast moving patterns