Exploring Flow-Lenia Universes with a Curiosity-driven AI Scientist: Discovering Diverse Ecosystem Dynamics

TL;DR

The paper tackles understanding emergent ecosystem dynamics in mass-conserving, parameter-localized cellular automata by applying Intrinsically Motivated Goal Exploration Processes (IMGEPs) to Flow Lenia. It defines system-wide metrics—$A$ (non-neutral evolutionary activity), compression-based complexity, and multi-scale matter distribution—and demonstrates that IMGEP-driven exploration illuminates far more diverse dynamical regimes than random search. The approach yields qualitative ecosystem phenomena such as feeding, colonies, and allopatric-like evolution, and is paired with an interactive human–AI exploration tool to facilitate analysis. While demonstrated on Flow Lenia, the framework is presented as a general strategy for discovering open-ended, collective dynamics in other parameterizable complex systems.

Abstract

We present a method for the automated discovery of system-level dynamics in Flow-Lenia--a continuous cellular automaton (CA) with mass conservation and parameter localization-using a curiosity--driven AI scientist. This method aims to uncover processes leading to self-organization of evolutionary and ecosystemic dynamics in CAs. We build on previous work which uses diversity search algorithms in Lenia to find self-organized individual patterns, and extend it to large environments that support distinct interacting patterns. We adapt Intrinsically Motivated Goal Exploration Processes (IMGEPs) to drive exploration of diverse Flow-Lenia environments using simulation-wide metrics, such as evolutionary activity, compression-based complexity, and multi-scale entropy. We test our method in two experiments, showcasing its ability to illuminate significantly more diverse dynamics compared to random search. We show qualitative results illustrating how ecosystemic simulations enable self-organization of complex collective behaviors not captured by previous individual pattern search and analysis. 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.

Figures (10)

• Figure 1: Snapshot of an advanced state of Flow Lenia. Each color represents a different set of localized parameters, determining how matter behaves. Regions with brighter colors represent higher concentrations of mass. In this paper, we present an AI Scientist method, based on IMGEPs, to systematically discover novel dynamics in large-scale Flow Lenia simulations.
• Figure 2: IMGEP approach for exploring a diversity of self-organization in Flow Lenia. The algorithm samples new goals in a system-wide metric space, selects parameters from previous similar simulations, mutates them, runs simulations, and computes metrics to guide further exploration. This process discovers a variety of dynamics, including dense matter clusters akin to colonies, behavior resembling allopatric speciation, feeding, and more.
• Figure 3: Examples of environments with multiple parameters co-existing and interacting in Flow Lenia. Gray bars represent walls blocking the flow of matter, creating environmental constraints that influence the system's dynamics.
• Figure 4: Goal space coverage comparison between IMGEP and random exploration, showing that IMGEP consistently illuminates more of the goal space than random search.
• Figure 5: Evolution of exploration metrics over time. The consistent divergence between IMGEP (blue) and random search (orange) indicates IMGEP's sustained ability to discover new evolutionary regimes.