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

Researchers present a curiosity-driven AI method for discovering emergent behaviors in Flow-Lenia, a continuous cellular automaton with mass conservation. Using Intrinsically Motivated Goal Exploration Processes (IMGEP), the study reveals ecosystem-level dynamics and self-organized patterns that resemble biological phenomena, demonstrating that AI-driven diversity search can efficiently scaffold complex systems research.

This research represents a methodological advancement in using AI to systematically explore complex, parameterizable systems rather than relying solely on random search or manual investigation. The team adapted IMGEP—an approach that prioritizes exploring uncertain goal-spaces—to study Flow-Lenia across multiple scales, revealing macro-scale organization patterns absent at base scales. The significance lies not in Flow-Lenia itself, but in demonstrating a generalizable framework for discovering emergent behaviors in bottom-up systems.

The approach combines simulation-wide metrics like evolutionary activity and compression ratio to guide exploration, then validates findings through scaling studies across six spatial scales and seven time horizons. This iterative loop—diversity search, inspection, redesign—mirrors scientific methodology itself, with the interactive exploration tool maintaining human scientist involvement throughout. The authors explicitly position this work beyond Lenia, suggesting applicability to any parameterizable complex system.

For AI and computational science communities, this demonstrates practical value of intrinsic motivation methods in scientific discovery. Rather than exhaustively testing parameters, IMGEP efficiently illuminates metric space and uncovers qualitatively interesting behaviors. The finding that macro-scale organization lacks base-scale analogues underscores emergent complexity's genuine novelty. This could influence how researchers approach systems biology, artificial life, and materials science simulations where exhaustive parameter exploration remains computationally infeasible. The work validates AI as a scientific collaborator capable of generating hypotheses worth investigating, rather than merely executing predetermined experiments.

• →IMGEP-based diversity search revealed significantly more of Flow-Lenia's metric space than random exploration, demonstrating efficiency gains in complex systems research.
• →Macro-scale organization patterns emerged with no analogues at base scales, indicating genuine emergent complexity requiring multi-scale analysis.
• →The iterative human-in-the-loop approach combines automated exploration with scientist inspection, enabling principled experimental scaffolding.
• →Findings suggest the methodology extends beyond Lenia to other parameterizable complex systems where bottom-up collective behavior matters.
• →Curiosity-driven AI methods can reduce computational costs for large-scale diversity searches, enabling more ambitious subsequent experiments.