Neural networks replicate themselves on pixel grids showing evolving diversity

Self-Replicating Neural Cellular Automata: Quantifying Emergent Phenotypic and Genotypic Diversity in an OpenEnded Substrate

Machine LearningNeural and Evolutionary Computing

Summary

The paper studies a digital system where each pixel on a grid contains a tiny neural network that represents a cell. These cells survive only by copying themselves to neighboring pixels with slight changes, making their state entirely determined by their internal 'genetic code' (network weights). Starting from a few initial cells, complex patterns and different cell types emerge and compete like species in an ecosystem. The authors introduce new ways to measure how diverse these cells are both in behavior and genetics, finding that increasing one often reduces the other. Their work helps understand complexity and diversity in self-replicating digital life.

What this means in practice

  • For artificial life developers: Create self-replicating digital organisms with measurable genetic and behavioral diversity across spatial grids to study evolution and adaptation.
  • For complex systems modelers: Analyze emergent species diversity and lineage structures in evolving systems by applying new diversity metrics at multiple scales.

Authors

Sanyam Jain, Felix Simon Reimers, Stefano Nichele

Abstract

We study an in-silico substrate in which every pixel of a two-channel cellular-automata grid carries a tiny neural network (an agent) that senses its Moore neighborhood. A cell persists only by self-replication: a living neighbor is cloned and its weights are mutated by a uniform perturbation, so that phenotype (cell state) is driven entirely by genotype (network weights). From a handful of seeded founders the system grows into a spatially organized ecosystem of coexisting, competing and dominating species. Our main contribution is a battery of coarse-grained diversity metrics that make such growth measurable at two scales: four phenotypic tools based on cellular-type frequency, entropy and cell variance, and two genotypic tools that colour each agent by a hash of its full weight vector versus a sparse random-weight probe. Across a five-fold sweep of 1680 small runs and 24 long (1000-generation, 200 x 200) runs, the substrate is persistent and self-maintaining in 20 of the 24 long configurations and exposes a clear phenotype-genotype diversity trade-off: raising phenotypic diversity collapses genotypic diversity and vice versa. Full-genome hash colouring further reveals lineage structure that a random-weight probe systematically misses. Code, data and animations are released as supplementary material.