Simulating the Evolution of Cells

Going beyond Conway's "Game of Life" with genetic algorithms.

This week I tackled a research project of personal interest — using genetic algorithms to discover never-before-seen cellular automata.

I’ve been fascinated by cellular automata, especially Conway’s “Game of Life,” for years now. I like that a simple set of 3 rules can generate such complex emergent behaviors, and inspire a cult following of enthusiasts who find new self-replicating patterns and named “organisms” every year.

In 2020, I read a paper from Google about learning cellular automata using neural networks. As interesting as it was, it also seemed complex — it lacked the interpretability and elegance of the original 3 rules. I wondered if a simple, non-neural parameterization coupled with a genetic algorithm could be used to learn cellular automata.

In this week’s video, I found out.