Self-Assembly Gets Automated in Reverse of 'Game of Life' (www.quantamagazine.org)

Google Research scientist Alexander Mordvintsev has advanced the field of neural cellular automata (NCAs) by automating the reverse engineering of self-assembly rules from desired patterns, such as regenerating pixelated monarch butterflies that spontaneously regrow damaged wings. Building on traditional cellular automata like Conway's Game of Life, Mordvintsev's key innovation is using neural networks to learn local interaction rules that cause simple “cells” to self-organize into complex, robust structures. This approach enables "complexity engineering," where instead of designing a complex shape directly, one designs the behavior of fundamental units (cells) that autonomously assemble into the target form. Technically, Mordvintsev replaced the binary on/off states of classical automata with continuous cell states and introduced random asynchronous updates and hidden variables to simulate more lifelike, organic development. Neural networks trained through backpropagation or genetic algorithms discover the rules to grow and regenerate specified patterns, demonstrating remarkable self-healing properties. This paradigm promises wide implications: it models biological morphogenesis more realistically, may inspire regenerative medicine approaches to tissue growth, and offers a new distributed computing architecture without centralized control. By mimicking how real cells collectively build and repair organisms, NCAs open avenues for resilient, adaptive AI systems and could accelerate understanding of evolution and natural complexity.