May 19, 2026
Ctrl+Z for digital life
Growing Neural Cellular Automata
AI grows pixel creatures that heal themselves — and the comments are split between awe and huh
TLDR: This article shows an AI system made of simple digital cells that can grow shapes and sometimes repair damage on their own. Commenters were torn between amazed nostalgia for Distill, confusion about what the demo is actually useful for, and big-brain jokes about biology being elite data compression.
A blast from the past just crash-landed into the comment section: an interactive 2020 Distill article showing tiny AI-driven cells growing patterns, holding their shape, and even repairing damage when you erase part of them. On paper, it’s about teaching simple digital “cells” to organize themselves into a picture and recover when it gets wrecked. In practice, the crowd immediately turned this into a feelings-fest, with one commenter practically having a nostalgia flashback over Distill itself, saying the site’s gorgeous explainers shaped their entire early understanding of machine learning. For some readers, this wasn’t just a paper — it was a reunion tour.
But not everyone was ready to join the standing ovation. The biggest pushback was a very relatable “Okay… but what is this actually for?” One skeptical commenter flat-out said it “looks cool” but they still don’t understand the usefulness, which is basically the eternal internet verdict on half of cutting-edge AI demos. Another commenter tried to pull the discussion back to practical territory, pointing to newer research suggesting neural cellular automata could help create training material efficiently. Meanwhile, a different hot take went full galaxy brain: if living things can build themselves from compact instructions, then biology is basically doing absurdly good data compression.
So the mood? Equal parts science wonder, old-school Distill fan club, and mild “please explain this in normal-person words” frustration. The demo’s healing patterns are undeniably mesmerizing — but the real action is the comment-section split between inspired minds, confused onlookers, and people casually comparing life itself to a zip file.
Key Points
- •The article introduces a differentiable model of morphogenesis using neural cellular automata.
- •It presents interactive demonstrations where users can erase parts of patterns and observe whether the system regenerates them.
- •The article connects the model to biological morphogenesis, describing how multicellular organisms self-organize from a single cell.
- •It emphasizes that biological development is robust to perturbations and uses regeneration in organisms as motivation for the model.
- •The article compares growing, persistent, and regenerating training regimes, showing how different objectives affect stability and repair behavior.