Scientists Built Digital Cells That Copy Themselves and Evolve

Imagine a giant grid of pixels, like an old arcade screen. In this experiment, every single pixel holds a tiny AI brain of its own. Each one can only sense the eight pixels touching it — the way a cell in your body senses its neighbors. And it can only survive by doing one thing: copying a living neighbor next to it, with a small random change baked in. That's it. No designer, no goal. Just copy, mutate, repeat.

The result, published on the research site arXiv, is that life takes off. Starting from just a handful of "founder" cells, the grid grows into a full ecosystem — different kinds of digital creatures competing, coexisting, and sometimes one type taking over. The authors ran 1,680 short experiments and 24 long ones (each 1,000 generations on a 200-by-200 grid). In 20 of the 24 long runs, the world kept itself alive without help.

The paper's real contribution is a new set of measuring sticks. The researchers borrowed a classic idea from biology: the difference between body and blueprint. In their world, the "body" is what a cell looks like on the grid, and the "blueprint" is the exact set of numbers in its little brain. They colored each cell by its full blueprint versus a random sample of that blueprint, then measured how diverse each kind of variety got.

They found a clear trade-off: when you push up the diversity of appearances, the diversity of inner blueprints collapses — and the reverse. They also found that coloring cells by a full blueprint reveals family trees and lineages that a random sampling approach completely misses. Notably, this is a simulation, not real biology, and it doesn't tell us anything about actual living cells. The code, data, and animations were released publicly, so other researchers can build on it.