Emergent Mind / Labs

A field guide to digital animals

Digital Creatures

Every creature here grew from a single cell by a rule small enough to fit on an index card. Nobody drew them. They were evolved, one random mutation at a time, to live as long as they could and then die.

Stephen Wolfram shared a sheet of specimens like these on X. Below is how they're made, step by step, with a lab where you can evolve your own.

Evolve a new creatureSee how it works

Eighteen specimens evolved for this page. Select one to inspect it.

How they're made

Five ideas, one cell at a time

01 A row of cells and a rule

A creature lives on a single row of cells. Each cell holds one of a few colors, and one color means empty (black here). At the start, every cell is empty except one.

To make the next row, each cell looks at three cells in the row above: the one directly above it and that cell's two neighbors. The rule is a lookup table that says, for every possible trio of colors, what color the new cell becomes.

That table is the creature's genome. Here is the complete rule for the creature in the figures below. It uses four colors, and each little icon is one entry: three cells on top, the result below.

39 entries, each standing for a trio and its mirror image. The all-empty trio always stays empty and isn't shown.

02 Apply it everywhere, row after row

Every cell updates at once, using the same rule. Time runs down the page: the top row is the single starting cell, and each row below is one step later.

Watch the outlined trio slide across the row. The highlighted entry in the table is the one being looked up, and its result fills the outlined cell.

03 Mirror-image rules make mirror-image bodies

The rules are symmetric: a trio and its mirror image, the same three colors in reverse order, always give the same result. Since the starting cell is symmetric too, the left half of a creature is always the mirror of its right half. That is where the bilateral, animal-like look comes from.

Symmetry also shrinks the genome. Four colors make 4 × 4 × 4 = 64 trios, but only 40 of them can be set independently.

On the right, the same creature with one entry changed for one trio . The body comes out lopsided.

Symmetric rule
One entry broken

04 Fitness is a long life that ends

A creature's lifetime is the number of rows it lasts before every cell is empty again. Most rules fail one of two ways: they die almost at once, or they never die, spreading forever or settling into a pattern that repeats forever.

The interesting ones live a long time and then stop. That finite lifetime is the creature's fitness, and immortal creatures score zero.

In general there is no way to know in advance whether a rule's pattern will ever die out. It is a version of the halting problem. So in practice, anything still alive after 1,000 steps, or more than 300 cells from where it started, counts as immortal.

Lives 1 step The null rule turns everything empty.
Never dies Grows forever.
Lives steps Then every cell goes empty.

05 Evolution: mutate, test, keep

Start from the null rule, whose creature lives exactly one step. Then repeat:

  1. Mutate. Pick one entry of the rule at random and change its result to a different random color (its mirror image changes with it).
  2. Grow. Run the mutant from a single cell and measure its lifetime.
  3. Select. Keep the mutant if it still dies and lives at least as long as the current creature. Otherwise, throw it away.

That's the whole algorithm. There is no designer and no crossover, just one random change at a time and a single number to judge it by.

Keeping ties matters. Most mutations that survive are neutral: they don't change the lifetime, but they let the genome drift until some later mutation finds a longer life. Without that drift, evolution gets stuck almost immediately.

The lab

Evolve your own

Watch evolution build a creature from nothing. It starts as a single dot that lives for one step. Then your browser repeats one simple loop, thousands of times a second:

  1. 1Change one entry in the creature's rule, at random. This is a mutation.
  2. 2Grow the changed creature from one cell and count how many steps it lives.
  3. 3Keep the change if the creature still dies and lives at least as long. Undo it if it dies sooner or never dies.

Most changes get undone. Every so often one lets the creature live longer, and the picture jumps to something bigger and stranger. When 10,000 changes in a row fail to help, it stops. Then press New creature: it starts over from a single dot with a different random history, and you get a creature nobody has seen before.

Changes tried
0
Kept
0
Steps it lives
1

How long it lives, as changes pile up

Flat stretches are changes that didn't help. Each jump is a lucky one. Changes tried run left to right on a log scale.

The change it just tried

Link to this creature N new creature M one change

Fossil record: the creature each time it started living longer. Select one to inspect it.

Notes

Where this comes from

The model

This is the minimal model of biological evolution Stephen Wolfram introduced in Why Does Biological Evolution Work? (2024) and developed in Foundations of Biological Evolution: More Results & More Surprises (2024): one-dimensional cellular automata with nearest-neighbor, mirror-symmetric rules, grown from a single cell, evolved from the null rule by single point mutations, with lifetime as fitness. He used the same kind of evolved creature as a model organism in Towards a Computational Formalization for Foundations of Medicine (2025).

His post doesn't give the exact settings behind the new specimens, such as the number of colors or the step limit. The settings on this page are ours, chosen to produce creatures with the same anatomy.

The specimens on this page

We ran 1,160 evolutions with the code on this page, each 15,000 mutations long, using four to eight colors. Then we picked eighteen for their looks, the way a naturalist picks specimens for a drawer. Evolution itself only ever selected for a long life. The eye for shape came afterward, and the Latin names are ours too.

Their colors were chosen to echo Wolfram's sheet. Color is just a label, so recoloring a rule doesn't change the creature. Open any specimen and choose Replay its evolution to watch the exact mutations that produced it.

Why it works at all

Random point mutations plus a single number to climb shouldn't obviously produce intricate structure. What evolution finds instead is whatever happens to work: long-lived creatures built from gliders, nested triangles and collisions, with no simple story for why they last as long as they do.

Wolfram's argument is that this is what biology looks like too: computationally irreducible systems, sampled by selection for a coarse objective, produce mechanisms that are effective without being explainable.

Built by Emergent Mind Labs. Inspired by Stephen Wolfram's specimens.

Specimen

Colors
Lifetime
Cells wide

Its rule