Two kinds of life evolving in one dish. Grazers eat the jelly's food and hold on to family to form bodies. Hunters, the hollow rings, eat grazers. Bodies are hard to eat and hit back. Nothing here is scripted.
step · season · generations deep
running on your processor
Tap a cell to follow it · tap empty jelly to feed · scroll to zoom
What just happened
Following
How this world works
Designed by us the world, the senses, the possible actions, the costs.
Left to evolution every brain weight, brain size, and whether to stick, bite, share, call or remember.
This run so far
Family tree
Each bar is a family line. A new line starts when a daughter's colour has drifted far enough from her family's, and ends when its last member dies. Thicker bars had more cells at their peak; hunter lines are outlined. Bars share the step scale above.
Why bodies stayed small, and what changed
Cells kept forming clumps of 3 to 5 and never grew into big organisms, so we measured why. Each line below is two test worlds (seeds 11 and 22), run headless and read at step 6,000.
| Rules | Cells alive | Living in a body | Cells per body | Largest body |
|---|---|---|---|---|
| Before: holds re-decided every step | 2,199 | 59% | 3.9 | 33 |
| Lasting attachments | 3,238 | 88% | 9.9 | 60 |
| … plus twice the food | 4,107 | 55% | 7.1 | 71 |
| … plus food flowing along attachments | 3,141 | 82% | 7.8 | 47 |
| … plus hunters can't tear cells out of big bodies | 3,420 | 82% | 11.7 | 64 |
| … with the family bonus switched off | 1,651 | 5% | 2.2 | 5 |
| Now: family bonus counts up to six neighbours | 4,298 | 92% | 21.7 | 125 |
| … the same bonus made 50% stronger | 7,207 | 97% | 37.7 | 332 |
In the old rules, 84–97% of broken holds were cells pulled apart while both still wanted to stick. A hold lasted only 4 to 6 steps. Now touching family form lasting attachments, and stickiness changes slowly. An attachment holds about 60 steps on average.
About 95% of deaths were starvation and under 1% were hunters. A body only feeds through its edge, but every cell inside it costs energy. Real organisms hit the same surface-against-volume problem. More food just made more cells, and sharing food along attachments didn't help.
Switch off the family bonus (family close by lowers a cell's running cost) and bodies vanish. It used to stop counting at three neighbours, so a cell deep inside a body gained nothing over one at the edge. Letting it count up to six, slightly weaker per neighbour, makes bodies about five times bigger than before.
With the bonus off, three times more hunters, and hunters that struggle to tear cells out of bodies, 10–27% of cells lived in groups of about three (3–9% without). Real algae did something similar: they evolved 8-cell colonies within about 100 generations of meeting a predator that could only swallow single cells (Boraas, Seale & Boxhorn, 1998).
A side effect: with nine in ten grazers held safely inside bodies, hunters now mostly starve. They die out roughly every 60 steps and a few newcomers drift in each time, so hunter brains never get the chance to evolve. We tried weaker stings, weaker shielding, more food per kill, no fighting back and cheaper upkeep for hunters. None of them reliably brought hunters back (two worlds each, 6,000 steps).
The honest summary: body size in this dish is mostly set by one rule we wrote, how much family neighbours lower a cell's costs. We picked a middle setting. Make it 50% stronger and bodies average 38 cells, but only because cells deep inside then live almost for free, which isn't how real bodies work. Predators and food on their own didn't build big organisms here, at least not within 6,000 steps.
30,000 steps later
The same two test worlds with today's rules, left running headless for 30,000 steps. In one, bodies kept growing: the average doubled from about 20 to about 40 cells, and the largest passed 300. In the other, they swung between 16 and 33 cells. In both, brains shrank, from 6 neurons to under 2 in one world and about 4 in the other. Every neuron costs energy, and a cell held inside a body seems to need fewer of them. Two worlds is a hint, not proof. (The earlier version, before lasting attachments, flattened out after a few thousand steps and stayed flat for 100,000.)
Do hunters drive an arms race? earlier version
The same four starting worlds with and without hunters, run headless for 10,000 steps and averaged over steps 4,000–10,000.
| Seed | Hunters (avg) | Grazers with / without | In a body with / without | Body size with / without | Grazer brain with / without |
|---|---|---|---|---|---|
| 11 | 782 | 1,970 / 3,117 | 74% / 75% | 6.7 / 6.8 | 5.8 / 6.1 |
| 22 | 789 | 1,699 / 2,690 | 72% / 74% | 6.1 / 6.0 | 5.6 / 6.0 |
| 33 | 81 | 2,833 / 3,053 | 80% / 67% | 7.8 / 9.8 | 5.1 / 5.5 |
| 44 | 101 | 2,762 / 3,020 | 74% / 72% | 6.6 / 6.5 | 5.0 / 6.1 |
| Mean | 438 | 2,316 / 2,970 | 75% / 72% | 6.8 / 7.3 | 5.4 / 5.9 |
Hunters survive on their own in every world, and grazers and hunters rise and fall in cycles, the classic predator–prey pattern (watch the chart in the Live panel). But within 10,000 steps hunters don't push grazers into bigger bodies or bigger brains. Grazer brains actually end up a little smaller when hunted. No arms race yet.
Left for 60,000 steps (two worlds), the hunters got much better at hunting: kills per step rose by about 65% (from 3.5 to 5.8 and from 5.1 to 8.5), and hunters came to outnumber grazers about two to one. Grazers didn't find an answer. Their bodies stayed small and both sides' brains shrank to save energy. So far the hunters are winning by getting cheaper and more efficient, not smarter.
When grazers fight back earlier version
Short run: four worlds with and without fighting back, 12,000 steps, averaged over steps 4,000–12,000.
| Hunters (avg) | Hunters died out and re-added | Grazers in a body | Body size | Grazer brain | Hunter brain | |
|---|---|---|---|---|---|---|
| Fight back on | 148 | 31 times | 62% | 5.1 | 5.7 | 6.0 |
| Fight back off | 462 | 3 times | 73% | 6.9 | 5.3 | 5.9 |
At first, fighting back swings things to the grazers. They often wipe the hunters out, then relax into smaller bodies because the danger is gone.
Long run: two worlds left for 40,000 steps tell a different story. By about step 8,000 the hunters adapt and stop dying out, and their brains grow and stay bigger.
| At step ~40,000 | Hunter brain | Grazer brain |
|---|---|---|
| Seed 22, fight back | 6.9 neurons | 5.3 neurons |
| Seed 22, no fight back | 4.0 neurons | 5.1 neurons |
| Seed 11, fight back | 6.7 neurons | 6.0 neurons |
| Seed 11, no fight back | 5.2 neurons | 4.9 neurons |
When prey can hit back, hunting takes more brain: hunters kept about 7 neurons instead of shrinking to 4 or 5. That is the first sign of an arms race in this dish. It's two worlds only, so treat it as a hint, not proof.
What each rule actually does earlier version
Each rule was switched off in the same four starting worlds (seeds 11, 22, 33, 44), run headless for 6,000 steps and averaged over steps 2,000–6,000. If a rule matters, switching it off should change the world.
| Switched off | Creatures | Living in a body | Cells per body | Largest body | Verdict |
|---|---|---|---|---|---|
| Nothing (full rules) | 2,714 | 74% | 7.7 | 63 | baseline |
| Holding on | 1,891 | 0% | — | — | matters: 30% less life without it |
| Daughters staying attached | 2,637 | 71% | 7.3 | 65 | no clear effect yet |
| Signals inside a body | 2,645 | 77% | 7.0 | 61 | no clear effect yet |
Holding on is the one rule evolution clearly uses: nearly every creature ends up sticky, and worlds where it works hold 43% more life. Daughters that stay attached don't make bodies bigger, because loose cells re-attach anyway. Cells don't seem to use the signals inside a body yet. In an earlier version, memory showed no clear effect either. Four seeds is a small sample; differences under about 5% can appear by chance.
Dig deeper
origin stands on decades of work in cellular automata, artificial life and real biology. These are the projects and studies behind its ideas, and good places to start.