Experiments
Everything this website explains comes from simulations: thousands of games where only one thing changes. Here are the findings, one by one, with their numbers and a button to reproduce them in the laboratory.
The numbers come from 192 × 192 boards and thousands of generations. In the browser, raise the speed and be patient: evolution takes time. Every game is different, so you will see similar values, not identical ones.
- 29colour families live side by side in the Evolutionary version when colour is inherited with small changes (without that, 3.7)
- +0.68preference for partners of their own colour: the highest the Brains learned on their own (they started at +0.02)
- 10 of 11head-to-head games are won by the small brain
- 93 %of the Evolutionary version's “corners” exist because their nest lacks food
First time here? «Start here» explains everything from scratch, with examples. And words with a dotted underline have their definition: tap them.
1 · Evolutionary
Do the permissive or the strict ones win?
The question
If every cell inherits how many neighbours it can live with, which genomes win? Does it depend on the environment?
How it was tested
Evolutionary with food (growing back 0.3 per generation); average of 3 games of 3000 generations for each cost value (extra food per permission).
What came out
| Cost per permission | Average permissions | Most common genomes |
|---|---|---|
| 0.1 | 3.0 to 3.4 | lives with 2 to 5 (or 1 to 5) |
| 0.25 | 2.2 | lives with 2 to 5 or 2 to 4 |
| 0.5 | 1.4 to 1.5 | lives with 2 to 4 |
Watch it here
gen. 0Average genome permissions
- Cost 0.1: –
- Cost 0.5: –
What it means
Being permissive has advantages (surviving anything) and a cost (eating more). With a low cost the permissive ones win; with a high cost, the strict ones. The environment decides.
2 · Evolutionary
Colour as a surname
The question
In the 2023 Mixing version everything ended up grey. What if colour were inherited with small changes?
How it was tested
Evolutionary with different amounts of colour drift, games of 3000 generations.
What came out
| Colour drift | Colour families |
|---|---|
| 0 (like Mixing) | 3.7 on average |
| 4 | 8 to 21 |
| 8 (default) | 29 on average |
| 10 | 25 to 37 |
Watch it here
gen. 0Colour families
- No drift: –
- Drift 10: –
What it means
With a little variation when inheriting, families drift apart and show up as moving patches: genetic drift, made visible.
3 · Brains
They learn to choose partners of their own colour
The question
If a partner that is too different may give a daughter that does not make it, do brains learn to avoid them? They only see colour, not genes.
How it was tested
Brains (small, no holding on or learning), 2 games of 6000 generations with incompatibility.
What came out
| Generation | Prefers similar | Wants to breed | Failed daughters |
|---|---|---|---|
| 100 | +0.02 / +0.11 | 0.91 / 0.82 | 13 % / 12 % |
| 2000 | +0.21 / +0.15 | 0.88 / 0.89 | 15 % / 15 % |
| 6000 | +0.52 / +0.68 | 0.50 / 0.22 | 9 % / 10 % |
Watch it here
gen. 0Preference for partners of its colour (−1 to 1)
- With incompatibility: –
- Without incompatibility: –
What it means
Without incompatibility, the colour preference stays at +0.08. With it, they learn to use colour as a clue to compatibility, become choosy and fail less. In biology this is reinforcement: one of the ways new species are born.
Reproduce it: With incompatibility Reproduce it: Without incompatibility
4 · Brains
Holding on only when there is food
The question
If they can stay alive although the rule says they die, paying food, what do they do?
How it was tested
Small brains that decide whether to hold on, 2 games of 6000 generations per cost.
What came out
| Cost of holding on | Hold on when the rule says they die | …more the more food they have |
|---|---|---|
| 2 | +0.98 / +0.99 (always) | ≈ 0 |
| 4 | +0.85 / +0.94 (always) | ≈ 0 |
| 8 | ≈ 0 | +0.96 / +0.85 |
Watch it here
gen. 0How much more they hold on the more food they have
- Cost 2: –
- Cost 8: –
What it means
If holding on is cheap, they always do it. If it is expensive, a conditional strategy appears that nobody programmed: they hold on only if they have food.
5 · Brains
Learning helps… and is then dropped
The question
If they can learn during their life (without inheriting what they learn), does it pay off? Is the ability to learn inherited?
How it was tested
Small brains starting with a fair amount of learning (gene 0.5), 6 games of 4000 generations, against 6 without learning (where the gene is useless).
What came out
| Variant | Failed daughters (gen. 200) | “How much it learns” gene (gen. 4000) |
|---|---|---|
| Learning | 6–9 % | falls in all 6: 0.08–0.14 |
| Not learning | 13–16 % | stays: 0.34–0.61 |
Watch it here
gen. 0“How much it learns” gene
- Learning: –
- Not learning: –
What it means
Learning halves the failures at first, but evolution gradually drops it once the preference is in the genes. It is the pattern of the Baldwin effect.
6 · Brains
Does thinking more help survival?
The question
Does a 14-neuron brain, with memory and more to look at, survive better than a 7-neuron one?
How it was tested
4 games of 6000 generations with each brain, on separate boards.
What came out
| Measure | Small (7 neurons) | Big (14 neurons) |
|---|---|---|
| Board occupied | 19.0 % | 18.1 % |
| Starve to death each generation | 4.4 % | 5.5 % |
| Prefer a partner of their colour | +0.22 to +0.29 (avg. 0.26) | +0.23 to +0.69 (avg. 0.50) |
| Hold on according to their own food | +0.81 to +0.95 | +0.61 to +0.82 |
| Hold on according to food in their area | cannot see it | +0.27 to +0.59 in 3 of 4 games |
Watch it here
gen. 0Preference for partners of its colour (−1 to 1)
- Small brain: –
- Big brain: –
What it means
It does not survive more, but it chooses partners with twice the discernment and discovers a strategy the small one cannot: checking whether its area has food.
7 · Brains
Head to head: big brains against small ones
The question
If big and small brains live on the same board and eat the same food, who wins?
How it was tested
Half of each size, mixed at random. Mothers with different brain sizes cannot have daughters (they are different species). 3 games on 192 × 192 and 8 on 128 × 128.
What came out
| Outcome | Games | How |
|---|---|---|
| The small one wins | 10 of 11 | the big one disappears between generation 1000 and 4000 |
| The big one wins | 1 of 11 | — |
| They coexist | 0 of 11 | — |
Watch it here
gen. 0Share of the population
- Big brains: –
- Small brains: –
What it means
Only one size is ever left: two species living on exactly the same thing cannot coexist (in ecology, competitive exclusion). And the small one almost always wins, even though in 5 of 8 games the big ones were ahead at first. Thinking more does not pay: the big brain starves a little more, and with the same food that is decisive.
8 · All
Which patterns appear on their own?
The question
Without drawing anything, which still lifes, oscillators and spaceships appear in each version? Are they Conway's or are there new shapes?
How it was tested
12 games per version from a random board (128 × 128, 5000 generations). Every 100 generations isolated objects are followed for up to 30 generations; if they recover their shape (only the shape), they are patterns. Each shape is then tested on its own with Conway's rules.
What came out
| Version | Objects that are patterns | Conway's | Do not exist in Conway |
|---|---|---|---|
| Conway | 63.0 % | 100.0 % | 0.0 % |
| Borrego | 64.7 % | 96.8 % | 0.0 % |
| Mixing | 61.1 % | 100.0 % | 0.0 % |
| Evolutionary | 0.2 % | 14.3 % | 85.7 % |
| Brains | 0.8 % | 9.8 % | 90.2 % |
Watch it here
gen. 0Cells that were not there 2 generations ago
- Conway: –
- Brains: –
What it means
Borrego and Mixing are still, in terms of patterns, Conway's game. In the evolving versions almost nothing repeats, and the little that is stable is new: above all the “corner”, three cells in an L that in Conway would become a block and here hold because in 93–94 % of cases the ground where the fourth would be born is exhausted.
9 · Brains
How much brain is worth it?
The question
If every active neuron costs food and which neurons are active is inherited (and mutates), does the brain grow or shrink?
How it was tested
Big brain starting with the 7 neurons of a small one switched on; each neuron can switch on or off at birth (1 % chance). 3 games of 6000 generations per cost.
What came out
| What each neuron eats | Active neurons (gen. 6000) | Board occupied |
|---|---|---|
| 0 (free) | 6.7 to 7.0 | 14–16 % |
| 0.05 | 6.0 to 6.3 | 10–11 % |
| 0.15 | 5.2 to 5.5 | 9 % |
| 0.3 | 4.2 to 4.3 | 7 % |
Watch it here
gen. 0Average active neurons
- Cheap neurons (0.05): –
- Expensive neurons (0.3): –
What it means
The more expensive they are, the more the brain shrinks: evolution trims whatever does not pay. But not even free neurons make it grow: a newly switched-on neuron starts without useful connections and adds nothing until its weights mutate, so nothing selects it. And letting the size change has a price: sometimes an important neuron is switched off, and more cells starve (9–14 % against 5.5 % for a fixed brain).
Reproduce it: Cheap neurons (0.05) Reproduce it: Expensive neurons (0.3)