communicoupling · concept 14 of 26
Why does a society settle into stable cultural regions — dialects, tastes, worldviews that hold their edges for generations — when no one draws the borders and everyone is free to change? Axelrod's model shows that the very force that should homogenize everyone, the tendency to become more like those we interact with, instead freezes a patchwork of distinct cultures in place.
Interaction makes people more alike. You pick up the words, the tastes, the assumptions of those you talk to; contact rubs off. Run that rule forward and the obvious prediction is a uniform monoculture — a single blur into which every difference eventually dissolves. Instead, we get stable diversity. Regions, subcultures, camps that persist and border each other for lifetimes, never merging.
Axelrod found the twist hiding inside the homogenizing rule itself. Because people interact more with those already similar, and copy only from them, similarity and interaction feed back into each other. Neighbours who happen to share nothing have no channel to influence one another — and once a border of zero overlap forms, it can never open again. The population freezes into bordered cultural regions that have gone permanently deaf to each other.
The homogenizing rule that fragments
Each agent carries a culture: a short vector of features — call them language, cuisine, faith, politics, dress — each set to one of q traits. The update is almost embarrassingly simple, and it only ever makes people more alike. Pick an agent and a random neighbour. Measure their similarity: the fraction of features on which they already agree. With probability equal to that similarity, they interact — and interaction means the agent copies one feature it didn't share. No one ever grows more different. Every single event is a step toward sameness.
The catch is the gate. Similarity doesn't just measure the outcome of interaction; it controls whether interaction happens at all. Two neighbours who already share most features talk constantly and converge fast. Two who share only a little talk rarely. And two who share nothing — similarity 0 — can never talk, so they can never copy, so they can never come to share anything. That is an absorbing state: a border across which influence is arithmetically impossible. The rule that says "grow alike" quietly digs a moat around every difference it fails to erase early, and the moat is permanent.
What to try
Hit Run to freeze and watch interactions still possible fall to 0%. The grid stops moving — not because it homogenized, but because every remaining border is a wall of zero overlap. Read off how many distinct regions survived.
Drag q up from 2. At low q everything merges to one culture. Somewhere in the middle the count jumps sharply, and at high q the map shatters into many tiny frozen cultures. More initial variety yields more lasting regions.
The bright gold walls mark edges where neighbours share nothing. Trace one: the region on each side can never again influence the other. Those walls are the borders no one drew — and once drawn, they do not move.
Borders no one drew
Let the simulation run to its end and you get a map that looks designed: contiguous regions of uniform culture, each internally identical, meeting along sharp boundaries. But nobody placed those boundaries. They are wherever the drift happened to leave two neighbours with no feature in common — and there the process simply ran out of channel. A frozen region is not a decision; it is the residue of a conversation that stopped being possible. Dialects hold their isoglosses, tastes cluster by neighbourhood, political camps talk past each other — not because anyone enforces the lines, but because influence across them has gone to zero and stayed there.
Unlike segregation by preference, no agent here wants to be apart; every agent is trying, always, to grow more like its neighbours. The regions are not the goal of anyone in the model — they are the trap the homogenizing rule builds for itself. Coupling, gated by likeness, decouples: the more the similar converge, the more sharply they cut off from the dissimilar, until the dissimilar are sealed away behind a border that no further contact can ever cross.
Diversity and its fragility
More possible variety yields more lasting diversity. With few traits per feature, agents start out sharing a lot by accident, so almost every pair can interact, and the whole grid drains into one culture before any wall can set. Crank q up and the initial soup is so varied that many neighbours share nothing from the start — those walls set immediately, quarantining little pockets before the homogenizing pull can reach them. The transition between the two regimes is sharp. Grid size and feature count push it around: a bigger world, or a shorter culture vector, tends to homogenize more, because there are more paths for influence to travel and fewer independent features to disagree on.
And yet the frozen map is fragile in a way the segregation grid is not. Everything here rests on interaction being pure copy-if-similar. Add a trickle of noise — the odd agent who spontaneously flips a feature, an innovation, a mistake — and a wall of zero overlap can suddenly regain a shared feature, reopen, and let the homogenizing flood back through. A single long-range tie, one bridge across a border, can do the same. Real culture is not a fixed vector of independent features, and real interaction is not pure copying. The model earns a genuine insight — likeness-gated influence can manufacture stable borders from nothing — while remaining a toy about how those borders can also, at the lightest touch, dissolve.
The mapping
| In the model | In the world |
|---|---|
| a culture vector | The bundle of features and traits a person carries — language, tastes, beliefs, habits. |
| similarity-gated interaction | Talking more, and copying more, with those already like you — contact is not evenly spread. |
| feature copying | Becoming more alike through contact: picking up a word, a taste, an assumption from someone you engage. |
| a frozen region | A stable cultural area — a dialect zone, a subculture, a camp that holds its edges for generations. |
| a zero-overlap border | A boundary across which influence is impossible: two groups with no shared ground left to build on. |
| q, traits per feature | How much initial diversity there is to freeze — the raw variety a society starts out holding. |
Where it tears
Real culture is entangled, hierarchical, and invented — not a tidy list of independent slots that only ever get copied. Faith shapes cuisine shapes politics; new traits are created, not merely transmitted; meaning is negotiated, not overwritten. The model's clean feature-vector is a deliberate cartoon, and the crispness of its borders is partly an artifact of that cartoon.
The absorbing state depends entirely on interaction being noiseless copying. This is a documented sensitivity: add even a small rate of random feature-flipping (cultural drift, innovation, error) and the frozen patchwork can melt back toward monoculture. So "stable diversity emerges naturally" is only half the story — the stability is real but brittle, and the model's own literature shows how little noise it takes to undo it.
It is seductive to read this grid as proof that cultural separation is innocent, emergent, nobody's doing. But much real segregation of tastes, dialects, and worldviews was enforced — by conquest, exclusion, redlining, schooling policy, and law. A model where borders arise from pure innocent drift must never be used to prettify boundaries that were, in fact, imposed on people who did not choose them.