Order Without a Plan: A Field Guide to Emergence

Here is a claim you can test tonight on anything alive: the whole does things none of its parts can do. A lone ant is close to stupid: it follows a trail, walks in circles, dies if you drop it ten centimetres from the nest. The colony, though, builds bridges out of its own bodies, regulates its temperature, finds the shortest path to food, wages war. No ant holds the plan. The plan is written nowhere, and yet it happens. That is emergence: a global order that arises from local interactions, with no conductor and no centre.

A working definition to keep: there is emergence when a higher level (the flock, the city, the mind) shows properties that the lower level (the bird, the resident, the neuron) does not have. The phrase to hammer throughout: no conductor. Nobody coordinates. The order is distributed in the rules each part follows, and repetition does the rest. Let’s look at three tiny worlds where it jumps out at you.

World 1 · The Game of Life: four rules, and a universe

In 1970 the mathematician John Conway invented a “game” with no player. A grid of cells; each cell is alive or dead; each cell looks at its 8 neighbours and applies the same handful of rules every turn: a dead cell with exactly 3 live neighbours is born; a live cell with 2 or 3 neighbours survives; otherwise (loneliness or overcrowding) it dies. That is all. Zero intelligence in any cell.

This is the first shock of emergence: you can conjure universal computation out of rules a child grasps in thirty seconds. Nobody “programmed” the glider; it is a consequence of the rules, discovered after the fact by watching the grid run.

World 2 · Rule 30: chaos out of nothing

Go simpler still. Take a single line of cells. Each cell, next turn, depends on just three cells: itself and its two immediate neighbours. Three black-or-white cells make 8 possible situations; a “rule” says, for each of the 8, whether the cell turns black or white. So there are only 256 possible rules. Stephen Wolfram numbered them all in the 1980s. Number 30 became famous. Start from one black cell in a white line, stack the generations downward, and watch:

That last point is the vertigo. It is called computational irreducibility: for some systems there is no formula that jumps straight to line 1,000. The only way to know what they do is to let them run, step by step. The computation is the answer; there is no shortcut. This is why weather, a crowd, a market predict so badly: not because we lack equations, but because sometimes you have to live through the system to know what comes next.

World 3 · Boids: three rules and a murmuration

In 1986 Craig Reynolds wanted to animate flocks of birds on film without drawing each one. His idea: give every bird (a “boid”) three rules that depend only on its close neighbours: separation (don’t crowd), alignment (take the average heading of your neighbours), cohesion (steer toward the local centre of the group). No leader, no global plan, no target shape. Release five hundred boids: they form a flock that ripples, splits and re-joins, indistinguishable from a real starling murmuration.

The common engine

Three worlds, one motor. Every time: (a) identical, simple units; (b) that see only their immediate neighbourhood; (c) a rule applied again and again. Nothing else. And yet out come objects, universal computation, chaos, coordination. It is the deepest reversal in the science of systems: you do not need a designer to get design. Good local rules and time are enough.

A word on where the interesting things live. Too much order: everything freezes (a crystal, a grid that dies in two turns). Too much chaos: everything dissolves into noise (the snow of an old TV). Life, computation and thought settle on a razor’s edge between the two, the edge of chaos: stable enough to keep a structure, loose enough to transform it. Picture a slider: concrete on the left, smoke on the right; in the middle, a living stream that neither freezes nor evaporates.

Two honest caveats. Weak emergence: the whole is surprising but fully contained in the rules: the glider is deducible from the Game of Life, you just had to simulate it to see it. This covers almost everything we handle. Strong emergence: the far more contested idea that a higher level has causal powers irreducible to the level below (consciousness is the poster child). Stay careful: most researchers lean on “weak plus computational irreducibility”, which already explains why systems surprise us so much.

And crucially: there is no free order. A snowflake, a flock, a cell self-organise by creating local order, while dumping disorder (heat) into their surroundings. The universe’s global books keep worsening. This is the resolution of Maxwell’s old demon, the imaginary gatekeeper that seemed to sort molecules and lower entropy for free: the sorting is paid for by the information the demon must store and erase, and erasing a bit has an unavoidable thermodynamic cost (Landauer’s principle). Emergence does not cheat thermodynamics; it pays for local order with exported disorder. These “dissipative structures” (Ilya Prigogine, Nobel 1977) stand only while energy flows through them. Cut the flow, the flock lands, the pattern dies.

This is why the deepest example of all is the mind. None of a brain’s ~86 billion neurons thinks; none contains “you”. A neuron is a slightly richer Game-of-Life cell: it sums its inputs, crosses a threshold, fires or not. From that local swarm, with no conductor and no locatable “centre of the self”, a mind emerges. A person may be less a thing you could copy than a pattern kept alive by a flow: a murmuration that holds only while the birds keep flying.

Where this stands in 2026

The idea is far from settled science history; it is a live front.

Researchers now build cellular automata that learn to self-organise. A paper presented at the GECCO conference in July 2025, “A Path to Universal Neural Cellular Automata”, replaces Conway’s hand-written rules with rules learned by gradient descent, aiming to make a universal Turing machine emerge inside a fabric of small, identical cells. These “neural cellular automata” can already grow a target shape from a single cell and regenerate it after you cut it away: a decentralised substrate, with no boss, that learns both to compute and to repair itself.

In artificial intelligence the word “emergence” itself is under fire. Large language models were said to show emergent abilities: a skill that appears suddenly past a certain size. A well-known result (Schaeffer et al., a best-paper award at NeurIPS 2023) argued this may be an artefact of the chosen metric: a hard, all-or-nothing score shows a cliff, while a continuous score shows a smooth slope. Others reply, through 2025–2026, that some abilities really do require a structural threshold. The debate is unresolved, and the discipline it teaches (telling “a genuine new property emerged” from “my measurement manufactured the illusion of a jump”) is worth far more than either verdict.

And the living systems keep giving up their secrets. By filming falcons hunting flocks of thousands of starlings, researchers have shown that the escape pattern depends on how fast the escape information propagates and on the flock’s previous state (a hysteresis: the past matters). The murmuration sits at the edge of chaos, tuned to react in an instant: a living proof that three local rules plus a critical state give collective intelligence with no central brain at all.

Further reading

• Growing Neural Cellular Automata (Distill, 2020), an interactive article: watch a “creature” grow from a single cell, then cut it and see it regrow. Emergence and learning, in animations you can poke.
• Play Conway’s Game of Life, drop a glider, press play, watch it walk. The fastest way to feel “the rules do the rest”.
• Parable of the Polygons (Nicky Case & Vi Hart), Schelling’s segregation model as a playable post: tiny individual preferences produce a total segregation nobody wanted.
• A Path to Universal Neural Cellular Automata (arXiv 2505.13058), learning the update rule so universal computation emerges.
• Emergent Abilities in Large Language Models: an explainer (CSET), a clear map of the “mirage” debate.