How Simple Rules Build Complex Worlds: From the Game of Life to Phantom Traffic Jams

When a murmuration of starlings swings across the sky as a single shape, no bird designed that shape. When motorway traffic grinds to a halt for no visible reason and then clears a few minutes later, no driver planned it. Each individual reacts only to what is right next to it, yet the whole displays a structure that was nobody’s intention. That is what people usually mean by emergence.

This article follows four classic models of emergence in the order they appeared. Each has rules short enough to fit on an index card, and each can be run in your browser.

1970: life on graph paper

The mathematician John Horton Conway invented the Game of Life, and it reached a wide audience through Martin Gardner’s “Mathematical Games” column in the October 1970 issue of Scientific American. The rules:

Cells on a square grid are either alive or dead.
A dead cell with exactly three live neighbours (out of eight) becomes alive.
A live cell with two or three live neighbours stays alive; otherwise it dies.

Today this is written B3/S23. Conway was after a balance: rules under which patterns neither explode trivially nor die out at once. Fill a grid at random and run it, and you see exactly that. The soup churns for a while and then settles into small debris: 2×2 blocks, beehives, three-cell bars that flip back and forth.

Some of that debris moves. The five-cell glider returns to its own shape every four generations, shifted one cell diagonally. Nothing in the rules mentions movement. Cells are only born and die, yet our eyes see an object walking across the grid. It is about the simplest possible demonstration of emergence.

A pattern that grows forever

Conway suspected that no finite starting pattern could grow without limit, and offered a prize to anyone who could prove him wrong. Late in 1970 Bill Gosper’s group at MIT did, with the glider gun: a structure that oscillates in place and fires a new glider every 30 generations. Each glider adds to the population, so the count of live cells climbs without end.

The gun raised a bigger question. If gliders can serve as signals and collisions between them as logic operations, could you build a computer inside Life? The answer turned out to be yes. Berlekamp, Conway and Guy sketched the construction in Winning Ways for Your Mathematical Plays, showing that Life is Turing complete: in principle it can carry out any computation. Enthusiasts later built working Turing machines inside it.

That result has a sharp consequence. In a Turing-complete system, questions like “will this pattern ever stop changing?” are in general undecidable, the same as the halting problem. Knowing the rules perfectly does not guarantee a shortcut to the outcome; sometimes the only way to find out is to run it. Simple rules do not imply predictable results.

Try it: in the Game of Life, stamp the glider gun, then switch the rule to HighLife (B36/S23) or Seeds (B2/S) and see how much one digit changes the world.

1986: one ant and the start of artificial life

In his 1986 Physica D paper “Studying artificial life with cellular automata”, Christopher Langton described an ant walking on a grid. Its rules are even shorter than Life’s:

On a white cell, turn 90° right; on a black cell, turn 90° left.
Flip the colour of the cell, then step forward.

Starting from a blank grid, the ant spends its first few hundred steps drawing small, fairly symmetric shapes. Then the symmetry breaks and an irregular blob grows for roughly 10,000 steps. Then, abruptly, the ant falls into a 104-step routine that repeats forever and builds a straight diagonal strip. It is known as the highway.

The ant shows a different face of emergence. In Life, many cells interact simultaneously; here there is only one agent, and it interacts with the trail it left behind. The grid acts as the ant’s memory.

The gap between what is proven and what is not is striking. Nobody has proven that the ant always ends up building a highway. What has been proven is weaker: from any finite starting pattern the ant’s path is unbounded, so it never stays trapped in a finite region. Meanwhile, in 2002 Gajardo, Moreira and Goles showed that with a suitably designed starting pattern, a single ant’s path can evaluate any Boolean circuit. A two-line rule can be a computer.

Try it: in Langton’s ant the highway appears after a little over two seconds at the default speed. Change the rule string to something like LLRR and the ant cycles through four colours and draws something entirely different.

1987: a flock without a conductor

At SIGGRAPH in 1987, the computer graphics researcher Craig Reynolds presented “Flocks, Herds, and Schools: A Distributed Behavioral Model”. His goal was animation, not biology. Scripting the path of every bird in a flock by hand was impractical, so he gave each bird rules and let the flock take care of itself.

His simulated birds, which he called boids, each look only at nearby flockmates and do three things:

Separation: steer away from neighbours that are too close.
Alignment: steer towards the average heading of neighbours.
Cohesion: steer towards the average position of neighbours.

No boid sees the whole flock and there is no leader. Still, the screen fills with a convincing flock that splits around obstacles and rejoins behind them. The approach became a standard way to animate herds, swarms and crowds in films and games.

Boids differ from Life in almost every detail: no grid, no cells, continuous positions and velocities. The underlying structure is the same, though: local rules, many agents, no central control. Put those three together and global patterns appear. In 1995 the physicist Tamás Vicsek and colleagues stripped the idea down further, showing that a rule of matching neighbours’ direction plus random noise is enough to produce a phase transition from disordered motion to collective motion. That work helped turn flocking into a research topic in physics.

Try it: in the boids simulation, drag Alignment to zero and watch the Order number drop; shrink the vision radius and the flock breaks into many small groups.

1992: a cellular automaton for the motorway

Traffic jams are emergent too. In 1992 the German physicists Kai Nagel and Michael Schreckenberg published a minimal model that divides a road into cells. Each car has an integer speed, and every time step it applies four rules in order:

If below the speed limit, speed up by one.
If the speed exceeds the number of empty cells ahead, slow down to that number.
With probability p, slow down by one.
Move forward by the current speed.

Rule 3 is the key: it builds in the human habit of easing off for no particular reason. Above a certain density, one car’s random slowdown makes the cars behind slow down in turn, and a cluster of stopped cars appears on its own. No accident, no lane closure, no bottleneck. Such jams are often called phantom jams.

2008: proving it with real cars

The model’s prediction was demonstrated with real drivers in an experiment by Yuki Sugiyama and colleagues, published in the New Journal of Physics in 2008. They spaced 22 cars evenly around a circular track 230 m in circumference and told the drivers to cruise at 30 km/h while keeping a safe distance.

At first the flow was smooth. But each driver’s speed wobbled slightly, the gaps became uneven, and soon cars bunched up at one point, stopped briefly and moved off again. The cluster travelled backwards, against the direction of the cars, at about 20 km/h. As the paper’s title puts it, these were traffic jams without bottlenecks. The authors concluded that a bottleneck is only a trigger, not the underlying cause: above a critical density, free flow is itself unstable.

Can the wave be removed? In 2018 Stern and colleagues ran a similar ring-road experiment with more than 20 vehicles and made just one of them an automatically controlled car. By driving smoothly instead of copying every wobble of the car in front, it damped the stop-and-go waves and cut the fuel consumption of the whole group. Emergence governs the whole, but changing one local rule can change the whole.

Try it: the phantom traffic jam simulator uses the Intelligent Driver Model, with continuous positions and speeds, instead of Nagel–Schreckenberg’s cells. On the default 1 km ring with 45 cars, one hard brake leaves a jam that never dissolves. With 30 cars, the same brake is absorbed within minutes.

Common misconceptions about emergence

“Emergence is inexplicable.” It is not. Every model above follows deterministic or explicitly stated random rules and can be reproduced exactly on a computer. Calling a pattern emergent means it is not written into any individual rule, not that it has no cause.

“If the rules are simple, the outcome must be predictable.” Life and Langton’s ant are both Turing complete, and for such systems there is in general no guaranteed shortcut faster than running them. Why Langton’s ant starts its highway after about 10,000 steps is still unproven.

“A flock must have a leader.” Boids show that a flock can form without one. Whether a particular animal species has leaders is a separate empirical question, but a leader is not needed to explain the shape of the flock.

“Somebody caused the jam.” In Sugiyama’s experiment every driver did as instructed. The jam was a property of the group, arising from density and how drivers respond to one another, not the mistake of a single person. Individual habits still matter, though: as Stern’s experiment showed, one smooth driver can calm the wave.

The four models side by side

Model	Year	Agents	Core rule	What emerges
Game of Life	1970	Grid cells	Birth and survival by neighbour count	Gliders, oscillators, computation
Langton’s ant	1986	One ant	Turn by colour, flip colour	A highway after chaos
Boids	1987	Hundreds to thousands	Separation, alignment, cohesion	Flocks without leaders
Traffic models	1992 onwards	Cars	Adjust speed to the car ahead, plus some noise	Phantom jams moving backwards

In each case the agents see only what is near them, nothing coordinates them centrally, and the interactions repeat many times. Each of the four tools lets you change the rule’s numbers one at a time, which is the quickest way to find where a structure appears and where it falls apart.