REF · CLASSIC MODEL

Cellular Automata & the Game of Life

Growing arbitrarily complex global behaviour from one fixed rule that looks only at neighbours — from the 256 elementary rules to the Turing-complete Game of Life

Cited in: Topic 4 · The Infinity of Minimal Rules

01The Question It Poses

Is it possible, with the barest possible device — a row (or sheet) of cells and one fixed rule that looks only at neighbours — to generate arbitrarily complex global behaviour? A cellular automaton (CA) is the minimal test bench for the causal chain "local rule → global behaviour": strip the mechanism down to "cells + local update," and see whether complexity can still grow.

Its origins show the weight of the question. John von Neumann invented CAs in the 1940s to study "a machine that can reproduce itself"; Conway turned them into a popular phenomenon in 1970 with four rules; Wolfram systematically swept all 256 one-dimensional elementary rules in the 1980s and distilled four classes of behaviour. All three were asking the same thing: how low is the floor of order — how few rules does it take before complexity begins to grow?

02The Rules

Elementary cellular automaton (1D):

  1. A row of cells, each 0 or 1 (white or black).
  2. A cell's next value depends only on itself and its two immediate neighbours right now.
  3. Three cells, two states each: eight possible neighbourhoods; assigning an output to each of the eight defines one rule.
  4. So the total number of rules = 28 = 256. Read the eight outputs as a binary number and you get the rule's index (Rule 30, Rule 90, Rule 110…).

Conway's Game of Life (2D), with even fewer rules — just four:

  1. A grid of cells, each alive or dead; each cell has 8 neighbours (including diagonals).
  2. A live cell with exactly 2 or 3 live neighbours stays alive; otherwise it dies (too alone or too crowded).
  3. A dead cell with exactly 3 live neighbours comes to life.
  4. All cells update by this rule at once, generation by generation.

03What You See When It Runs

Start from a single black cell at the top and stack each moment's row downward to get a spacetime diagram. Both "look at three cells only," yet Rule 90 grows a fractal (the Sierpiński triangle) while Rule 30 grows noise that passes randomness tests — the difference is only that eight-row lookup table.

One line of rule, drawn downward through timeRule 90 — a fractal (Sierpiński triangle)Rule 30 — chaos (used as a random-number source)Same setup, one black cell at the top. Only the rule differs.
Same start, same "look at self and two neighbours": Rule 90 grows a fractal, Rule 30 grows noise.

Taken as a whole, the 256 rules sort into four classes: Class I quickly dies to a uniform state; Class II settles on fixed or periodic structures; Class III is perpetual noise (Rule 30 lives here); Class IV, the rarest, grows localized structures that move and collide, poised at the "edge of chaos" (Rule 110 lives here).

In two dimensions the most famous local structure is the glider: five cells that return to their original form every four steps, shifted one cell over. Using gliders as signals and "glider guns" as sources, you can build logic gates and from there a full universal computer — which is what it means for the Game of Life to be Turing complete.

A glider: four rules, and a shape that walksgen 0gen 1gen 2gen 3gen 4 = gen 0, moved one cell ↘Nothing “moves” — cells switch on and off by a fixed local rule. The motion is an illusion the rule prints.
The glider translates one cell every four steps. It "moves," yet nothing moves — cells just switch on and off by the rule.

04What It Explains

A CA's greatest use is an existence proof: if you can reproduce some macroscopic pattern with one minimal local rule, you have proven that the pattern needs no central command and needs no complex cause. This is the cleanest floor for "self-organization."

Concrete real-world correspondences run in a string: the pattern on a cone snail's shell (the spacetime diagram of a 1D rule, its time axis frozen into calcium), the discrete version of reaction–diffusion patterns (Topic 13's Turing patterns), lattice models of forest fire and percolation (Topic 17), and abstractions of parallel computing and lattice gases. Its core falsifiable expectation: given a macroscopic pattern, if a simple local rule can reproduce it, the "complex design / central control" hypothesis is weakened; if nothing can, that in turn signals long-range coupling or a global constraint in the system.

What It Cannot Explain

Further Reading