sciencebriefs
13:00in productionCh. 1 · A question about copying/ 13:00 · ceiling 15 min
Computing & AI

Cellular automaton

Von Neumann built a grid of simple cells following fixed local rules to answer a question about machines copying themselves, and the idea outgrew its original purpose entirely.

Working at Los Alamos with Stanislaw Ulam, John von Neumann designed a cellular automaton, a grid of cells each following the same simple rule based on its neighbours, as a way to ask whether a machine could construct a working copy of itself without outside help. His design used 29 possible states per cell and produced a genuine universal constructor, a theoretical proof that self-reproduction did not require anything mysterious, years before the structure of DNA was known. The idea of a grid governed by local rules later took on a life of its own, most famously in John Conway's Game of Life and in Stephen Wolfram's later classification of how such simple systems can behave, from settling down to running forever without repeating.

Chapters & takeaways6
  1. 0:08
    A question about copying

    Von Neumann wanted to know if a machine could build a working copy of itself using only local, mechanical steps.

  2. 2:10
    Cells, states, and neighbours

    A cellular automaton is a grid where every cell updates at once, based only on its own state and its neighbours', under one fixed rule.

  3. 4:20
    The 29-state constructor

    Von Neumann's design, worked out with Stanislaw Ulam, used 29 states per cell and produced a genuine self-copying construction, on paper, in the 1940s.

  4. 6:30
    Conway's simpler game

    Decades later, John Conway's two-state Game of Life showed that far simpler rules could still produce open-ended complexity.

  5. 8:40
    Wolfram's classes

    Stephen Wolfram sorted these systems into four behavioural classes, and one rule, Rule 110, was later shown capable of any computation.

  6. 10:50
    From toy grids to real models

    The same local-rule approach now underpins simulations in biology, chemistry, and physics, far from its origin as a thought experiment about machines.

Worth your time?

Yes. Study the whole thing.

4/ 5
What works
  • it traces a clean line from a 1940s thought experiment to a still-active field of study
  • the self-reproduction problem is explained without requiring biology or computer science background
  • it gives proper credit to Ulam's role alongside von Neumann's
What does not
  • the mathematics of why Rule 110 is computationally universal is asserted rather than shown
  • the practical applications section reads as a list rather than a developed argument
Study it if
  • anyone who has played with Conway's Game of Life and wondered where the idea came from
  • readers interested in the history of computing before computers existed
  • people curious how simple rules produce complicated outcomes
Skip it if
  • readers wanting a technical treatment of automata theory
  • anyone expecting the story to be mainly about von Neumann's other, more famous work
The written brief4 min read

A question about copying

A cellular automaton is a grid of cells, each in one of a small number of states, that all update together at every tick of a clock according to a single fixed rule applied identically everywhere. The rule looks only at a cell’s own state and the states of its immediate neighbours, yet from that narrow local view, patterns can spread, stabilise, or churn indefinitely across the whole grid. John von Neumann built the first serious version of this idea at Los Alamos in the late 1940s, working with Stanislaw Ulam, not as an exercise in pattern-making but to settle a specific question: could a machine, built from simple mechanical parts, construct a working copy of itself using nothing but local instructions, without any outside designer stepping in to help.

Cells, states, and neighbours

Von Neumann’s answer took the form of a cellular automaton with 29 possible states per cell, laid out on a grid, encoding both a constructor and the instructions the constructor would follow to build a copy of itself and its own instruction tape. The construction was never built physically; it existed as a formal design, worked through on paper and later analysed as a rigorous piece of computation theory. What mattered was that it worked in principle: a self-copying machine could be specified using only local rules and finite information, with no need for anything beyond what mechanism and logic could supply. Von Neumann’s analysis of what self-replication actually requires structurally came before the discovery of DNA’s own structure, which later turned out to solve the same problem by different means.

The 29-state constructor

The theoretical result holds up as exactly what it claimed to be: a proof that a universal constructor is possible within a cellular automaton framework, given enough states and a sufficiently large grid. It does not depend on any assumption later found to be shaky, and the framework itself proved durable enough to be picked up again decades later by researchers studying artificial life and self-replicating systems more generally. The broader concept of a cellular automaton, cells on a grid updating by a shared local rule, has likewise held up as a genuinely useful and general modelling tool, used well outside its original context to represent physical and biological processes that share the same basic structure of many simple, locally interacting parts.

Conway’s simpler game

What the original 29-state design does not offer is practicality: nobody has built von Neumann’s constructor as a physical machine, and it was never intended as an engineering blueprint. It is also worth separating von Neumann’s specific self-reproduction proof from the later popular fascination with cellular automata in general, which took off mainly through John Conway’s much simpler Game of Life in the 1970s and Stephen Wolfram’s systematic study of one-dimensional rules in the 1980s. Those later developments, including the finding that a rule as simple as Rule 110 can perform any computation a general-purpose computer can, are extensions of the framework rather than confirmations of von Neumann’s original construction, and the two threads are often blurred together in popular retellings.

Wolfram’s classes

The lasting significance is less about self-replicating machines, which remain largely theoretical, than about what the framework revealed regarding complexity itself: that astonishingly rich, unpredictable behaviour can emerge from rules simple enough to state in a sentence, applied uniformly and without any central coordination. That insight has since been put to work modelling how shell patterns form, how chemical reactions oscillate, how traffic jams propagate, and how physical systems undergo phase transitions, all treated as grids of simple interacting units rather than as their original, very different subject matter. It also fed directly into later questions about what minimal ingredients are needed for a system to be capable of universal computation, connecting a Cold War-era thought experiment to present-day theoretical computer science.

From toy grids to real models

This rewards time from anyone who enjoys watching a simple idea turn out to be far more general than its inventor intended. The self-reproduction puzzle is a genuinely elegant piece of reasoning, and it is satisfying to see it laid out before jumping to the more familiar Game of Life, which most readers will have encountered without knowing its intellectual ancestry. It asks less patience than some foundational computer science topics, since the core grid-and-rule idea is visual and easy to hold in mind, even if the 29-state details of von Neumann’s own construction take more effort to follow. Readers chiefly interested in von Neumann’s other achievements, in game theory or computer architecture, will find only a narrow slice of him here, but that slice is a good one.

Same field · Computing & AI4 of 46
Up next in Science

Chargaff's rules

· 13:00

Erwin Chargaff measured DNA base ratios across dozens of species and found a pairing rule steady enough to become the numerical clue behind Watson and Crick's double helix — credit for which he never received.

13:00