Artificial Life and Emergent Systems

FeedbackFeedback Control and Requisite Variety covers cybernetics' regulation half — a system steering itself toward a known goal. This page covers its other founding thread: what happens when many simple parts, each following a small, local rule with no awareness of the whole, are left to interact — and complex, sometimes lifelike, behaviour emerges that no individual rule specifies or predicts.

Conway's Game of Life

JohnJohn Conway's Game of Life, introduced to a wide public through Martin Gardner's Scientific American column, is the clearest possible demonstration of emergence¹. The entire system is a grid of cells, each either alive or dead, updated simultaneously every step by three rules: a live cell with fewer than two or more than three live neighbours dies (of isolation or overcrowding); a live cell with two or three live neighbours survives; a dead cell with exactly three live neighbours becomes alive. Nothing in those three rules mentions gliders, oscillators, or self-replication — yet starting from the right initial pattern, all of them appear, unplanned, purely as a consequence of the same three local rules applied everywhere, every step.

That small three-cell pattern (a "blinker") oscillates forever between a vertical and horizontal line, a pattern nobody wrote directly — it's a consequence of the same underlying rule applied uniformly, discovered rather than designed. Life belongs to a broader class called cellular automata — grids of cells, each with a small set of possible states, updated in parallel according to a fixed local rule — a modelling tool used well beyond recreational mathematics, for phenomena from crystal growth to traffic flow, wherever local interaction rules are known but the resulting large-scale pattern isn't.

A cellular automaton is, in a real sense, the purest possible case of the connectionist side of that page's distinction: no cell holds an explicit symbol or rule about the pattern it's part of, no central process is reasoning about gliders or oscillators, and yet the large-scale behaviour is exactly as real and predictable as if something had planned it. Every connectionist system on this site — a neural network's weights, an ant colony's pheromone trails — is a more elaborate version of the same basic move.

Daisyworld: Emergent Self-Regulation

WatsonWatson and Lovelock's Daisyworld model connects emergence directly back to the feedback and homeostasis covered on the previous page, but produces the regulation as an emergent property rather than designing it in directly². The model: an imaginary planet seeded with two flower species, one black (absorbs more sunlight, warms its surroundings) and one white (reflects more sunlight, cools its surroundings), competing for space as the sun's output gradually increases over the simulation. Neither species "wants" to regulate planetary temperature — each simply grows fastest at its own preferred local temperature and out-competes the other wherever it's warmer or cooler than that optimum.

The emergent result: as the sun warms, black daisies initially do well (their extra warmth helps them thrive at what would otherwise be a too-cold planet), but as it warms further, white daisies gain the advantage (their cooling effect keeps their local patch survivable while unchecked black daisies would overheat), and the shifting balance between the two species holds the planet's average temperature far more stable than the sun's steadily increasing output alone would predict — global thermostatic regulation, with no thermostat, no controller, and no species acting with the "goal" of regulating anything. Daisyworld was built specifically to demonstrate that Gaia-style planetary self-regulation doesn't require anything mystical or teleological to work — ordinary competitive selection acting on individually self-interested organisms is sufficient to produce it as a side effect, provided the right feedback structure (each species' own success being tied to the temperature it itself helps create) is present.

Cybernetic Robotics: Grey Walter's Tortoises

WilliamWilliam Grey Walter's "Elmer" and "Elsie," built in the late 1940s, were among the first autonomous robots ever constructed, and deliberately so — Walter built them specifically to demonstrate that complex, purposeful-looking behaviour didn't require complex control, just the right feedback wiring³. Each tortoise had exactly two sensors (light and touch) and two "neurons" (vacuum-tube circuits), yet the resulting behaviour looked unmistakably purposeful: seeking moderate light (avoiding both darkness and blinding brightness), navigating around obstacles, and returning autonomously to a recharging hutch when its battery ran low — all from two sensors and two neurons wired as feedback loops, with nothing resembling a planning module or an internal map anywhere in the circuit.

The tortoises are the direct ancestor of a design philosophy still very much alive in robotics: reactive control, where complex-looking behaviour emerges from simple sensor-to-actuator feedback loops rather than from an internal model of the world being consulted and reasoned over — the same "simple local rule, complex emergent result" idea as the Game of Life and Daisyworld above, just wired into physical hardware instead of simulated on a grid or in a population model.

References