A small random deviation prevents emergent systems from collapsing into a uniform state

In agent-based or rule-driven generative systems, elements following identical behavioral rules tend to synchronize and cluster — the system becomes homogenous. Adding a small, bounded random deviation (jitter) to each element’s update prevents this collapse without destroying the system’s overall structure. The result is a system that maintains global texture and visual identity over long time spans while individual elements keep shifting. This is related to the statistical concept of entropy: without perturbation, the system drifts toward a low-entropy attractor state. Too much noise destroys structure; too little causes stagnation. The principle applies to autonomous-agent simulations, flocking algorithms, and any iterative generative process.

Examples

Reas’s line-element simulations would ‘group together, move in the same direction, and then it becomes basically unified’ without jitter. Adding a small deviation keeps them ‘homeostatic.‘

Assessment

Take a simple agent system (e.g. Hydra or p5.js boids) and observe what happens at three noise levels: 0, small, large. Describe the visual difference.