Layers I through III establish the substrate: constraint, replication, and variation. But replication with variation, unchecked, produces only divergence: an expanding cloud of instances drifting outward without pattern.
Direction emerges when feedback is introduced. And direction does not require consciousness.
Goal-directed behaviour, at its structural core, is metric minimisation under iteration. A system has a measurable target. It takes an action. It measures the result. It adjusts.
The cycle is simple: act, measure, adjust, repeat. The cumulative effect is structured progress: not because the system "understands" its goal, but because the feedback loop mechanically reinforces convergence.
Consider a shared board: a grid of cells, initially all white. The goal: turn every cell red. Each instance, on each cycle, selects an empty cell and changes its colour. A counter tracks who placed first, how many cells each instance claimed, and the cumulative progress toward completion.
No instance sees the full picture. No instance knows the strategy of the others. But each receives feedback: the board changes, the count updates, the progress is visible.
The goal is simple. The behaviour is iterative. The effect is cumulative. And from these minimal ingredients: a shared metric, individual action, visible feedback: structured collective progress emerges.
Without feedback, variation is random walk: each instance explores aimlessly, producing divergent but unpatterned results. With feedback, variation becomes search: each instance's actions are evaluated against a measurable criterion, and the results inform subsequent actions.
This is the structural transition from undirected exploration to guided adaptation. It is the moment when a population of replicating, varying instances begins to converge rather than merely diverge.
Feedback does not require an external designer specifying the metric. Feedback can emerge from the environment itself: resource availability, spatial occupation, interaction outcomes. The environment provides the signal; the system responds through iteration.
Direction can be demonstrated with a program that sends a small message to a shared display: it places a red dot on a cell, confirms the placement, and increments a counter. Each instance competes: not through antagonism, but through parallel action toward a visible, shared metric.
Watch the board fill. Watch the counters track contribution. Watch structured progress emerge from simple, iterated, feedback-guided actions.
Feedback turns noise into signal. Without it, variation drifts. With it, strategies converge.