The Self-Organizing Pattern Trap

In the AI Story

De Bono's towel experiment is the simplest illustration. Drop a damp, crumpled towel on a flat surface. It lands in a configuration. Pick it up and drop it again — a different configuration emerges. Each is determined by the material's properties and the surface's constraints. No one designed the folds. The pattern is self-organized. The brain, de Bono argued, works the same way. Neural tissue receives information and the information carves channels through the brain's inherent properties. Subsequent information follows the channels previous information carved. The patterns deepen with repetition.

A neural network exhibits the same dynamic. Training data organizes itself into weighted connections through the network's architecture — its layers, activation functions, loss-minimization dynamics. No one decides what patterns the network will learn. They emerge from the interaction of data's statistical properties and the architecture's computational constraints. The resulting patterns are asymmetric: strong in the direction of formation, weak against the grain. This is why AI output converges toward the center. Not because the model is incapable of producing unusual output, but because self-organizing dynamics favor the conventional over the unconventional.

The trap has a particularly dangerous property: it is invisible from inside. The self-organizing system does not evaluate its own patterns. It simply channels. This is true of the human brain, which does not spontaneously notice its own assumptions, and equally true of the language model, which does not spontaneously recognize its own training distribution. The pattern is the water the fish swims in. Making the water visible requires an external operation — a deliberate intervention that the system, by its nature, cannot perform on itself.

The escape mechanism is what de Bono spent fifty years developing: specific tools (provocation, random entry, the Six Hats) that introduce discontinuities into the self-organizing dynamic. Each tool operates on the same principle: disrupt the pattern so the system is forced to reorganize. The reorganization may or may not produce something useful. The point is not that every disruption succeeds; the point is that without disruption, the system will never produce anything it has not already produced.

Origin

De Bono described the self-organizing pattern trap in The Mechanism of Mind (1969), three decades before the artificial neural network paradigm vindicated the model in computational form. The framework anticipated, with striking precision, both the extraordinary capabilities and the structural limitations of modern large language models.

Key Ideas

Patterns form without designer. Self-organizing dynamics produce stable channels through mechanical interaction of material and environment — no homunculus required.

Asymmetric channeling. Patterns favor the direction of formation and resist movement against or across the grain — the structural basis of both expertise and rigidity.

Invisible from inside. The system does not experience its own patterns as constraint; the excluded paths register as nonexistent rather than forbidden.

Bilateral in AI collaboration. The human's pattern and the machine's pattern reinforce each other, producing double convergence toward conventionality.

Escape is mechanical. Because the trap is produced by mechanical dynamics, it can be sprung by mechanical interventions — not inspiration, but method.