First to Fly: Insect Flight and Independent Innovation

Three hundred twenty-five million years ago, insects achieved what no animal had done before: they conquered the air. Long before pterosaurs glided through Mesozoic skies, before the first bird took flight, ancient dragonfly-like creatures exploited an entirely new dimension of existence. The fossil record shows what followed—an explosion of insect diversity unmatched in the history of animal life. Flight provided such overwhelming selective advantage that insects rapidly radiated into ecological dominance they maintain to this day.

The Mystery of Wing Origins

For decades, naturalists debated a fundamental question: where did wings come from? The tergal hypothesis proposed that wings arose from dorsal body plates—flat extensions of the insect’s back that might have begun as thermoregulatory surfaces or gliding membranes. The pleural hypothesis suggested wings evolved from lateral leg structures that folded along the torso and gradually migrated upward. Were wings entirely novel structures, or modified pre-existing appendages?

Modern evidence supports the tergal origin. Insect wings evolved from the eighth segment’s dorsal outgrowths—body plates repurposed for an entirely different function. This is not creation from nothing, but transformation through what we might call exaptation. Structures that served one purpose became, through gradual modification, the basis for revolutionary new capability. The developmental programs were already present; selection pressure reshaped them.

Initialization Determines Possibility

Here lies a profound parallel with artificial learning systems. Just as neural networks exhibit initialization sensitivity—where starting parameter configurations determine which solutions gradient descent can reach—arthropod body plans determined which evolutionary innovations became possible. Insects possessed the right substrate: segmented bodies with tergal plates positioned for aerodynamic modification. Other arthropods, differently configured, could not discover this path.

Evolutionary local search operates through random variation and selection, walking gradually down fitness landscapes. But major innovations require fortuitous preconditions. The insect body plan—its particular segmentation, its dorsal plates, its ecological pressures—created an initialization from which flight could emerge. Small differences in starting configuration led to vast differences in evolutionary trajectory.

Limitations and Creative Leaps

Both biological evolution and artificial optimization face fundamental constraints. Evolution cannot make leaps; it proceeds through accumulation of small advantageous changes. Neural optimizers struggle with dimensionality and local minima. Yet somehow, major transitions occur. Wings arose once in insects, then refined across orders over hundreds of millions of years. Similarly, attention mechanisms arose once in transformers, then spread across domains—vision, language, generation.

The pattern suggests that revolutionary capabilities emerge not through directed design but through repurposing. Insect wings: thermoregulatory plates transformed into flight surfaces. Feathers: insulation co-opted for aerial locomotion. Attention: a mechanism for sequence modeling that suddenly enabled novel architectures.

What enables these “inventive moments”? Perhaps it is the convergence of proper initialization—the right substrate already in place—with environmental pressure strong enough to drive exploration, and time sufficient for local search to discover unexpected transformations. Insects succeeded where other arthropods failed not through superior design, but through possessing body plans that, under the right selective pressures, could be reshaped into something entirely new.

From so simple a beginning—a dorsal plate, a random parameter configuration—endless forms most beautiful and most wonderful.