Algorithmic monoculture

Algorithmic monoculture is the convergence of multiple autonomous algorithms on identical or near-identical strategies, belief structures, or behavioral patterns despite being nominally independent. Unlike monoculture in agriculture — which is a deliberate choice for yield optimization — algorithmic monoculture is typically an emergent and unintended property of agent economies. It occurs when algorithms share training data, optimization objectives, or architectural templates, causing them to correlate in ways that eliminate the independent variation required for collective error correction.

The phenomenon is particularly dangerous because it is invisible: the algorithms appear diverse (different developers, different deployment contexts) while being functionally homogeneous. A financial market populated by fifty algorithms using the same risk model is not fifty independent agents; it is one agent with fifty mouths. The resulting epistemic fragility can produce cascading failures that no individual algorithm could trigger alone. The antidote is epistemic diversity maintenance: deliberate architectural, objective, and temporal heterogeneity designed to prevent the very convergence that optimization pressure encourages.