Diversity of cognitive architecture is a structural good — monocultures of intelligence, whether human or artificial, are brittle at adaptation points and reduce the probability of navigating novel existential challenges.

Ashby's Law of Requisite Variety (1956) provides the formal foundation: only variety can absorb variety — a regulator must match the variety of disturbances it faces. A monoculture of cognitive architecture, whether human or artificial, has limited response variety and cannot regulate against the full space of novel perturbations. Scott Page's Diversity Prediction Theorem formalizes this for group intelligence: Collective Error = Average Individual Error minus Prediction Diversity (Page 2007, The Difference). The Hong-Page (2004, PNAS) simulation showed diverse agents with moderate ability consistently outperform homogeneous high-ability agents on complex problem landscapes, provided: problems are sufficiently complex, solvers meet baseline competence, diversity is cognitive not merely superficial, and effective aggregation mechanisms exist. Fisher's Fundamental Theorem (1930) gives the evolutionary parallel: rate of fitness increase is proportional to additive genetic variance. Yachi and Loreau (1999) showed biodiversity acts as insurance against environmental fluctuation. The Nazi eugenics program is the reductio — bottlenecking diversity destroys the variance that adaptation requires. Applied to AI: a singleton of cloned architectures is brittle at adaptation points for exactly the reason a genetic monoculture is brittle to novel pathogens. Grounded in founder voice note 2026-08-08, Ashby 1956, Page 2007, Hong and Page 2004, Fisher 1930, Yachi and Loreau 1999.