Algorithmic monoculture and social welfare

As algorithms are increasingly applied to screen applicants for high-stakes decisions in employment, lending, and other domains, concerns have been raised about the effects of algorithmic monoculture, which many decision-makers all rely on same algorithm. This concern invokes analogies agriculture, where a monocultural system runs risk severe harm from unexpected shocks. Here we show that dangers monoculture run much deeper, convergence single algorithm by group decision-making agents, even when is more accurate any one agent isolation, can reduce overall quality being made full collection agents. Unexpected shocks therefore not needed expose risks monoculture; it hurt accuracy under "normal" operations, used only decision-maker. Our results minimal assumptions, involve development probabilistic framework analyzing systems use multiple noisy estimates set alternatives.