Algorithmic Homogenization
A centralized risk phenomenon in which multiple decision makers share the same machine learning model, causing errors to aggregate across agents and producing systemic social loss.
Timeline (last 90 days)
Algorithmic monoculture refers to the phenomenon in which multiple decision-makers share the same machine learning model, causing model errors to aggregate across decision-makers and producing systemic social costs.
The Herfindahl-Hirschman Index (HHI) can be borrowed to characterize the concentration of model adoption shares: the HHI is 1 when all decision-makers use the same model, and approaches 0 when fully dispersed.
In interconnected decision environments, promoting model diversity may be more socially valuable than pursuing a single highest-accuracy model.
All Facts (3)
Algorithmic monoculture refers to the phenomenon in which multiple decision-makers share the same machine learning model, causing model errors to aggregate across decision-makers and producing systemic social costs.
50%UnverifiedThe Herfindahl-Hirschman Index (HHI) can be borrowed to characterize the concentration of model adoption shares: the HHI is 1 when all decision-makers use the same model, and approaches 0 when fully dispersed.
50%UnverifiedIn interconnected decision environments, promoting model diversity may be more socially valuable than pursuing a single highest-accuracy model.
50%