MIT researchers Manish Raghavan and Brian Hedden systematically evaluated major objections to algorithmic monoculture — the use of one algorithm for all decisions in a domain — in a paper reported Sept. 29 in Philosophical Perspectives. They concluded that objections including systematic exclusion, agency concerns and gaming either fail or are not decisive against all forms of monoculture.

The researchers instead prove mathematically that monoculture tends to create informational echo chambers that can hinder exploration. In hiring, they write, this could make it less likely the best candidates get jobs, though bundling hiring algorithms into a single ensemble could overcome this limitation and sometimes perform as well as, or better than, a polyculture of differing algorithms.

They note findings are contextual, that domains such as generative content creation or AI-guided science may differ, and that real-world feasibility of ensembling remains unexplored. They hope the work spurs further empirical study of long-term consequences.