MIT researchers are challenging the assumption that companies sharing one hiring algorithm inevitably make opportunities worse for job seekers. Their analysis finds that the consequences depend on how decisions are organized and which algorithm is used.
Brian Hedden and Manish Raghavan examine these questions in “Algorithmic Monoculture and Its Critics,” published in Philosophical Perspectives on September 25.
One concern is that an applicant rejected by one employer will be rejected everywhere. The researchers argue that this objection is less decisive when hiring happens sequentially and successful applicants leave the available pool.
They identify a stronger concern around exploration. When employers repeatedly favor candidates with similar characteristics, they may overlook people whose abilities could challenge the system’s existing assumptions.
The researchers also examine combining multiple algorithms into a shared ensemble. Their models and simulations show that such combinations can sometimes match or outperform arrangements in which employers use separate algorithms.
That result comes with limits. Practical feasibility remains unresolved, and the research does not demonstrate that a shared system eliminates bias in actual workplaces. It instead gives researchers more specific questions to test before recommending either uniformity or diversity.