MIT study examines the trade-offs of shared hiring algorithms
MIT researchers have found that an “algorithmic monoculture” in hiring — multiple firms relying on the same decision-making system — is not inevitably harmful to job seekers. Their study, published in Philosophical Perspectives, evaluates objections to shared algorithms and shows that the consequences depend on the setting and on the quality of the system used.
The research was conducted by Brian Hedden, a professor in MIT’s Department of Linguistics and Philosophy and the Department of Electrical Engineering and Computer Science, and Manish Raghavan of the MIT Sloan School of Management and EECS. They examined hiring models, while noting that the framework may also apply to fields such as lending.
Shared screening does not automatically mean exclusion
A common concern is that an applicant rejected by one company’s screening algorithm would be rejected everywhere if all employers used the same tool. The researchers argue that this does not by itself establish systematic exclusion: the overall number of jobs filled is unchanged simply because firms use one algorithm rather than several.
In their models, shared screening can also strengthen candidates’ bargaining position. When employers compete for the same selected pool of applicants, competition may drive wages upward. That conclusion does not mean every version of monoculture is desirable; it shows that the usual objection requires more detail about how the hiring process operates.
The study also considers agency. A system that forwards a candidate’s application to every participating employer could deny that person an opportunity to revise a résumé after an unsuccessful outcome. But this objection does not apply where the platform allows candidates to update and resubmit their materials. Likewise, applicants may try to optimize a résumé for a known system, but the researchers say it is not obvious that this incentive is stronger under one algorithm than under several competing ones.
The larger concern is reduced exploration
Hedden and Raghavan identify informational homogenization as a more fundamental issue. Diverse and independent decision makers can generate the “wisdom of crowds,” improving the chances that firms identify strong candidates. If every firm receives the same recommendations, candidates with similar profiles may repeatedly be chosen while potentially better alternatives are not discovered.
The researchers mathematically show that monoculture tends to form informational echo chambers that can hinder exploration. In hiring, that can make it less likely that the best candidates obtain positions. The concern may be especially relevant in areas such as science, art, and writing, where the value of discovering alternatives can be particularly important.
Accuracy and algorithm design shape the result
Monoculture can perform differently when a shared algorithm is substantially more accurate than the separate algorithms individual firms would otherwise use. The researchers also tested an ensemble approach: multiple hiring algorithms are packaged into one system that scores candidates using an average. Across simulations of hiring situations, such an ensemble could sometimes perform as well as, or better than, a market in which firms use different algorithms.
Randomness in a shared platform could also encourage greater exploration. However, the practical feasibility of algorithmic ensembling remains an open question, and the authors call for more empirical work on the complexities of real job markets. For businesses deploying automated hiring, the implication is to assess accuracy, candidate revision pathways, and exploration mechanisms together rather than assuming either uniformity or diversity of algorithms will produce the better outcome.