MIT researchers Brian Hedden and Manish Raghavan conclude that algorithmic monoculture in hiring is not uniformly harmful. In a paper published in Philosophical Perspectives, they tested objections including systematic exclusion and found them unconvincing or dependent on details such as domain and accuracy. Their modeling shows that a single ensemble algorithm can match or outperform a polyculture where firms use different tools. For business, attention shifts from whether vendors converge to how well the shared system is built.
What MIT researchers tested in hiring models
The authors modeled several hiring situations to test the fear that a candidate rejected by one screening algorithm would be rejected everywhere. They conclude the total number of hires does not fall, because all open roles are still filled. Firms then compete for the same pool of highly ranked candidates, which according to Raghavan can strengthen candidates bargaining power and lift wages. They also tested objections linked to agency, such as the inability to revise a resume shared across firms. According to Hedden, that concern fades when candidates may revise and resubmit materials.
The central limitation the authors prove mathematically is informational homogenization rather than exclusion. Drawing on the wisdom of crowds theory, they argue that diverse independent decision makers can produce a stronger set of hires than a single decision maker. A shared algorithm can repeatedly select candidates with the same characteristics and miss better alternatives, creating an echo chamber that hinders exploration. Performance then depends on accuracy: one precise algorithm can beat several weaker ones. In simulations, an ensemble that averages scores from multiple hiring algorithms sometimes outperformed separate tools, while added randomness could support discovery.
Monoculture in decisions predates current AI hiring tools. Lending once depended on independent bankers at individual banks and now relies on standardized credit scores derived from the Fair Isaac Corporation (FICO) algorithm. In hiring, a handful of resume screening algorithms are already common across many Fortune 500 companies. Raghavan notes that wider use of AI for information and decisions creates more channels for such correlation. The authors stress that hiring and lending may behave differently from generative AI content creation or AI-guided scientific research, where monoculture could inhibit discovery in science, art or writing and prove more problematic.
What this means for companies using hiring AI
For employers, the finding changes how to assess a market where competitors use the same screening vendor. Convergence does not by itself mean fewer people will be hired, but it concentrates competition on candidates the model ranks highly. Large employers with many openings may feel this as faster bidding for a narrow shortlist, while small firms without brand advantage may need clearer compensation or faster offers to stay in contention. An ensemble approach that combines several signals into one score could offer a practical path to keep quality stable without maintaining entirely separate systems.
The paper does not give approval to every shared system. Outcomes hinge on whether candidates can revise and resubmit, whether the workflow allows exploration instead of repeating the same profile, and whether accuracy has been measured against alternatives. Gaming also needs attention: if a certain resume format scores better, candidates may adapt, though Hedden notes it is unclear whether one shared algorithm invites more gaming than targeting a few firms in a diverse market. The authors add that the feasibility of ensembling in real hiring operations remains unexplored and calls for empirical study.
The marker to watch is whether vendors and employers start testing ensemble scoring and publishing comparative results. If pilots show stable or improved hire quality alongside data on revision rates and wage dynamics, the MIT framing will gain support. If shared systems without randomness or accuracy checks narrow discovery, the risks the authors flag for creative and scientific work will apply to hiring as well.
