Algorithmic Monocultures in Hiring

A new study of 3 million job applicants finds that widespread use of hiring algorithms from the same vendor produces racially disparate outcomes, with Black and Asian applicants disproportionately steered toward rejection, according to research posted on arXiv [1]. The paper, titled "Algorithmic Monocultures in Hiring," analyzed 4 million applications screened by tools from a single vendor [1]. It found that 14.74 percent of applications from Asian applicants and 25.87 percent from Black applicants were submitted to positions that adversely impacted those groups under U.S. employment discrimination standards [1]. The authors argue that when many employers rely on identical screening software, the same candidates face repeated rejection across firms [2]. The concept of monoculture in computing describes a community of systems running identical software, which share the same vulnerabilities and can fail catastrophically [3]. The researchers applied this lens to hiring, finding that individual outcomes were strikingly uniform: 4 percent of applicants who applied to 10 positions were recommended for rejection from all of them, a rate higher than expected by chance [1]. To probe the mechanism, the team exploited the deterministic nature of the algorithms, simulating what would happen if each applicant had applied to every available position [1]. The results indicated that candidates would need to submit applications widely to have a meaningful chance of being evaluated by a human reviewer rather than an automated filter [1]. The dataset did not include direct employer decisions, only the algorithmic recommendations that precede them [2]. The findings arrive as regulators and lawmakers scrutinize automated employment tools. The U.S. Equal Employment Opportunity Commission has issued guidance on algorithmic fairness, and several states have proposed or enacted laws requiring audits of hiring technology [2]. The paper does not name the vendor whose software produced the data, but the authors note that a small number of firms supply screening algorithms to a large share of major employers, amplifying the effects of any embedded bias [2].