{"text":[[{"start":6.8,"text":"AI hiring tools used by large employers are repeatedly screening out some of the same job applicants, with researchers finding that Black and Asian candidates are disproportionately affected. "}],[{"start":19.15,"text":"A Stanford-led study of 4mn job applications across 156 employers who used the Pymetrics hiring platform, which assesses people using a series of online games, found evidence of “systemic rejection” linked to the algorithms it used."}],[{"start":35.55,"text":"The research showed that jobseekers would need to apply for at least 25 different positions to be almost certain of receiving at least one recommendation to proceed to the next stage of an application."}],[{"start":46.75,"text":"The study is the largest examination of AI hiring algorithms to date, and adds to growing concerns that widely used automated recruitment tools risk embedding bias across employers. "}],[{"start":58.95,"text":"Game-based assessments have become popular with employers, who increasingly rely on companies such as Pymetrics or HireVue to screen the large volumes of applications they receive. Jobseekers have complained they spend hours completing tests with little prospect of their application receiving human scrutiny. "}],[{"start":76.4,"text":"“As a single vendor comes to dominate decision-making in a space, their quirks or shortfalls can be present across that entire sector in a way that wasn’t possible before,” said Kathleen Creel, a co-author of the study and assistant professor of philosophy and computer science at Northeastern University. "}],[{"start":96.35000000000001,"text":"Pymetrics owner Harver did not respond to a request for comment."}],[{"start":100.65,"text":"The study, which was led by the Stanford Institute for Human-Centered AI, looked at 4mn job applications submitted via Pymetrics between December 2018 and December 2022. The dataset spanned 156 employers, the majority of which had annual revenues of $5bn or more. "}],[{"start":119.95,"text":"The research found there were “clear racial disparities” in outcomes. Analysing individual roles, they found that one in 10 of the positions in the dataset demonstrated “adverse impact” against Black applicants, while one in 20 roles did so for Asian applicants. "}],[{"start":136.8,"text":"Adverse impact is a term used by US federal agencies to describe a selection rate for any race, sex or ethnic group that is less than four-fifths of that of the most selected grouping."}],[{"start":147.5,"text":"A previous study by researchers from the University of California, Berkeley and the University of Chicago found that “distinctively Black names” reduced the probability of employer contact by 2.1 percentage points relative to “distinctively white names”."}],[{"start":163.95,"text":"The Stanford-led research also found that several employers were, for some roles, using identical algorithmic models to screen candidates. The researchers identified 42 models that were “shared across” different employers, meaning candidates who had been rejected from one company were likely to fail at others using that model. However, the data showed few candidates had been caught by the issue. "}],[{"start":185.54999999999998,"text":"Four per cent of applicants who applied for 10 roles were recommended for rejection across all of them by the platform’s algorithm, a higher rate than would be expected by chance."}],[{"start":194.89999999999998,"text":"“When applying to two positions at two different employers, applicants might reasonably expect that they are receiving two separate evaluations and therefore two chances,” the researchers wrote. “But if both positions share the same model, their numerical score will be identical.”"}],[{"start":210.24999999999997,"text":"The algorithms used by Pymetrics assess traits such as propensity for risk and speed of response, as well as characteristics such as trust and care for others. Applicants whose performance most closely matches that of top employees are recommended to advance, while others are rejected."}],[{"start":228.49999999999997,"text":"The study’s researchers cautioned that their findings may not generalise to all algorithmic screening, noting that the game-based approach of Pymetrics may differ from other approaches such as résumé screening."}],[{"start":null,"text":"