How job search sites such as LinkedIn, ZipRecruiter, and CareerBuilder are working to remove gender and racial bias from their AI-powered matching algorithms
Sheridan Wall / MIT Technology Review : Tweets: @docstefflbauer Tweets: Doc Stefflbauer / @docstefflbauer : Isn't geographical data one way they block people in Paris suburbs from interviews? Isn't it so bad that whole of the suburb-dwelling population is included in “socially disadvantaged” categories (as a proxy for race)? Maybe this is only in France. Surely the US is different 🙃 https://twitter.com/...
Context & Ripple Effects
The industry has been here before: Amazon scrapped an internal machine-learning tool in 2017 after it learned to downgrade female candidates — a failure that became the cautionary tale for algorithmic hiring — and researchers later found Facebook's ad delivery disproportionately steering certain job ads by gender, with no such skew detected on LinkedIn's job ads. Against that backdrop, MIT Technology Review reports LinkedIn, ZipRecruiter, and CareerBuilder are now working to strip gender and racial bias from their AI-powered candidate-matching algorithms.
The timing is not incidental: regulators are moving from exposés to enforcement, with New York City passing a bill requiring a yearly bias audit before employers can deploy AI hiring tools at all. The debate also reaches back further than gender — as far back as 2016, reporting flagged how application-sifting algorithms tended to screen out the poor, and readers in the piece raise geographic proxies for race in France's Paris suburbs as the same problem wearing different features.
First-order effects
- Job seekers on LinkedIn, ZipRecruiter, and CareerBuilder face matching systems being actively retrained to stop encoding gender and racial proxies into who gets surfaced to recruiters.
Second-order effects
- Employers using these platforms gain a compliance argument ahead of mandates like New York City's yearly bias-audit requirement — audited, de-biased matching becomes a selling point rather than just a cost center.
Third-order effects
- If the pattern holds, bias auditing moves from scandal response (Amazon's abandoned résumé-rating tool) to standard procurement criteria, structurally favoring platforms that can document fairness over smaller tools that cannot.
The trend: Algorithmic hiring is shifting from unexamined automation toward audited, fairness-tested matching, pushed by documented failures like Amazon's scrapped tool and regulatory moves like New York City's audit requirement.