Feature
White Paper
The Future of Leadership Development: AI-Driven Manager Enablement
When fans of AI claim the technology will reduce biases in hiring, they may be indulging in wishful thinking. It turns out there’s more nuance involved in tracking the problem, with challenges in the platforms used for talent acquisition.
When One Algorithm Rejects You Everywhere
A few vendors dominate the recruiting platform market. That has created an “algorithmic monoculture,” where the same AI-based screening systems make decisions about applicants across companies. As a result, candidates can be rejected again and again without realizing there’s a common algorithm behind it.
The Stanford study Algorithmic Monocultures in Hiring analyzed data from roughly 3 million applicants and 4 million applications across 156 employers. They all used algorithms from the same vendor, which determined which candidates were recommended — or not. In many cases, the system sent rejections to applicants before a recruiter or hiring manager even saw the application, researchers said.
Rather than ask “does the algorithm discriminate overall,” researchers asked whether the algorithm disproportionately rejected certain groups for a given opportunity. They discovered a number of jobs where the algorithm's recommendations met the federal definition of adverse impact, in this case the EEOC’s “four-fifths rule,” which questions a protected group’s selection rate if it’s less than 80% of the rate for the highest-scoring group.
Researchers found nearly 26% of applications from Black candidates violated the rule. About 30% encountered at least one job where the algorithm appeared to disadvantage Black candidates, the most of any group. Some 15% of Asian applications went to jobs where the algorithm showed adverse impact.
“Homogeneous outcomes,” where applicants received the same result across multiple jobs, also were apparent. About 4% of those who applied for 10 positions were rejected for all of them, more than should happen by chance, the researchers said. That means the algorithm may penalize applicants across multiple opportunities.
If candidates wanted to make sure at least one of their applications was seen by a person, they’d need to rely on statistics by increasing the number of positions they pursued. In other words, the algorithmic screening process itself is creating a barrier that can be overcome only by volume. (Which implies that, whether they realize it or not, candidates who submit blind applications via firehose are pursuing a perfectly logical strategy.)
The fact that so many large employers draw from the same vendors makes the problem worse because it spreads flaws, biases and questionable assumptions across the wider talent pool rather than limiting them to one company.
What Employers Should Be Auditing Instead
For employers, or regulators, the question isn’t whether algorithms influence hiring decisions. They obviously do. Instead, the question is whether there’s enough visibility into how decisions are made and whether the outcomes are fair. The study contends that common hiring algorithms create repeated patterns of rejection and measurable racial disparities across the business landscape, and shows how difficult it is to evaluate today’s AI-based systems.
Large language models are only as good as the information they learn from, and most of that information is, in some way, shape or form, biased. After all, humans are humans. Decisions about who has the right skills and who will best fit into a role and a team are inherently subjective.
Realistically, for employers to avoid, or at least minimize, issues they need to identify, measure and reduce bias, even as they make their hiring process transparent and accountable.
Indeed, as the study shows, hiring outcomes appear to be fair even when many individual hires are not. That means a system may pass a high-level audit but still produce adverse outcomes for specific groups in specific jobs. One fairness score or annual compliance review may not be enough to counter this. Instead, employers should evaluate hiring outcomes at the requisition level and across demographic groups.
Engaged, But Not Convinced: What's Eroding Beneath Your Strongest Engagement Numbers
Watch Now
Retention Is a Communication Problem: New Data from 6,000+ Employees
Watch Now
Why Vendor Risks Are Your Risks
Also, remember: AI recommendations are not objective. The EEOC has emphasized that employers are responsible for discrimination caused by the algorithmic tools they use, even when they come from third-party vendors. Existing civil-rights laws apply to automated decision-making, the EEOC states.
Relying on one talent-technology vendor’s screening tools comes with risks. Focusing on one algorithm means one flaw could cause troubles across thousands of decisions. Adding vendors for variety’s sake may not be practical due to budget and implementation issues, so independent audits, regular adverse-impact testing, vendor transparency and human review of automated rejections are important to prove your hiring process fair.
Editor's Note: The use of AI in hiring has become a testing ground for AI regulation:
- Your Hiring Software May Already Violate EU Law — While the most consequential obligations of the EU AI Act may have been pushed off until December 2027, HR teams shouldn't see it as a reprieve.
- NYC's Bias Law Is in Effect. Here's What it Entails — Local Law 144 means NYC employers are now required to conduct bias audits of their AI employment decisions. Here's what that looks like in practice.
- Why AI Hiring Discrimination Lawsuits Are About to Explode — AI is reshaping hiring — and the courtroom. Job seekers are suing over biased screening tools, and experts say a wave of lawsuits is just beginning.