AI Bias in Hiring: How Algorithms Perpetuate Discrimination

The promise of artificial intelligence in hiring is compelling: efficiently sift through mountains of applications and identify the most qualified candidates. But, a growing body of research reveals a troubling reality – AI-powered resume screening tools are often riddled with bias, potentially exacerbating existing inequalities in the job market. While intended to be objective, these algorithms can perpetuate and even amplify discrimination based on race, gender and other protected characteristics.

The Persistence of Bias in AI Hiring Tools

Recent studies from the University of Washington demonstrate that even state-of-the-art large language models (LLMs) exhibit significant bias in how they rank resumes. Researchers tested three leading AI models across over 554 real resumes and 571 job descriptions spanning nine different occupations. The findings were stark: the models consistently favored candidates with names associated with White men.

In direct comparisons, resumes with White male names were preferred in 100% of 27 bias tests against those with Black male names. University of Washington research also showed that White-associated names were favored 85% of the time, while female-associated names received preference only 11% of the time. Black male names were never favored over White male names.

Interestingly, the bias wasn’t tied to specific job types. Even in roles traditionally held by women, like Human Resources, the models still showed a preference for male names. This suggests a systemic bias embedded within the algorithms themselves, rather than a reflection of real-world workforce demographics.

The Impact of Resume Length on Bias

The University of Washington research also revealed that bias intensifies when resumes are shorter. When resumes were reduced to just a name and job title, biased outcomes increased by 22.2%. This is particularly concerning for entry-level candidates, career changers, or those re-entering the workforce, who may have less extensive professional histories to differentiate themselves. StudyFinds highlights that these populations are most vulnerable to discrimination based on name alone.

The “Human in the Loop” Illusion

Many organizations rely on human reviewers to mitigate potential bias in AI-driven hiring processes. However, a second study from the University of Washington, involving 528 participants and 1,526 resume-screening scenarios, casts doubt on the effectiveness of this approach. The study found that human reviewers overwhelmingly followed AI recommendations, even when they expressed skepticism about the AI’s accuracy.

When the AI favored a particular racial group, human reviewers followed that preference up to 90% of the time, particularly for high-status jobs. Crucially, exposure to biased AI recommendations altered what reviewers considered a “qualified” candidate, effectively absorbing the AI’s bias as their own. KUOW reported that this pattern held even when participants didn’t trust the AI, shifting their decisions by nearly 50 percentage points in some cases.

What Can Be Done to Mitigate AI Bias in Hiring?

Addressing AI bias in hiring requires a multi-faceted approach:

• Independent Audits: Regular, structured audits of screening outcomes, broken down by demographic group, are essential to identify and address hidden biases.
• Unconscious Bias Training: Unconscious bias training before the screening process, rather than as a one-time compliance exercise, can help reviewers resist absorbing AI-driven biases. Research suggests that engaging with implicit association tests (IATs) beforehand can increase the selection of candidates the AI was biased against by 13%.
• Skills-Based Assessments: Shifting the focus from resume-first screening to skills-based assessments, structured work samples, and blind application reviews can reduce the influence of demographic signals.
• Question the Default: Organizations should critically evaluate whether resume-first screening is the most appropriate method for every role.

The question isn’t whether your AI is biased – it is. The critical question is whether your organization has the mechanisms in place to challenge that bias and ensure a fair and equitable hiring process.