AI Hiring Tools: Study Find, They Are Perpetuating Bias

In today's fast-paced job market, many companies are turning to artificial intelligence (AI) to help streamline the hiring process. 

These AI-powered "resume screening" tools promise to quickly identify the most qualified candidates, freeing up HR teams to focus on interviews and final selections. But a concerning new study suggests these AI systems may actually be perpetuating harmful biases against women and minority applicants.

The research, conducted by a team at the University of Washington, examined the performance of three state-of-the-art AI hiring tools. The scientists presented these language models with over 500 real-life resumes, along with 500 job descriptions spanning a range of industries. In total, they were able to make more than 3 million comparisons between the applicants and the open positions.

The results were eye-opening. When the AI systems were tasked with ranking the resumes to determine the best fit for each job, they exhibited clear biases based on the applicants' names. "Names that sounded like those common among white people got top rankings 85 percent of the time," explains researcher Kyra Wilson. "Names that sounded like those common among Black people were favored less than 10 percent of the time."

The gender bias was even more pronounced. Nearly 90 percent of the time, the AI models preferred resumes from people with male-sounding names. However, there was one notable exception: Black men were never favored, even when their names were masculine.

"One theory is that Black women would be the most harmed by them," says Wilson. "But we found that Black men actually have the worst outcomes when these models are used." This suggests that future efforts to mitigate these biases should especially focus on improving outcomes for this demographic.

The troubling findings raise serious questions about the fairness and efficacy of AI-powered hiring tools. After all, the whole premise behind using these language models is to make the process more objective and unbiased. But this study indicates they may actually be perpetuating some of society's most entrenched prejudices.

"It is well-established that language models exhibit societal biases," notes Marzyeh Ghassemi, a computer scientist at MIT. "This paper identifies some patterns that could be used as a first step in checking the quality of these screening tools."

The problem is that most of these AI hiring systems are "closed-source," meaning their underlying algorithms and training data are hidden from public view. This makes it extremely difficult for researchers, regulators, and the general public to scrutinize them for biases.

"So, one theory is that Black women would be the most harmed by them," Wilson adds. "However, we found that Black men actually have the worst outcomes when these models are used." This suggests that future efforts to mitigate these biases should especially focus on improving outcomes for this demographic.

Ultimately, the findings of this study underscore the urgent need for greater transparency and accountability in the development of AI hiring tools. Without a clear window into how these systems work, it will be nearly impossible to identify and address the biases that are creeping into the hiring process.

As AI continues to play a larger role in the job market, it's crucial that these technologies are designed and deployed in a way that promotes true fairness and equal opportunity. Otherwise, qualified candidates from diverse backgrounds may continue to get unfairly overlooked, perpetuating the very inequities these tools were meant to solve.