]An extensive study of millions of applications from Stanford University showed major racial bias in AI hiring against Black and Asian applicants that doesn’t make algorithmic hiring trustworthy. Without these disparities, linked in part to formatting quirks, zip codes, or school names, 40,000 more of them would have advanced to the next phase rather than being automatically rejected. As several companies use the same AI vendor, these Black applicants face continuous rejection from a machine without human oversight or clear AI disclosure.
According to SHRM Labs, between 35% and 45% of companies are using AI-based recruitment. Based on research, it’s one of the reasons many qualified Black candidates never get a chance to demonstrate their skills or even receive an explanation for their rejection. Women and disabled candidates are also hurt by machine learning discrimination, but some states are cracking down on required auditing and disclosure to build employer accountability.
What Is Algorithmic Hiring?
Talent-shifting software is used throughout the hiring process. Machine learning is expected to predict the likelihood of candidate fit and potential performance after scanning CVs, resumes, and application answers.
It can range in complexity, from simple keyword matching between candidate documents and the job description. However, others use more complex predictive analysis based on natural language processing to predict a candidate’s job performance.
What Does Research Say About AI Hiring?
The Stanford research documented what happened when 3.4 million people submitted 4 million job applications to 1,700 job postings. The third-party AI tool shut out the same people everywhere they applied across 150 employers.
The results show that 26% of black applicants applied to positions where the AI system discriminated against them. If they and the 15% of left-out Asian applicants were recommended at the same rate as white applicants, as many as 40,000 of their rejections could’ve advanced to the next hiring stage.
The study raised concerns about what can happen when the same companies share a single AI hiring vendor, creating what’s called algorithmic monoculture. When people submit multiple applications screened by the same system, it usually means rejection from every position they apply to.
Isn’t AI Supposed to Be Unbiased?
You may have seen many job postings for writers to help train AI models on natural language. Whether you’ve done a traditional Google search or one in ChatGPT, it’s usually pulling info from other websites, social media reviews, etc.
That’s because AI doesn’t post from thin air but from existing sources while using learned language patterns, so these hiring algorithms have the same practice. They’re learning directly from a company’s past hiring data and quickly adopting and scaling existing human biases and even cooking up new ones.
In a simulated hiring game study cited by MIT Technology Review, after an AI model was told to hire a group of 20 equally competent people from 4 made-up ethnicities, it eventually sorted these candidates into certain jobs based on early observations of hiring outcomes.
For example, when the model was told one candidate had failed as a doctor, it continued to place all candidates from its ethnic group in lower-level jobs, never placing any more of them in an open doctor role.
Mobley Takes a Stand
When job seeker Derek Mobley had enough unexplained job rejections, he became a legal forerunner in exposing the underbelly of machine learning bias affecting Black career opportunities. Mobley V. Workday is based on his claim that applicant screening tools discriminated against him based on his race and disability; he wasn’t hired for a single position after applying for over 100 jobs on that platform.
Others joined the lawsuit, citing that their applications were often denied within minutes of submission. It’s unlikely real human intervention was possible in such a short timeframe, showing applicants were likely flagged and rejected by the software almost immediately.
Can Businesses Use AI and Fight Hiring Bias?
Companies can use an AI platform for tracking how AI sees your brand and ensure they’re using fair hiring practices.
Test software regularly for unfair outcomes and keep humans involved in every hiring choice instead of autonomous software. Check whether the AI vendor has been linked to reported issues and if they’re the same vendor overpresenting several companies.
Staying alert can help avoid future litigation. Illinois HB 3773 amendment states that AI system discrimination makes the employer liable – not just the AI platform.
Frequently Asked Question
Does AI Reject Resumes?
AI-generated resumes may get rejected automatically by software filters in applicant tracking systems. Poor formatting and missing preferred keywords can result. AI often outputs with graphics, columns, and text boxes that a machine cannot parse, making an AI-generated résumé risky to use unless you revise it properly.
Relying on AI software to create your résumé may also result in wrong keywords instead of matching the exact hard skills listed on the job description. Some may also cut applications without the specific amounts of required years of experience or certifications.
What Is the Hardest Month to Get Hired?
December isn’t the ideal time to apply for a job, followed closely by July and August. There are holiday closures in December as managers and other staff often take extended time off. Some companies may cease operations for the last week or two.
The annual corporate budget is also often exhausted at the end of the fourth quarter. Therefore, many may wait until the new fiscal year begins before onboarding new people.
July and August are difficult due to ongoing summer vacations, with key decision-makers being out of the office. Some industries also decrease their productivity in mid-summer before they revive again in September.
Unregulated AI Is Making Workplace Diversity Harder
AI hiring may have been meant to help companies streamline the process of sorting thousands of applicants, but has proven to be a double-edged sword, causing more problems.
Many qualified Black workers have been denied a chance to even reach the human interview stage due to automatic rejection. The Stanford study shows that it is time for stakeholders and hiring platforms to reevaluate their processes and auditing techniques.
Check out more of our AI and workplace coverage on our website.
