This simple AI error may be keeping you from landing your dream job

While many share understandable doubts, reservations, and fears, there’s no denying that we all find ourselves living in the age of AI nowadays. Predictions of AI taking over virtually all industries and aspects of everyday life have persisted for decades upon decades, but recent years have seen many of those at-one-time farfetched premonitions materialize into full-fledged realities. 

From automated personal assistants to content creation, there’s no shortage of AI programs available today that claim to make working life infinitely easier for the humans behind the screens. Of course, just like any other form of emerging technology, a considerable number of bugs and issues remain that must be rectified before such algorithms and programs can be legitimately trusted and relied upon on a regular basis.

Case in point, applicants currently on the lookout for a new job or professional beginning are going to want to know about poignant and relevant new research just released by the University of Florida regarding the use of AI algorithms to screen potential job candidates. Read on to learn more about a frustratingly common AI error that may be keeping countless qualified candidates from landing a great new job.

Applicant tracking systems, AI, and automated candidate screening

One of the most widespread and ubiquitously adopted forms of automation sweeping business practices today are applicant tracking systems, used by countless HR departments to help sort through myriad job applicants and pick out the most desirable candidates. Recent research even tells us that close to all Fortune 500 companies now use at least one form of ATS to parse through applicants long before a human pair of eyes ever land on a single resume. 

Applicant tracking systems, as well as other related AI-centric candidate screening tools, are intended to assist employers in selecting the best applicants for an open position as fast and efficiently as possible. This is usually accomplished by such programs via searching for specific keywords, identifying knock-out questions or details, or grading trial task performances. 

Troublingly, however, the aforementioned new piece of research released by UF and published in the scientific journal Production and Operations Management suggests that in many scenarios these automated hiring tools may be doing more harm than good. More specifically, researchers point to a fundamental misunderstanding between what human HR managers and recruiters think they’re getting with AI assistance, and what the algorithms actually consider to be their primary objective.

Man, machine, and misunderstandings

According to the report, while AI screening technologies are programmed to pick job candidates to hire, the vast majority of hiring managers only want their automated tools to assist in the decision-making process by providing a short-list of promising candidates to interview. Virtually no human HR managers or corporate decision makers are relying solely on AI to make personnel decisions, but these programs aren’t privy to that information.

Put another way, instead of providing a list of a dozen or so promising job candidates featuring an eclectic variety of professional backgrounds and professional skill sets to move forward in the interviewing process, AI screening tools instead simply produce lists of the absolute safest candidates possible.

“When we ask these algorithms to select the 10 best resumes, we know we are not directly hiring these 10 people. We know there is a second stage of interviewing, but the AI doesn’t know that,” says Heng Xu, a professor of management in the University of Florida’s Warrington College of Business and lead author of the new study, in a media release. 

“If you don’t specify that there are additional steps, the AI system might select 10 candidates that are good, but safe, choices. But if you tell the AI system there will be another round of screening by people, it might different, and potentially stronger, candidates,” Prof. Xu adds.

Selection vs. screening

The research team explains that this potentially major misunderstanding between man and machine largely comes down to the distinction between screening for candidates versus actually selecting the right person for the job. Human managers just want AI to help with screening applicants, while most algorithms are designed to select the best candidates according to a stringent set of predetermined rules, keywords, etc.

For example, in any collection of resumes there are going to be stronger and weaker applicants. If it is under the impression it is selecting the absolute best people for the job, an algorithm will simply reject all candidates deemed risky. But, if the algorithm is instead explicitly asked to screen initial applicants and produce a list of potential hires to move forward with interviews or job assessments, the program can then provide a more comprehensive collection of candidates featuring a few more high-risk high-reward options, allowing human decision makers to take the hiring reins from there.

Moreover, just like an annoyed HR manager reading through resumes in a bad mood, the study also notes AI screening tools are by no means immune to introducing their own biases into the hiring process.

Crafting a solution

So, in pursuit of a solution to this glaring issue facing countless employers and job seekers alike, researchers at UF set out to put together a new algorithm of their own better suited to address this misunderstanding. To that end, they focused on a distinct type of algorithm intended to reduce biases in the hiring process (fairness-aware algorithms).

Study authors modified a pre-existing fairness-aware algorithm to ensure the program understood its role was to screen candidates, not select or hire applicants. Sure enough, a series of subsequent tests involving data pertaining to nearly 8,000 employees from a Fortune 500 company showed the modified algorithm was indeed able to identify stronger candidates, avoid bias, and lower hiring costs (saved 11% of human interviewing expenses).

“We’re trying to send the message that we have to rework the language we use for these human-AI teams, otherwise the AI may optimize on poorly specified goals,” Prof. Xu concludes. “Our point is, sometimes we need to take responsibility, because we didn’t specify the task clearly.”