In a recent article published by the World Economic Forum (WEF), Keith E. Sonderling, commissioner of the U.S. Equal Employment Opportunity Commission, raises questions about the use of artificial intelligence (AI) as companies seek to grow their post-pandemic workforces.
In what he calls the “Great Rehiring,” many organizations − particularly those who needed to temporarily downsize during the first year of the pandemic − will be under pressure to staff up quickly to meet booming demand spurred by the economic recovery. And, as happens with all economic downturns, companies have invested in automation to reduce their reliance on human labor. In the case of the Great Rehiring, one area of automation that comes to the forefront is AI-based recruiting, screening and interviewing systems.
At face value, these systems would seem to be an enormous benefit to companies. Every business owner, at one time or another, has said “I wish I could clone that person” when they have an employee who does excellent work. With AI, “cloning” existing employees seems achievable – simply search and screen people who demonstrate similar backgrounds and character traits as the best existing employees … and voila! Clones!
But it’s actually not that simple. In fact, in an experiment similar to our cloning fantasy, Amazon once trialed an AI resume-screening system where the source data was the resumes of its employees, along with all the resumes that had been submitted to the company over the past 10 years. Using this data, the system was able to detect patterns in resumes that indicate whether or not the person will become a productive employee.
This seems like a reasonable way to find new candidates who would be a good fit for the company. There was only one problem – an outsized majority of resumes from the past 10 years were from men, so the system concluded that men are better candidates than women and rated resumes accordingly. Oops!
This is an excellent example of how demographic trends may render AI systems obsolete before they’re even turned on − particularly if they rely too heavily on historical data. For example, many organizations today have strategies and programs in place around diversity and inclusion. In some cases, this is an existential undertaking − as a case in point, there are millions of open jobs in information technology and cybersecurity, yet women make up only 20-30% of the workforce in those fields. There is no way for companies to even begin to fill those jobs unless more women are brought into those careers.
An AI screening system that does not have this context, and that is basing its screening scores solely on historical data, may actually homogenize the organization rather than diversify it. In the case of IT jobs, the system could determine that men make better employees than women (like the Amazon system did), which would run counter to the critical goal of bringing more women and overall diversity into the profession.
Context will be especially important during the Great Rehiring, as companies rely on automated screening tools. The WEF and U.S. Department of Labor have both found that women and people of color were disproportionately impacted by the pandemic − therefore they likely will be represented disproportionately among the population of workers re-entering the workforce during the recovery.
Employers using automated recruiting and screening tools need to hold their vendors’ feet to the fire to prove their AI is not skewed against these or other demographic groups. The AI software needs to be modeled on datasets where the bias can be reduced / removed entirely to provide outputs that reflect the real world. If they don’t, they may find that AI-driven automation doesn’t just speed the pace of processing candidates; it also speeds the pace and severity of violating employment laws.
But this is an ideal opportunity in which humans working alongside AI will result in the most desired solution. Automation tools, like AI, don’t need to reduce oversight to be effective and efficient – they can be designed with these considerations in mind to ensure proper checks and balances. With automated hiring and screening processes, it would make the most sense to have humans continually looped into the process to infer context, since that’s what we are good at. AI is great at finding patterns, while humans are great at making decisions around them. Pairing the two will make for the ultimate “Great Rehiring” process.