I am biased, as are you. Right or wrong, we all have our individual biases, because we’re human, and that’s how we’re wired. Given this, what is the likelihood that we, here in 2019, could program an artificial intelligence (AI) to be completely bias-free? And if we are unable to do this, would it be prudent to pump the brakes on driving headlong into letting AIs dominate our hiring decisions?
To be clear, I’m not saying we should remove AI from hiring — far from it. But I do believe we’re in the midst of an AI infatuation that, in the rush to remove humans from the hard work of applicant filtering, may inject more bias than it removes.
AI: Accelerated Inequity
A few months ago, Amazon, one of the world’s top AI powerhouses, revealed just how long those odds could be when it scrapped its AI-based hiring software project. According to Reuters, Amazon’s project went live in 2015. A year later, Amazon figured out that the software had taught itself to discriminate against women based on the preponderance of male applicants over the preceding 10 years. Apparently, the cost and/or complexity of the problem was more than Amazon was willing to bear, as the company powered down the project in 2017.
My years as Symantec’s chief human resources officer continue to make clear the need for better tools in hiring, especially in the top talent tiers. Enterprises regularly receive hundreds, even thousands, of applications for a single posting, in part because submission tools can now reduce the application process to check-and-send simplicity. No HR department has the bandwidth to thoroughly vet every application manually. The impetus to employ AI in this role keeps climbing, specifically because it’s getting harder to identify top candidates within every opening.
Thus, we have what might appear to be an insoluble dilemma: We need AI in hiring. The ethical and financial positives tied to promoting diversity and inclusion in that hiring are amply documented, and yet AI algorithms seem bound to magnify the biases we build into their datasets because of our inherent biases.
Perhaps the lesson to take away from Amazon’s effort is that diversity and inclusion must be accounted for at the outset. Once bias is baked into an AI’s foundations, it becomes extremely difficult to remove.
I’m not implying that it’s too soon to begin using AI in hiring. For example, my company uses AI that analyzes job descriptions and reports back a score indicating the level of gender-neutralization in a listing. The more gender-neutral we can make job descriptions, the less likely we are to lose good applicants based on subtle gender prejudices.
Do such AI measures work? Honestly, it’s too soon to know. We’re in the middle of an experiment. When we have two years of data in hand, we should be able to look back and see if the tools delivered the desired benefits. And therein lies my advice for others interested in bringing AI into their hiring practices: This is not an all-or-nothing proposition. Adopt with measured, incremental prudence.
1. Start with one AI tool in one business group. As with other software tools, start small, and roll out more broadly once the technology proves itself within your company.
2. Build your data set from real-world data, not some idealized, artificial construct of the “ideal” workforce. Nothing is ever going to be ideal, and trying to make it so will inherently introduce exclusion biases.
3. Establish look-back checkpoints after deployment, perhaps at six-, 12- and 24-month marks. Judge whether your AI tool is delivering intended hiring benefits — or is at least moving the needle closer to your target(s) relative to your pre-AI baseline.
4. Always be adjusting. Just as an airliner spends most of its journey correcting course en route to its destination, you’re going to use those checkpoint observations to tweak your AI’s algorithms and decision weighting.
Essential Human Elements
I’ll stick my neck out just a bit farther: I do believe that AI can be a valuable tool for HR in streamlining the early parts of the hiring process. I do not believe that AI can replace humans in the later stages of hiring, at least not in the foreseeable future, because we are a long, long way from knowing how to program an AI to have a “human connection” to an applicant, that undefinable element that differentiates one candidate from another during the hiring process’s last stages. It’s how that person meets your eye and interacts, the tone of their voice, the sense you get of their drive and commitment. We have no clue how to make algorithms around these qualities; we have a hard enough time as it is divesting our AI algorithms from gender, race and other elements of diversity.
Harken back to when Thomas Friedman of The New York Times revisited an interview with then-SVP of people operations at Google, Lazslo Bock. Bock described two of the biggest checkbox items — GPAs and test scores — as “worthless” and noted the rising number of employees at Google with no college education whatsoever. Rather, Bock valued flexible and synthetic thinking, situation-appropriate leadership and “intellectual humility.” Like drive and commitment, these are also very difficult factors to quantify for an AI algorithm. That’s why we need humans at and near the hiring helm. They see the deep, truly critical human factors that software cannot.
My decades in human resources confirm this. We need to keep the humanity in HR, especially if we’re going to pursue equity in hiring. We need systems that de-emphasize checkboxes and prioritize curiosity, collaboration and mental agility. Companies that fail to recognize this, and that charge into AI without taking care to minimize bias while simultaneously preserving humanity in hiring, will fall astray and be left behind.
Source : https://www.forbes.com/sites/forbeshumanresourcescouncil/2019/05/29/ai-diversity-and-keeping-human-in-human-relations/#73bd35873a8b