Podcast Recap: AI to Fight Hiring Bias

Michelle Hancic, pymetrics’ Global Head of Consulting Psychology, participated in a podcast with the AI Asia-Pacific Institute about how the rise of AI has presented important challenges for recruiting and society at large. This podcast series invites leaders from different industries and creators of new AI to debate the most pressing questions in the space, from the future of work to the future of our humanity. Continue reading below the conversation with Michelle explaining how and why pymetrics was founded, how we audit our algorithms to ensure optimal predictiveness, and the variety of applications of the pymetrics platform that exist today and are in the works for the future. 

(The conversation has been edited and condensed for clarity and length)

‍

APIP:

I thought we could start by discussing your work in pymetrics...and the history of the company, which I believe is a good example of using AI for good.

Michelle:

My role at pymetrics is the head of industrial and organizational psychology, in the APAC region. I’ve been working with pymetrics for the past 12 months. I was actually a client of pymetrics previously, at the ANZ Bank. Really inspired by the work that was being done by pymetrics from the perspective of fairness and prediction in particular, and I have been working [in the assessments space] my whole career, as well as in talent acquisition. Fairness, accuracy and prediction really sit very dear to my heart, and so I was really inspired to join the business and continue the great work.

So for a little bit of background about pymetrics: it was co-founded by Dr. Frida Polli, who is still our CEO, and a very active driver of fairness and prediction and ambition. She started pymetrics in 2012.  She spent about a decade as an academic neuroscientist at Harvard and MIT and she ended up receiving an MBA at Harvard and it was while she was there that she actually came up with the idea for pymetrics because she experienced firsthand how the hiring process is actually quite broken. It wasn't uncommon for her to see her classmates spend six months plus preparing to land their ideal job only to find that their ideal job actually wasn't so ideal after all. She was thinking that if it was happening to Ivy League grads, surely it must be happening elsewhere as well. And sure enough, we know that recruiting is not always the most seamless process in the world. This gave rise to the idea to really leverage her neuroscience background and create a solution that would actually create a fairer, more accurate and predictive, as we like to call it, selection process.

pymetrics has now established a talent matching platform that uses gamified assessments derived from the Behavioral Sciences to build custom algorithms, which we actually call success models. We create that for each role within each company we work with--we're very focused on custom models. These models can then be used for recruitment. They can also be used for internal mobility to help people move within the organization, as well as provide insights into the cognitive, social and emotional traits of an entire workforce. And our core drivers are really fairness and then predictability. It's our fundamental belief that everyone has their place in the world of work. Every part of our approach focuses on this by matching people to jobs where they are going to be successful and thrive.

APIP:

Just so I understand a little bit better for the benefit of [our audience]... you have some type of monitoring, is it a continuous monitoring, after, let's say a solution has been tailored for one of the clients?

Michelle:

Yes, absolutely. There are two key ways that we check the fairness of our algorithms. First of all, when we build the algorithm, we actually test it rigorously against a sample of data that we have from people who've played the games previously. And for those individuals, we have gender and ethnicity data, and we test the algorithm to see whether females are being recommended at different rates than the males or whatever it might be. And then we will make adjustments to the algorithm until we get to the point where it's fair. And we won't deploy or release an algorithm until we're confident and comfortable that the algorithm is fair. But we don't rest on our laurels -- we also conduct validity analyses. So, once the algorithm has been active in the recruitment process for some time, we will then test that algorithm to check on the fairness of that algorithm so that if we need to make any further adjustments at that point we can.

We follow the US Equal Employment Opportunity guidelines around fairness and we abide by the four-fifths rule. We also look at statistical significance as well. So once again, in compliance with the US EEOC guidelines. That's really our core framework in terms of actually measuring the fairness of our algorithm, not only up front but also when we conduct the fairness analysis as well.

APIP:

When you think about the power of this technology, what are you excited about in the future? Shall we talk a little bit about what the future holds? Especially in respect to HR?

Michelle:

Yeah, absolutely. I am really excited. I think that there's a lot of possibility in terms of what technology and AI can actually bring, and everyone in HR right now is talking about the future of work, and future capabilities that are required to be successful for individuals and organizations to be successful. At the same time also, there's a little bit of fear around massive job losses due to automation-- robots taking over the world, taking over our jobs, etc. We probably need to take a little bit of a step back and look at the fact that through history, we've actually eliminated jobs several times over because of a variety of disruptive technologies like the printing press, the steam engine, the car. We've always been innovating, and this is not new and I don't think it's catastrophic either. I'm actually quite optimistic because we've dealt with these changes before and will do so again.

From an organizational perspective, my view is that the automation that we're seeing today is definitely going to continue across all parts as organizations try to become more efficient. And the focus on automation will definitely be much more around those, which we're already seeing. They're those repeatable tasks that a machine can actually easily do and do so with accuracy. And so the way I see it, and I'm not alone here, is that we can expect to see jobs being changed by our AI and not necessarily fully replaced. It's not as if machines will do everything and humans will do nothing...If we look at radiologists, we know that AI can now detect tumors on an X-ray. Can they do it better than humans in some cases? Yes, but what the research actually shows is that AI is better at detecting some types of tumors while doctors are better at detecting other types. So we're seeing a real benefit of machines and humans working together rather than necessarily being used in isolation.

APIP: In a sense, everyone really has the power once they have the data to create something that can be quite harmful as well. Right?

Michelle:

Absolutely. I don't think it’s necessarily the intention. I think it's just sometimes things can be done with good intention, but negative outcomes can occur, right? If you're not careful enough -- but we are. We're very active with government departments and independent bodies, and Frida (pymetrics’ CEO) is doing a ton of work in terms of informing the ethical use of AI because we are so passionate about doing the right thing. With the use of facial recognition, not only in recruitment, but also in law enforcement and those sorts of things…particularly in the US it's [been under] significant levels of scrutiny around the fairness, as facial recognition technology isn't quite at that level yet where it is as accurate as we would like it to be. And yet, we've got some very high-stakes decisions being impacted by that data. We need to really think about the data we use and the confidence that we have with that data. You're absolutely right, that's where those ethical, human centric principles come in to make sure that people's work is safe. It is of paramount importance.

Our philosophy and approach are very much around that there isn't one size fits all... Everyone has their place--it's about finding that right fit.

‍

APIP:

Yeah, absolutely. So in terms of you guys thinking about the future, what's next for you guys? Who do you think needs to be focusing more in this area? Or do you think you would start to open up into different sectors?

Michelle:

Our passion is definitely around helping [everyone] find their place in the world of work in a way that is fair and predictive. That will always be our North Star. In terms of additional products, every innovation really focuses on that and how we can make the hiring process better, but also how we can actually help people move internally and how we can help organizations understand more about their workforce and, and their people. So, for example, we have a talent marketplace where individuals who may have applied to a client played the pymetrics games, and they may have been unsuccessful with that client, now have a new opportunity. The candidate would obviously consent to this to go into the marketplace where they can be matched to other opportunities with other pymetrics clients… One door closes, but many others can open. Our philosophy and approach are very much around that there isn't one size fits all. So It's not as if only the smartest people who score X on a particular assessment are the ideal candidate. Everyone has their place--it's about finding that right fit.

We also have a new innovation around providing graduates with guidance upfront before they apply to a graduate program…and we have some clients actively using that technology right now. Because often for grads, it's very confusing, right? I'm applying to an IT organization, I'm really not sure which graduate stream to apply for. And I'd like some guidance and advice. With technology, they're able to have a clearer picture of where they might best fit and apply accordingly. We're continuing to focus on providing our clients with further insights around their workforce and the traits that individuals will bring in, and how we can move people, across from one industry or one segment to another within the organization.

I worked for the ANZ bank for 13 years, and I always stayed in HR. At one stage, I managed internal mobility for the organization. When people were displaced by business restructuring, they were really quite confused, even though they had been there for 10 plus years. They didn't necessarily have visibility across the whole organization and where else they could fit in a business that's 40,000 people plus. You're not necessarily going to know every single possible job that you might be suited for. And so, having technology, able to drive and support that just takes a lot of the guesswork and a lot of the anxiety, quite frankly, out of that process. 

__________________________________

If you would like to learn more about pymetrics’ AI- & mission-driven approach to recruiting, feel free to visit our Mission page here.