TRANSCRIPT
EPISODE 29: FRIDA POLLI
Jim Freeze Hi! And welcome. I’m Jim Freeze, and this is The ConversAItion, a podcast airing viewpoints on the impact of artificial intelligence on business and society.
[UPBEAT MUSIC]
On today’s episode, we’re joined by Dr. Frida Polli, Co-founder and CEO of pymetrics, a talent-matching platform that combines AI and behavioral neuroscience to pair people with the job openings that match their skillset. Frida launched the company in 2013, after noticing shortcomings with the recruiting and hiring process. It can be biased; it’s often inefficient; and it relies too much on what a candidate has achieved in the past, rather than what they can achieve at a new company. Today, she’ll share how companies can improve that process resulting in significantly higher retention rates and more diverse hires. Frida, welcome to The ConversAItion.
Frida Polli Thank you, Jim. Thank you for having me.
Jim Freeze We’re thrilled to have you here. So you have a pretty impressive background as a former MIT and Harvard cognitive neuroscientist before your turn as an entrepreneur. What inspired you to found pymetrics and how did your early experiences in research and business school shape the company?
Frida Polli Sure. I transitioned out of academia through the MBA program at Harvard. It was there that I started recruiting firsthand for two years, because that’s what MBA students do. I was just struck by a couple of things. First of all, what people wanted to know about somebody was not what was on their resume. That was very easily accessible to everyone. It was what was not on their resume. It was their potential to do a job, not their pedigree. That raw human potential is what we had been measuring in the lab for decades, essentially as cognitive scientists; it’s their cognitive, social and emotional aptitude inherent abilities that we’re all born with. I was just really struck by that, that really there was no way to measure people’s potential for a role rather than simply their resume experience pedigree.
Then on top of all of that, I just could see that this very resume driven process was leading to a lot of mistakes. We had classmates of mine who you could have told them from a potential perspective, they were the wrong fit for the role. I’ll never forget someone who was applying for investment banking, but really liked to sleep 15 hours a day. But she’s a smart girl, got the job and then, two days in, hated it. There are so many, many stories like that. That was one problem that I saw is that not assessing potential for a role led to a lot of bad fits. Then the second piece that I saw was really biased. The way that people were assessing folks was leading to biased outcomes where, I’ll take myself, I wanted to be a tech entrepreneur. There certainly aren’t that many women who are single parents that have graced the covers of entrepreneur magazines. So it really was this issue that this pedigree based approach to hiring also yielded a lot of bias in the process.
Jim Freeze Yeah, that’s really interesting. I love how you characterize the raw human potential. I think back over my own career. I think some of the most talented technologists I’ve ever met in my life that I’ve worked with were liberal arts majors; and to your point, if they’d had maybe a computer science degree, they would have been obvious for a role, but not necessarily if they had a degree in sociology. It’s an interesting concept. How do you take that notion of human raw potential? How does the pymetrics platform blend AI and neuroscience to optimize talent-matching? Can you walk us through it?
Frida Polli Yeah, sure. So again, all of the technology that we brought to bear on this problem is not something we developed. The cognitive science behind measuring human potential is something that has been done now in labs across the world for decades. That’s how I became exposed to it. So what cognitive scientists have learned is that is an evolution from the paper and pencil questionnaires of the 1950s, is that if you measure people’s behavior while they are doing certain activities, that is a much better signal of their raw potential for certain things than asking them to self-report on something. The analogy that I have is if you really want to know how much someone weighs, you’re far better off putting them on a scale than asking them to self-report on their weight. So what we do is have them go through a series of behavioral activities, again, all of which have been developed in academic labs for the last couple of decades; they’ll look at things like planning, sequencing, generosity, emotional EQ, risk profile, and on and on; a whole series of what we would consider people’s raw attributes.
There’s no right or wrong way to behave on these activities. It’s really telling us whether you’re more on this end of the spectrum; more risk-taking, or whether you’re less risk-taking, for example. Or for planning, whether you’re more of a spontaneous planner, or a very methodical one. Then what we do, the AI piece comes in when it’s about matching somebody’s profile to the right opportunities. So Frida Polli has a particular raw potential profile. The next thing I want to know is, what does that make me more versus less well-suited for?
It doesn’t mean that I can’t go do something that I’m not as well suited for. It’s just going to point me in the directions of things that I’m more well suited for. That’s where the AI piece comes in, that we’ve collected millions of profiles now; people that are gainfully employed in different industries, and jobs, and roles, and created hundreds of models to then map people to say, “Hey, Jim, you just went through our system. Here are all the jobs that you’re a high fit for, these are the ones that you’re a medium fit for, and these are the ones that you’re a low fit for.” And then again, it doesn’t block you or prevent you from pursuing a career that pymetrics says you’re maybe not as well a fit for, because obviously companies are evaluating things other than just your raw potential as measured by pymetrics. But we hope that it gives somebody a unique data signal that is also unbiased. That’s another piece by which to evaluate someone on.
Jim Freeze That’s really interesting. You’re also a strong advocate for ethical AI, and tackling bias in technology. You’ve touched on it a little bit, but how have you brought that kind of an ethos to life in the pymetrics platform?
Frida Polli Yeah, absolutely. The problem with ethical AI is that everyone has a slightly different definition. I’ll be very clear in my definition. Our technology is ethical in that it has no disparate impact. A disparate impact is essentially a big fancy word for gender or ethnic bias, right? A lot of tools out there, unfortunately still have a lot of disparate impact and it’s well-known, but people are just not addressing it in the way that they should. These are AI based tools and non-AI based tools. For example, cognitive testing, which is used by 50% of companies, and is administered by companies like Wonderlic that have been around since the 1930s, use cognitive tests that produce very poor racial outcomes. For every 10 Caucasians that pass a cognitive test, only three African-Americans are passing. It’s not because what we’re measuring is the problem, it’s because of the way the test has been designed.
Because there are a lot of tools with disparate impact out there in the market, our whole thing is that we will not release any algorithms that have disparate impact, and there’s a threshold that’s legally defined as fairness between genders and races. We abide by that threshold. That’s our commitment to our clients, to our candidates that go through it, is that you can be assured that if you go through our platform, there is no racial or gender bias in the outcomes that we have. We’re actually in the process of publishing aggregate data across hundreds of thousands of candidates that have gone through a platform in order to actually have some evidence out there so that people don’t just have to take my word for it.
Jim Freeze It’s a very compelling value proposition to be able to assert that and have the data to back it up. The pandemic has changed everything. It has certainly upended the job market leading to “the great resignation.” I think a situation that probably nobody could have foreseen, and it’s also completely transformed the way we work. What shifts or trends in hiring have you seen over the course of the last year?
Frida Polli Yeah, I think the shifts we’ve seen are positive. One is less of a reliance on traditional methods that I think could lead to bias. For example, a lot of people in the early career space are relying less on physical job fairs, which I don’t think anyone enjoyed, whether you were on the recruiter side or the candidate side, and are relying much more on digital processes. I think the good part about that is that it really opens up a lot more opportunities. With a physical job fair, it’s far more limiting in terms of the variety and diversity of applicants that you can consider. With a virtual one, the sky’s the limit in terms of who you can consider.
I think that’s a really positive trend that, in this reliance on digital tools, it really opens up the door of possibility and opportunity for folks. Now the flip side is, I think, then we have to be really certain that the tools we’re using to then make decisions along the way don’t have disparate impact because, then if not, you’re throwing all sorts of diversity at the top of the funnel, but if your tools are still exhibiting a lot of disparate impact, then you’re clearly not going to end up with a different outcome than what we had before.
Jim Freeze Absolutely. One of the things we talked a lot about on this podcast in many different episodes is consumer comfort with AI. We’ve talked about it in a number of different episodes. Leveraging technology in the hiring may seem a bit of a polarizing idea for some who don’t understand the ins and outs. How does pymetrics think about the issue of consumer comfort and general acceptance? What do you think of a way for the everyday person to appreciate and value AI tech delivers?
Frida Polli Yeah. Well, again, Jim, I would go back to the fact that first of all, the reality is… And actually Joe Fuller at HBS wrote a paper with a number of folks that came out recently, basically saying that look, 99% of fortune 500 companies are using automated decision-making, oftentimes with AI associated with it at the top of their funnel, and it’s been happening since the 1990s. The fact that it’s all of a sudden being discussed, I think is more a fact that a factor of us becoming more interested in it rather than that it just started happening. It’s been happening now for decades. Three decades.
Jim Freeze Yeah. That’s a really good point, an interest in and an awareness of it, right?
Frida Polli Yeah, I think that it’s been happening for three decades and the more antiquated technologies that are in much broader use actually are far more problematic than some of… I mean, we’re a social impact venture. Our whole premise is that we’re improving outcomes and we have to report on that to our social impact investors. I think that the point is there’s much more heightened focus on it. Now I think it’s really about gaining or regaining trust of the consumer. I think the way you can do that is through transparency and explainability. That’s really what we focus on. That’s why I was mentioning to you that we’re going to be publishing our disparate impact results, because at the end of the day, like anyone can say, “Oh, my tool is fair,” but if we don’t actually have to provide any data to support that, I’m with the consumer. How do I know that’s actually true? So it’s about providing transparency into the systems that you’ve built.
Jim Freeze Couldn’t agree more. In our business, we apply AI as the front door to our customer service operations, and it’s often the case that consumers don’t know that they’re dealing with an AI bot, so to speak. We always recommend to our customers that they let the consumer know at the beginning so that the consumer doesn’t feel tricked and they know that they’re dealing with artificial intelligence and we’ve got primary research that suggests that transparency helps with consumer comfort.
Frida Polli And the other thing I would note, Jim, is that AI has been cast as the big, bad villain, but I think actually I view it as much more. It’s a little bit like laparoscopic surgery, right? If the resume parsers of the 1990s, which are still in widespread use are like traditional surgery where you’re cutting up half the body and it’s a giant mess. I think AI can actually be far more fair than some of these other systems and much less… Not destructive at all basically. I think it’s unfortunate that the narrative on AI has been cast in such light.
Although, quite frankly, there are problematic uses of artificial intelligence. There’s been a lot of stuff written on facial recognition and hiring. I don’t think that’s a great idea, but I think again, it’s about enabling the consumer to be more discerning about the different types of technologies that are out there.
Jim Freeze Absolutely. So one last question for you. And I think I have a sense as to what part of your answer will be for this. Let’s see if I’m right. Where do you see room for AI to expand its role in hiring?
Frida Polli We haven’t thought about it so much as where it could expand. Obviously it could go into videos, in video interviewing and all the rest of it; it could go into performance management ratings after someone’s been hired. I think we’ve been really focused on expanding into internal mobility and re-skilling and learning and development because all of those areas lend themselves very well to what our platform does at its core.
Jim Freeze Yeah, you kind of touched on it earlier, too, where you talked about knowing, focusing just on the top of the funnel for purposes of recruiting, but actually probably further down in the funnel for purposes of hiring as well. After the fact too, as well.
Really fascinating episode, a really interesting technology. And we very much appreciate you joining us on this episode of The ConversAItion. Thank you.
Frida Polli It was a pleasure to be here.
[UPBEAT MUSIC]
On the next episode of The ConversAItion, we’ll be joined by Bridget Frey, the Chief Technology Officer at Redfin. We’ll talk about how the company combines AI and agent savvy to make the home buying process faster, easier and less stressful, and the diverse technical team behind the scenes that make it all possible.
This episode of The ConversAItion podcast was produced by Interactions, a conversational AI company. I’m Jim Freeze, signing off, and we’ll see you next time.
