Diversity, Inclusion and AI — Conversation with Frida Polli, CEO, pymetrics (transcript)

By BCG Henderson Institute

Martin Reeves, Global Director of BHI talked with Frida Polli, CEO of pymetrics, a start up which has developed a more cost effective, accurate and less biased substitute for hiring and promotion decisions, based on AI and neuroscience. Our discussion covered AI, diversity, human bias, the future of work and the connection to economic growth.

Listen to the podcast here.

Martin: I’m here with Frida Polli who is CEO and founder of pymetrics, which is an AI and neuroscience startup out of an MIT. Frida, thanks for joining us today.

Frida: Thanks for having me Martin.

Martin: We’re at WEF in Davos and one of the themes is Globalization 4.0 and inclusiveness. The idea that we don’t get the growth potential of the economy unless we solve for diversity and inclusiveness. And I think that’s somehow connected with your offering. Can you tell us a little bit about the pymetrics offering?

Frida: Yes, sure. And actually that’s part of why we’re here at WEF. pymetrcs is a WEF technology pioneer. And I think that part of the reason that they were interested in what we’re doing is because we solve a fundamental problem, which is, how do companies find the right workforce? And historically, we’ve relied on talent signals that are in our mind, fairly archaic, such as resumes and things of that nature. So what pymetrics does is it leverages behavioral neuroscience. Looking at people’s fundamental cognitive, emotional and social attributes and uses those as talent signals to understand somebody’s potential fit for a role. So that’s how we do the person to role matching.

Frida: And I think part of the reason that the WEF and others are interested in our offering is because not only does it make the matching process far more accurate in terms of people performing better, staying longer in roles and so on, it also makes the workforce chosen much more diverse. We often see increases in gender, ethnic and socioeconomic diversity, all three really important aspects of diversity. And I think what’s really fundamental to note about that is that, for the last couple of decades, we’ve always seen the best person and the more diverse person as being an opposition. There’s this very sort of strongly held belief that if you’re talking about diversity, somehow you’re lowering the bar, you’re letting people in that shouldn’t be. But actually what our products and our research shows is that it’s quite the opposite, that those two things go hand in hand.

Martin: So essentially you’re using neuroscience games and AI to match people’s behavior rather than our judgements about their behavior with potential job opportunities. And you’re saying that this results in a lowering of bias?

Frida: It results in a workforce that is far more likely to perform better and stay longer and also reduces bias. Those two things. Again, I want to just emphasize the fact that those two things go together. So we see huge ROI for companies that we work with, as well as benefit to applicants because they’re more likely to stay longer and be happier in their jobs. But as an additional benefit, side benefit or in our mind, very core benefit, there’s a very significant increase in diversity. And again, I would say that it’s not just gender and I think diversity, socioeconomic diversity is something we don’t often think about when we talk about diversity. But if you’re thinking about inclusion, it has to be part of the conversation.

Martin: And do you think about diversity as primarily an ethical issue or do you also think about it as a performance issue?

Frida: I think it’s both. Honestly, I think of it more as a business issue because I think that all of the studies have shown that a more diverse workplace performs better. There’s just been tons of research that’s been done on that subject. And the question is more how do we get people to cross that hurdle? We were talking about this yesterday. So unfortunately, as humans, our brains are created to be biased, because what we’re constantly doing is taking in small pieces of information and assimilating that new piece into an existing corpus of information, right?

Martin: And some people would say that that’s what AI does. It takes past data and essentially is very good at pattern recognition in relation to past data that may actually reflect human biases. For example, if all engineers in a certain category happened historically to have been male, a lot of data will show that correlation. So how does your technology in simple terms remove that bias?

Frida: Yeah. So what we do essentially is we audit all of the algorithms that we create for bias. And we’ve open sourced it on Get Hub, you can go download it if you want to geek out on it. It’s called Audit AI and essentially all it does is audit every algorithm that we have built. It’s basically, a tree search for the least biased alternative. So what that means is any algorithm we build for a company, we say, okay, is it producing equal outcomes for men and women and for people of different ethnic backgrounds? And if the answer is no, we go back and we said, okay, what’s the next least biased alternative? And that’s essentially how we arrive at what we call an unbiased algorithmic solution.

Martin: Just to get technical for a bit — is the bias in the algorithm or is it in the training data set?

Frida: It’s in the training data set. That’s the thing that I always find so amusing when people are lambasting AI. They’re like, “AI is horrible.” And I’m like, AI, the definition is a computer that basically is mimicking human behavior intelligence. So if you don’t like what the computer is spitting out, you really should be more concerned about the fundamental underlying human behavior that it’s mimicking.

Martin: But your experience is that, even given dirty and perfect real world data, using your approach, we can get a better approach than using the traditional interviewing and CV based approach.

Frida: Absolutely. Because again, I’m back to the point of the human brain was created to be biased, because you’re always taking little bits of information and you’re assimilating it to existing information. So there is no way that I can remove bias from Martin Reeves or Frida Polli. That’s just it. That’s it. You know, Don Quixote task. However, you can actually test an algorithm for bias and remove it, which is what’s so amazing about the whole idea of incorporating this technology in hiring, not to mention that is more predictive.

Martin: Right. Predictive of performance in a particular job category.

Frida: Yes, that’s right. Yep.

Martin: And you’ve now accumulated the data to prove that.

Frida: Yeah. We’ve worked now with over 80 companies and the data just is very consistent in that fact.

Martin: Do you think that you’re working on the bottleneck? Because obviously, let’s assume that everything you’ve said is perfectly right. You get a less biased result, a more accurate result and a more predictable result with your technology. It could be that something else is the bottleneck. And I say this because in the WEF kickoff press conference, that was a very interesting fact that cropped up for me, which is WEF is a very earnest and intelligent organization that prizes diversity, has increased its female participation from 21% last year to a forecast 22% this year. So it can’t be that easy.

Frida: Right, it’s not.

Martin: So, number one, do you think you’re working on the bottleneck or one of the key bottlenecks and number two, why is it so hard to execute such a simple agenda?

Frida: Of removing bias?

Martin: Well, of in this particular case, equalizing gender ratios. So removing that particular bias.

Frida: Well, let me answer that question first. Again, it’s back to the idea that bias took decades and each one of us to be created. Therefore, it’s just like what we’re talking about an unconscious bias training. Unfortunately it’s a four hour seminar, it’s not going to remove bias from a person in four hours that took four decades to create. So it’s just about … it’s like anything else that’s been learned over a long period of time. It’s very hard to remove and therefore you really need to institute, in my opinion, something that’s much more systematic in order to really get at that. So I think it’s just fundamentally changing human behavior is very difficult. However, as we know with orchestras, for example, when they instituted the blind audition with the curtain that came down, they were in much more swift manner able to go from 5% participation to 35% participation. So I do think it is working on the bottleneck to your point.

Martin: So supposing a company adopts a technology like yours and so at least, the selection or the matching part of the HR process becomes less biased. What else do they need to solve for to get a final result? To get say, gender equalization?

Frida: You mean across and up into senior management role?

Martin: Up into senior management, yes.

Frida: I think then you have to really use the similar type processes and/or technologies all the way up even including in your promotion strategies, your lateral mobility strategies, all the way up. I think that’s a piece of it. I think then obviously there are policies that companies are putting into place to make the workforce more flexible for whether it’s mothers that need time- whatever flexibility people might need — that’s different from the traditional person that they would employ. So I think it’s a question of technologies and processes that need to be put in place.

Martin: I guess another thing that needs to happen is behavior change. In other words, the processes of companies, are sort of like habits. They have a certain inertia, if the HR department has been doing things a certain way for years and management has been doing it a certain way for years, technically you may have a solution, but you’ve got to change behaviors. Do you find that the science, the data alone is sufficient to shift those habits? Or does it require a change management plan?

Frida: It’s a combination of things. I think the data can be very compelling, but I think that again, people are stubborn. We’re all stubborn, and people don’t like change and we all don’t like change. And so therefore it is a question of implementing change management. And again, it’s back to the idea that technology, in the way that the WEF envisions it, and I think that I envision it, probably you as well is not here to replace humans. It is to elevate the status of the activities that we do. And again, I always like to say why would a recruiter want to hold onto the glorified status of a resume reader? How is that even remotely exciting as a profession? It’s not.

Martin: Well, it may be a comfort to historical habit or it may be reassuring.

Frida: Well, sure, but what we see for example, when we are able to have that be a pymetrics thing and then the recruiter does something more elevated is that really they go on to have much more strategic roles within the company rather than simply reading resumes. So it is really about explaining and in convincing people through showing them past examples of companies we work for — that it’s not about removing humans — it’s about elevating a human to do a much more interesting task.

Martin: It’s about elevating the cognitive surplus through cognition at higher levels.

Frida: Yes, cognitive, creativity, empathy and all of those things. If you can put, in this case a recruiter in front of applicants to speak to them about the roles, to engage with them on a much more human to human level, that’s a far better use of their time than sitting at a desk reading resumes.

Martin: This is very interesting. Essentially, you have an example, which is the opposite of the popular meme. The popular meme is that AI is going to replace us all in all tasks. Whereas here you have a task where actually it’s complimentary.

Frida: Absolutely. And that’s what came out of the WEF research that was done last year or last year. Where they basically showed that Net-Net job … and actually it was research that was done with WEF and BCG, Net-Net there were going to be job gains as a result of automation in AI. But I think it was like 80% of jobs were going to be changed or transformed. So it’s not that there’s going to be this massive loss of jobs. It’s that some jobs will be lost, the vast majority will be changed, but in the outcome there will be more jobs. They will just be quite different. And I think that’s what people are afraid of because they don’t see the future. It’s very hard for us to perceive a future that we don’t know what it’s going to look like.

Martin: So if we broaden the question now, given your unique vantage point, given that you’re in the business of this intersection of AI and organizational routines, do you have a vision of the future of work? When all of the short term factors have played out and we’re in some end state where AI is playing a more dominant role…

Frida: Absolutely. I always say that if everyone used the technology like pymetrics, the workforce would be much better functioning and much more diverse. And then that’s really the vision that we imagine. Where people’s capacities are utilized to their full potential because you’re in a role that you’re well suited for. Where everyone is included in the economy, not just the selected few that historically have had access to those opportunities. And again, I think that it works better for everyone. I think that I’m a huge believer in the fact that a lot of the problems that we see, a lot of the challenges that we have today are because there has been unequal access to opportunity historically. And that’s changing over time slowly and I think things are improving. But I think that the future of work, a vision where everyone is best suited for the roles that they’re in and you look around and it’s a much more diverse place in terms of gender, ethnicity and socioeconomic status is, in my opinion, a much, much better vision of the world of work than where we are today.

Martin: Christine Lagarde yesterday in the IMF press conference, she said that there were three priorities to deal with the current economic situation where we’re seeing depressed growth, but vastly increased risk. She had, one of them was resilience to the unknown, this was basically advice to policy makers. The other one was collaboration, international collaboration so that we don’t end up with some sort of a stalemate or lose-lose with respect to trade. But interestingly the third one was inclusiveness, which she said it was necessary to unlock the full dividend of the digital revolution. So it sounds like you agree with that proposition.

Frida: Yes, I completely agree, completely. And I think the more we exclude people or leave people out, whoever that group is, I’m not thinking of a particular group. The more we really limit the growth of the economy, in my opinion, it’s pushing out the pain. As much as we can include people, we’ve always thought about it as a zero sum game — and I just don’t believe that’s true. I think that what we would see if we were a more inclusive society is that people that had the most access to opportunity would then go on to be entrepreneurs and build new companies. Because they would have the most access to capital and social capital and all the rest of it. And then, we would just expand the economic pie for everyone rather than thinking about it as like, oh, I’ve got to fight over this one job because there’s this finite number of jobs that we’ll only ever have. I just think that’s a very antiquated notion.

Martin: Finally, I wonder whether you could done share any thoughts about the joint research that we’ve recently undertaken where we looked at using the pymetrics, neuroscience games, and the BCG strategy games to look at the neuroscience capabilities underpinning strategic skills. What were your reflections on that research project?

Frida: Well, I thought it was super interesting because I think that your way of looking at different types of strategies that are successful for different types of environments is actually unique because you go to business school. I went to HBS, they teach you that there’s Porter’s Five Forces and that’s how you create a strategy. So I thought it was very interesting to think that different environments require different strategies. And then to go back and think, okay, what are sort of the cognitive, emotional and social aptitudes that map to those. I think was fascinating to find different profiles it kind of made sense that those would be the profiles that would be needed. So it was very complimentary tools, that put together, created an outcome that made a lot sense.

Martin: Well, for me it brought cognitive diversity alive in the sense that I was familiar with the concept. But interestingly, when we search for the skills underpinning strategic capability, we actually found that there was no such thing as a universal strategist. It’s just 2% I think of the sample were facile in and had all of the skills necessary for all the environments. So actually it made the conversation much more specific. What sort of person do we need for this specific problem, which naturally gives you a very performance-based rationale for actually embracing a diversity of cognition.

Frida: Absolutely. I think that I’m a pretty good entrepreneur. I think I’d be a pretty horrible lots of other things and that’s the whole idea behind pymetrics, it’s not a one size fits all. It’s really about helping people understand where their highest probability of success is. And I think if we start thinking about work in that way, rather than thinking there’s a good employee out there and a not so good one, I think we’re just opening up a lot more opportunity for everyone.

Martin: Well, thanks very much Frida for sharing your interesting perspectives with us today.

Frida: Yes, thanks for having me, Martin.

About the speakers

Dr. Frida Polli is a Harvard / MIT trained neuroscientist turned startup founder. She is the CEO and co-founder of pymetrics, a company using neuroscience and AI to improve the accuracy, fairness and diversity of hiring. With global clients including Unilever, Accenture and LinkedIn, pymetrics has matched hundreds of thousands of people with their ideal jobs, while removing bias from the hiring process.

Martin Reeves is a senior partner and managing director in the New York office of Boston Consulting Group and the director of the BCG Henderson Institute. You may follow him on Twitter @MartinKReeves and contact him by email at reeves.martin@bcg.com

About the BCG Henderson Institute

The BCG Henderson Institute is the Boston Consulting Group’s strategy think tank, dedicated to exploring and developing valuable new insights from business, technology, and science by embracing the powerful technology of ideas. The Institute engages leaders in provocative discussion and experimentation to expand the boundaries of business theory and practice and to translate innovative ideas from within and beyond business. For more ideas and inspiration, follow us on Twitter: @BCGHenderson

The Boston Consulting Group's strategy think tank, dedicated to sharing ideas and inspiration that help forward thinking leaders shape their next game.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store