Advancing Health Care with AI: Humana’s Slawek Kierner Talks Synthetic Data and Real Lives
25:03 Share

Today, we're airing an episode produced by our friends at the Modern CTO Podcast, who were kind enough to have me on recently as a guest. We talked about the rise of generative AI, what it means to be successful with technology, and some considerations for leaders to think about as they shepherd technology implementation efforts. Find the Modern CTO Podcast on Apple Podcast, Spotify, or wherever you get your podcast.

There can be sizable gaps between people working directly with technology and people working in product teams. Today's episode with Slavik Kerner, Senior Vice President Digital Health and Analytics at Humana, illustrates how diverse experience can come together in novel ways to build value with AI. Hello and welcome to Me, Myself, and AI, a podcast on artificial intelligence and business. Each episode, we introduce you to someone innovating with AI. I'm Sam Ransbotham, Professor of Information Systems at Boston College.

I'm also the guest editor for the AI and Business Strategy Big Idea program at MIT Sloan Management Review.

And I'm Shervin Kodavande, senior partner with BCG, and I co-lead BCG's AI practice in North America. And together, BCG and MIT SMR have been researching AI for four years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and deploy and scale AI capabilities and really transform the way organizations operate.

We had a great discussion with Prakhar Mahotra from Walmart last time. Prakhar himself has a fascinating background from organizations like Twitter and Uber to his current job at Walmart. There's quite some differences between those organizations. It's also interesting to learn some aspects that are similar. Check out our last episode for the fun details. So for today, let's go on to something new. Shervin, I'm really looking forward to today's episode. Slavic, thanks for taking the time to talk with us today. Part of our focus is on you, really. Can you...

Introduce yourself, tell us what your current role is and we'll go from there. My name is Slawek Kierner and I'm a Senior Vice President in charge of data and analytics at Humana. Humana is a Fortune 60 company.

really focused on healthcare and helping our members live longer and healthier lives. I have roughly 25 years of experience in data and analytics across consumer technology and healthcare industries. And you know, I've always been interested in data. My first money I spent on a PC and got MATLAB and Simulink loaded there and I would create all kinds of simulations, get them run overnight and

see what's coming out in the morning and use all kinds of ways to visualize data. At that time I was fascinated with process simulations and autonomous control systems and adaptive control. And this fascination got me a job with P&G, Procter & Gamble.

And the good thing is that I could continue to use my passion. So there was a lot of permission space to innovate and bring advanced algorithms to these spaces. And that's how I started. And that was also a fascinating time because you could start to have feedback from your algorithms, from your early AI-based systems.

And with this moment, I got hired by Nokia. It was already the moment when iPhone was launched and that was another fascinating transformation.

And as you probably know the story, that part of business was acquired by Microsoft. And I moved all of my team and I was running data and analytics for Nokia at that time to Microsoft, looked around for a while, helped run some of the same operations across Microsoft's retail and devices business, and then moved to the cloud and AI unit.

And that got me to Humana. So roughly one and a half years ago, I got to this moment when I thought, okay, like, you know, I built quite a bit of experience in data and analytics and can make money with it. But now let's try to use it for some good purpose. And that's something that we can do in healthcare where next to having a lot of data,

we also have a very noble purpose. I think you've almost run the table at almost every sort of industry and every sort of application. Sounds like you're just right ahead of whatever's happening next. That's what I hope so. I'm looking for these challenges. This is pointless, but I have to follow up. Were you a chemical engineer back in the P&G days?

I worked very close to chemical engineers, by the way. So part of my projects were to actually rewire large chemical factories and think about really, really huge chemical operations that...

P&G has in major cities. But to your point, I actually did two majors. One was in mechatronics, but was essentially at that time in a fascination with AI in my part of the world. And the other one was in business management. I just mentioned that because both Sheridan and I were chemical engineers back in the dark ages and thought we'd found a compatriot in the whole thing. Because I actually got interested in, you mentioned simulation as your beginning as well. And that's where I started too, was

You didn't have to build a plant. You could just simulate it and just really show some of the opportunities from data. And it sounds like you saw some of those same things.

Exactly. And you mentioned simulations or factors. That's exactly what I was doing. It's interesting because that fascination that I had early on during my studies with simulation, I could bring to PNG. So a factor is very different than you are in Humana. That's a completely different... Chemicals don't mind if they sit for a while in a vat, but patients do. How do the things you've learned from that past experience influence your current?

I think, you know, there are a number of things that actually you can learn in supply chain and chemical processes that actually do apply to healthcare. It's fascinating. But let me just list a few. So first of all, the whole basic setup of process and process control, I think, applies to chemical factory. And when you think about...

you know, market, you know, it's very similar. You can, you know, apply so much marketing and, you know, when you do too much, of course you're wasting money, not enough is not going to give you the results. So you kind of meet a similar problem. And I think the troubles that you have with keeping chemical processes in control also actually we see in healthcare. So for example, you know, if chemical control has a lot of lag,

So the time from when you apply certain force or temperature to a time when you start seeing a result of it on the output of a process, the more this lag is there, the more difficult it is to keep the system in control.

Classic thermostat problem. That's what it is. And when you think about healthcare, we have exactly the same. So we were trying all kinds of treatments and programs for our members on the onset. So think about diabetes or issue of, let's say, having a heart disease and then needing to adhere to medications. Now, we try to convince you to do it, but very often we need to wait very long until we see that actually your health has improved and

And again, here we have this long lags and these lags are both inherent to the process itself and clinical domain. But quite often they also result from just poor data interoperability. I think you're making a super interesting point that there are archetypal

sort of problems in different disciplines, different industries, different fields altogether, but similar approaches once customized to that particular industry or company would give a lot of good results. And I resonate a lot with that. You listed some very good examples. Do you feel that diversity across having seen different industries

archetypal problems in different industries and disciplines. Do you feel like that is a important attribute of someone like yourself who's leading an AI organization for a company, just coming from sort of different backgrounds and different, having seen it across very different disciplines?

Yeah, I think it is useful. I mean, it's useful for me, of course, and I have a lot of respect for leaders that emerged from healthcare industry. And of course, they have that background, which I am learning only on how to really operate in a healthcare context. But to your point, I think specifically in healthcare, there is a need for people that will come from other industries and bring knowledge because it feels that healthcare is still a little bit behind.

Certainly from data, transformation and availability of data, certainly from usage of AI, and also from a platform point of view. So you mentioned several different examples through there. Can you give us some more detail about some particular success that you've had at Humana, particularly around, obviously, artificial intelligence is what we're interested in. Is there some story of success that you're proud of?

Yeah, there are a few. So we certainly are testing and learning a lot. I think the key progress that I'm really proud of is the fact that we have created our own internal machine learning platform, which helps our data scientists have access to modern technologies, to all of the open source capabilities.

have cloud accessibility such that computing power is not anymore limited by what you have in your data center. But essentially you can tap and run any kind of algorithms out there. We are starting to see the benefits coming through way better, way more accurate models that predict retention in our business.

that help us predict inpatient admission situations and therefore act hopefully way ahead of a time when our member needs to visit ER or get into a hospital and hopefully be there early enough so that we can help this person stay in better health and avoid needing an inpatient treatment.

There's also a lot of progress in terms of usage of AI algorithms and sophistication of this. We had to overcome a lack of proper

security in the cloud to handle PAI and PHI data. So as we're building those capabilities and helping also build those with our vendors, we had to generate high quality testing data that would be differentially private. We are able to create a new model, an AI model based on synthetic data, which is of similar accuracy to the one that is created on real data.

Using generative AI, we created high-fidelity synthetic profiles and populations of our members and could use those to ingest that to our platform. We started to learn how to use it. We trained our data scientists. We have

More than 200 PhD-grade data scientists at Humana, and they could already get access and start using systems ahead of our readiness for PHI and PAI data handling, which happened in the meantime. But the fact that we have the synthetic data creation capability actually is helping us in many other fronts as well.

So make sure I understand, you're using this tool to help your organization learn how to handle the real data. So you use AI to generate synthetic data that then lets everybody practice and learn on before it becomes a real patient. That's correct. Actually, I really like that example. Can you describe why, you know, I might be, I think I'm trivializing it, but why is that an AI problem and not a statistical sampling problem? Like what makes AI fit there well?

It's a good question. And I think this became an AI problem when AI became better. So I don't know, you probably have seen some of the work of NVIDIA that creates the synthetic faces of people.

So essentially, you use deep learning to train your network to essentially learn on how a face of a person looks like. And when you ask it to recreate, taking away the original data and controlling for overfitting in such a way that it can ensure that none of the training pictures is recreated exactly. So essentially, all the faces that synthetic generator generates are unreal and never existed.

And that field has started many years ago, but initially these faces, they kind of always had two eyes, but an eye could be in the middle of the forehead. And so you'd right away see that it's not real. But over the last two years, they have improved so much that when you look at this synthetic face right now, it's hard to recognize that it's not a human. It could be easily tricked.

So the parallel there is that you're performing the same trick, but with data rather than with images. Exactly. We look at a human's record, at the history, health history, but also actually a complete history of your demographic and health data. And we recreate the same population through an approach like this one. So fully differentially private, very high quality. When you look at this, take a physician looking at health history of a synthetic virus,

individual and a physician cannot tell that it's unreal and it looks like real. Who had the idea to do that? I would not have thought of that. So it's always a mix of people that are sitting around the table to try to solve a tough issue that we have. That's part of it. And quite often we invite our partners. So in particular, that technology came from a collaboration with a

with a partner from Europe who interestingly enough also worked with me at Nokia. So a very talented individual that created the synthetic data set capabilities and synthetic data creation capabilities and got a lot of success with this in Europe which as you probably know of course is much more concerned with personal privacy

And then another set of partners we are working very closely with right now is Microsoft and their advancement of differential privacy in this space. And then finally, we quite often connect with academia and we have those connections as well. This is a great example of how in a real technical topic from a different discipline with deep learning and image,

you know, recognition makes a tangible difference in a completely different industry. And I could imagine maybe 20 years ago, 15 years ago, clinicians and business folks running a company like Humana would say, well, that's not the same. It's not real patients. It's all synthetic. We can't trust it, et cetera, et cetera. My question is what level of

education and sort of knowledge sharing do you feel you've had to go through both at Humana and you know in your prior careers to sort of bridge that gap between the art of the possible on the technical side and the business where the understanding is not the same and do you feel like there is still a gap and you know how do you bridge that and narrow it

Yes, there is a gap. And I think there's a gap between technology and business understanding and there's a gap between technology and ourselves. We all in this particular field need to continue to learn. Every few years, things change and sometimes they change totally on us. So part of it, and I think the first skill is how do you continue to learn? Continuous learning

that is necessary for all of us and us as leaders who need to inspire our teams to do the same. Because even if you hire PhDs in data science that have been recently trained, they need to continue to learn. They need to have their workbench where they can tweak data and they can learn with others. And then the other part of it, which...

I spearheaded at Humana is a much tighter link with product teams in this large technology companies that we collaborate with. So what we are trying to do is to make sure that we are closely in touch with those product teams. We follow what they're doing. We participate to the customer advisory boards and through this both help them shape their products and for us and get excited and hopefully drive accelerated adoption

of these new features and functionalities ourselves. So that's one part of your question. So how do we stay ahead? And then, of course, we have a huge role of helping our business teams and our partners in enterprises where we happen to work to also understand the art of the possible and help us turn this technology knowledge into reality that actually advances our experience both internally and also for our members and customers.

My takeaway from what you're saying is,

Hiring the team and keeping the team at the forefront of the art of the possible and inspiring them is one part of it. But also organizations have to take steps to actually bridge these gaps through all these things that you're talking about so that there is more collaboration and sort of cross-functional teaming and much closer to the product and

management, with analytics, with AI, with voice of customer, all of that so that these ideas are allowed to even incubate and go somewhere. Is that right? Yeah, 100% agree. Well, if we've gotten 100% agreement from Slavic, I think that's a great time to end. Thank you for taking the time to talk with us, Slavic. Shervin, let's do a quick recap. Sounds good, Sam. Slavic made some very interesting points. One of the things that he mentioned a lot was

how past experiences led to his current role. And he had so many different past experiences and yet he found ways to apply them. I thought that was a pretty fascinating point. Yeah, I really resonated with that. He talked about some archetypal problems like the chemical engineering, the problem of system control and the lag in the system and how that has to be managed. And he likened this to a problem of marketing, right?

And then more importantly, to the problem of managing the care of millions of the patients, because as they propose different treatments for different members, there will be a lag between what's working and what's not working. And the ability to understand what's working and what's not working and what that lag time is, I mean, that's almost a standardized thing.

chemical engineering or control systems, you know, electrical engineering problem. And his ability to see these archetypes and sort of transcend from discipline and domain and industry to, you know, health insurance is really, really important. Gave me a little hope that humans will still be around for a bit. It's interesting as well that in a lot of these examples that

The specific industry details were obviously different and they, you know, you can't just blindly apply them from one industry to another. And I think there's a role too, that creativity and being smart about what, what fits and what's smart about what doesn't fit. That again, seems very human.

Yeah, completely. And the other thing, you know, building up on that is importance of, again, building trust, right? And building trust of the humans in the AI solutions, right? And he talked about synthetic data to do synthetic tests of different treatments. And then he talked about the process of showing clinicians how that synthetic data actually mimics

the real data and how it does. And so I think that's also very important. Again, building trust and building trust in areas where, you know, human judgment really, really matters.

I liked how pragmatic it was too. You know that computer are going to have problems. And so you'd much rather find those problems out with generated data. And I thought what was creative about his solution was using AI to generate that data that smelled just like real data, but they could afford to screw up with. I like things like that where it makes complete sense once he says it, but I would never have thought of it myself. Yeah, that's a great point. Yeah.

That's all the time we have today, but next time join us as we talk to Gina Chung from DHL. Thanks for listening to Me, Myself, and AI. If you're enjoying the show, take a minute to write us a review. If you send us a screenshot, we'll send you a collection of MIT SMR's best articles on artificial intelligence, free for a limited time. Send your review screenshot to smrfeedback at mit.edu.