Jerrold Jackson, who is the Vice President & Head of Machine Learning & Data at EXOS, shares how his team is revolutionizing the way data is used to enhance human performance.
What if you wake up, let’s say tomorrow and your fitness app, or your wellness app, suggests that you should go on a run because it will help you meet your calorie goals. Well, that’s great. But what if you don’t feel like going for a run? What if your body hurts? What if you are sore from the two days before when you blasted yourself for 45 minutes each day with really rigorous work. Those nuances and those feedback loops have to be accounted for, and we have to have adaptive technology.
[Audio transcription of the full interview]
Hi, welcome to The Tech Show. I’m Aziza with Focus GTS. Focus GTS is a specialized technology staffing and recruiting firm, focusing on data science and analytics, digital marketing technology, cloud technology, and business intelligence and analytics.
Today, we are here with Jerrold Jackson. Jerrold is a Vice President and a Head of Machine Learning and Data at EXOS. And today we want to talk about how to use data and technology to amplify human performance.
-Welcome Jerrold, and thank you for joining us.
-Thank you.
-Jerrold is currently working as a Vice President and Head of Machine Learning and Data over at EXOS. Correct?
-That’s correct.
-That’s awesome.
How is EXOS impacting human performance?
Can you tell us a little bit about what you do at EXOS and just the company in general as well as a little bit about your role there?
Thank you again for having me. So you’re correct. I am the vice president and head of machine learning and data at EXOS. We are a human performance company. Our founder, Mark Verstegen, is a pioneer in this industry of human performance. He spent the last 20 plus years, really just working on upgrading people’s lives. And what we mean by that is, the bedrock of this company is really in elite athletes and elite military training.
So being kind of the go-to, the de facto company and in programming and methodology. We’ve since evolved to also working with corporations. So we work with 25% of the fortune 100. And so that’s the history kind of the first two, two-plus decades of EXOS. Moving into this next decade, Sarah Robb O’Hagan, who’s our current CEO and an industry veteran, has been brought on board along with a few other executive-level leaders, to bring the company kind of into the evolving consumer landscape.
Four basic pillars of methodology that EXOS sort of prescribing to its mindset, nutrition, movement, and recovery. We can now bring this expertise of coaching and methodology to the consumer space. And so that’s a big part of what my role has been designed to do, which is to think about how do we personalize that entire experience. Now that we’re in this world where everything is digital, the idea is that we can have tech-enabled personalization. And so a big part of what I’m doing is bringing machine learning and other techniques to help do that.
How do you use data and technology to amplify human performance?
That’s interesting. I’m passionate about health and wellness and all that space as well. So I’m excited to have this conversation with you and learn a little bit more about everything you’re doing over your answers right now.
You know, nowadays we have a lot of technology that can track our performance. We have Apple watches, we have Fitbits, and then we have equipment in the gyms, right? And we also have equipment at home now with Peloton and Mirror. We have all kinds of different data points being collected at all times, but how can you tell a story? And how do you personalize the data? Because everyone’s different, right? I can go and have a workout and might have a certain heart rate, but I don’t know how is this data helping me, right? So can you elaborate a little bit on that?
The goal here really is to think about a couple of things. One is, how do we even get better at using data? We know that averages don’t serve you as an individual. So you are an individual with your individual goals, whether they’re athletic or not athletic.
How machine learning impacts the model
And so how do we use data to not only understand who you are, what you need but also how to get you there? The promise of approaches like machine learning and other technology, in general, is that we can process through a whole bunch of your data and a whole bunch of data of folks just like you to serve up some of the best, once again, personalized either recommendations or sort of methodological approaches that we have a notion will work for you.
We’re getting into a sort of models and how those all work and the need for data to feed those models. That’s effectively what it is, right? You can build a process that’s tech-enabled that kind of takes, what’s been data-driven for EXOS and really the entire industry for a while. And you can scale it in many ways to even supplement the in-person experience.
Is there enough data out there to personalize everything that we do in the fitness world?
Yeah, and some of it’s conceptual. I think that the industry, in general, has a lot of aspirations, and then there are a lot of folks, EXOS included, who are beginning to chip away at what the possibilities are? You mentioned a few of the very common and popular connected devices out there, which certainly can help, right? So how as an industry, do we take data that we know is dynamic and fast-moving and integrate that into tech-enabled processes to then serve up, as you said, all sorts of personalized environments for people to thrive in?
So I think part of what we’re focused on here at EXOS is creating those digital experiences and digital solutions. Once again, from the elite athlete, all the way down. Because we do believe that there’s a performer in all of us. All of us have some goal of some sort. To your point, if we can harness data and process enough. That’s I think the promise of this field, of human performance, and I think the use of technology and machine learning specifically.
What is the main challenge when it comes to building this type of personalized model?
I would think about this maybe as not just a single model, but almost sort of a group of perhaps even an ensemble of models. So I think it wouldn’t be correct, and probably just a little bit naive to think that there’s one model that serves every use case here. There’s certainly going to be a number of sorts of models or processes, methodologies working in tandem. To not only process your data, but to process the larger universe of data to inform your experience specifically.
And there are a couple of different levers that can be optimized and can be sort of impacted. Because there’s both the consumer, there’s the person who’s doing the exercise or doing the mindfulness or, you know taking part in recovery activities. There’s also this concept of a coach, right? And so there is multiple points kind of felt the value chain where a model or models can be implemented.
At the end of the day, any of these approaches are extremely data-hungry. And what I mean by that is, you do need vast amounts of data to do this in ways that are most optimal. There are approaches that you can take that certainly aren’t a substitute for having either large amounts of data on an individual or people like an individual. So segments of people or having this long history on an individual person or a group of folks or both. So there are different methodological approaches that you can take kind of in the absence of one versus the other.
Leveraging the full stack of Tech Talent
Also what would be vital to this field as it develops, it’s not just data science and thinking conceptually and academically about building models and algorithms, but it’s also a heavy engineering task as well. So any company that is endeavoring to take on this journey has got to think about leveraging the full stack of talent.
From data scientists to machine learning engineers to data engineers to software developers. To create these experiences that include personalization. Personalization as an exercise is great, but if it’s not wrapped into a larger experience, that I think it’s kind of missing the mark. So there are a lot of components that go into it.
How can this model be trained to be personalized to each individual?
From what you’re saying I understand that it’s a lot of data points, and the challenge comes in when you have to put it all together and the experience of the particular person or people that are similar to that person. I think that these types of models are pretty sophisticated and they require sophisticated techniques to build them. So there’s also dynamic data that goes into this human performance amplification model. Can it be trained to include other data such as the way a person’s body evolves as they continue their fitness journey or mind journey or recovery journey is not something that’s in the model?
Oh, it has to be, absolutely. It’s a non-negotiable from where I stand. What’s exciting to me in this field is that humans are incredibly complex. With the point I made earlier about averages versus more individualized data points, we need all of those dynamic changes to truly inaccurately serve up a product or a solution, a digital product or digital solution that meets the demands of the human.
Example of personalized experience for this model
So, you can go from, let’s say basic linear regression to some more sophisticated machine learning techniques. To give an example, linear relationships between data points, to nonlinear relationships between data points, right? So, it’s really important to kind of think beyond just basic associations and to have enough data to capture all of the nuances of the human experience.
What if you wake up, let’s say tomorrow and your fitness app, or your wellness app, suggests that you should go on a run because it will help you meet your calorie goals. Well, that’s great. But what if you don’t feel like going for a run? What if your body hurts? What if you are sore from the two days before when you blasted yourself for 45 minutes each day with really rigorous work. Those nuances and those feedback loops have to be accounted for, and we have to have adaptive technology.
Are you just going to start building this model out? Or You already have some things underway?
I can’t dive into all the details at this point, but what I will say is for EXOS specifically, I mentioned Mark and Sarah earlier on our founder and CEO, they’ve been focused for a number of years on sort of honing in on what the model is. What’s the best way to enhance human performance, right? Regardless of what your goal is. And it’s those same techniques that I simply am going to work to sort of leverage in a digital sort of space with technology.
It’s easy to say, but It’s a very difficult thing, of course, to continue to sort of build upon. But I’m joining a team of folks who are super passionate about doing, and who have been working at these things for quite some time.
Going back to data points, when you were speaking about some of the dynamic data. When you have certain data points, for example, like heart rate, it varies. So as you do cardio exercise, you have a certain heart rate and you burn a certain amount of calories. But when you go and do, for example, some kind of weight-lifting exercise that also gives you heart rates. Does that human performance amplification model take all that data into consideration?
Does the model take into consideration all the variables and data points to enhance human performance?
Any in-depth technology in the space has to account for a variety of things. Nothing is univariant. The world is not just in an individual data point. We’ve had to not only know that your heart rate is fluctuating going up and down, but we’ve got to know the exercise that you’re doing at that moment.
We also need to ideally be accounting for as we’re driving a personalized experience for you. What exercises you did historically, and what your heart rate was throughout those as well. We encourage you to go really hard, let’s say for a certain amount of time or not with a certain exercise with a certain frequency over a certain period of time. All of those factors are critical here. Not every data point is the same. So there will be data integrity issues. There will be lots of noise in the data. A big part of what is inherently difficult about this space is accurately measuring these things.
Let’s say you’ve got a connected scale at home and you weigh yourself. Whatever frequency that you do that scale may or may not be accurate, but hopefully it’s at least consistent. It may or may not be your actual weight as defined in science, but it, hopefully, be consistent because you step on the same exact weight each time.
So should we personalized to that specific number on the scale, or should we personalize to your overall goals of losing a certain percentage of weight? It’s a question that I think the industry has to grapple with in general, and it gets to some of the nuances and trade-offs that ultimately as a technologist, we have to think about. Do we want to optimize being accurate? For being consistent or being both?
What are some of the different data points that you would need to build this human performance amplifying model?
I think I can list a couple of ideas just to keep its broad strokes, but I think the high-level answer is you need enough data to once again, capture the nuances of the human experience specifically for the example that you were giving around body composition.
It’s a fairly complex thing to measure. If you go to a high-end proper fitness facility like EXOS, there are over 400 globally, right? Across six different continents or to a nice gym of another brand or even just a doctor’s office. They presumably have tools that can accurately measure things like body composition.
If you are at home on one of any number of connected devices that you got online somewhere, who knows what condition it’s in, how accurate it is? It becomes a challenge to build technology that accounts for the fact that there are even a variety of different types of signals in terms of the accuracy of them, let alone the signal itself. So, of course, we want things like heart rate. Of course, we’re on different biometrics. There are different kinetic or movement-based metrics that the industry has also begun to collect.
So connected devices in general, really all of them, should be used and should be inputs into any model or set of models that attempt to personalize an experience for you. In EXOS, our particular challenge, but also an opportunity, is that we’ve got 20 plus years of chipping away at those four pillars; mindset, nutrition, movement, and recovery. And across all four of those, there are ways to measure and capture that data. We’ve been doing it for quite some time. And once again, the real opportunity here is enabling it with technology in the digital world that we live in.
How do you deal with data quality issues?
Why do you deal with data quality issues when it comes down to that, because there are all kinds of data. But from what I heard from data sciences, you know, if you don’t have good data, you can’t build a good model. How do you deal with it?
Yeah, I would almost argue that you should plan to not have good data. It’s very rare. I don’t know anywhere, any phenomenon in the world, whether you’re in a pilot flying a plane or whether you’re a data scientist or data engineer building personalization, you’re never going to have a hundred percent visibility into what you’re aiming to accomplish, what you’re flying towards. And so you have to kind of do with what you have.
The advantage of having more data points than not, so having an excess of data, is that kinda like finding a needle in a haystack. There are more likely more needles in the haystack, the bigger your haystack is to use that metaphor. So, you know, it goes back to the point I made earlier about the models that you have to use in these environments being data-hungry, right? Erring on the side of more data points is certainly better than not. And the data quality issues really never go away, but you can make better estimations, right?
You can build models that make better estimations that have higher levels of certainty, the more data points that you have. There’s a phenomenon called the “Cold Start” problem within machine learning, and the general idea is that it’s difficult to build something when you don’t have anything right to start with.