AI is No Longer Knocking, It is Inside the House!

Rethinking Healthcare takes more than disruption. It takes more than thought leaders. It takes change makers and doers. That's who we'll be speaking to on the Healthcare Rethink Podcast, giving you our dedicated listeners a rich body of insights to make your own change. This is the Healthcare Rethink podcast.

Brian Urban:
[00:00:30]    Yes, this is the Healthcare Rethink podcast. I'm your host, Brian Urban, and today we're getting deep inside the world of AI trends, barriers, and how it's going to ultimately impact everyone across the healthcare ecosystem. And who better to do that in this journey here than CEO of Protege, Bobby Samuels.
Bobby, thank you for joining our show here today.
Bobby Samuels:    Yeah, thanks for having me.

Brian Urban:
[00:01:00]    Hey, this is going to be a lot of fun. We've only gotten to know each other a teeny bit before our recording here, but that's probably the exciting thing about most podcasts, now. You almost meet your guests or the interviewer right on the show, and then you just get to know each other. So that'll be fun in this sense.

Bobby, I love your background. I got to touch on it a little bit before we started. An ex-Datavanter for a long time you led a lot of the privacy hub work there and academically you are an absolute stud. You've done so much academia across the US and the most elite universities. It clearly shows that now you're leading your own AI organization.

[00:01:30]    All these things are like a culmination of when it was going to happen, you really stepping into your entrepreneurial world. Before I take any more out of your background, our audience loves to get to know our guests, so tell us a little bit more about your journey and how you've come here to start up Protege.

Bobby Samuels:
[00:02:00]    Yeah, well thanks for having me. Really excited for the conversation. I've spent my whole career in data, data connectivity, data privacy. First at LiveRamp, which is an ad tech, and then most recently at Datavant did a bunch of different things at Datavant including leading the privacy business like you mentioned.

[00:02:30]

[00:03:00]    One of the clear trends that we saw and left Datavant in January and started thinking about what I wanted to do is this massive tailwind of AI and the need for a platform focused on AI and AI training data. There's all this data that was not as impactful or was not as lucrative five years ago, that now is incredibly impactful for AI. It remains very illiquid and so the need for someone to help enable data holders to let the data be accessible for AI, I think was pretty clear to us.

Brian Urban:
[00:03:30]    That's exciting because it speaks to your ability to look across different desperate data sets and how they can be connected and how they can be more fluid across different users so they can be more accessible so it's not just sitting and collecting as well. That's really exciting. Thinking about AI and healthcare, there's a ton of use cases. Not all of them are being flushed out. I think there's a lot of baby steps being taken from healthcare delivery, from health plans, pharmaceutical manufacturers, device, everyone in between. But what are you seeing as the most, I guess, emerging trends that are starting to stick and really make sense and show value in the market today?

Bobby Samuels:
[00:04:00]    A couple trends that we're seeing. One is AI for radiology and things related to imaging. I think imaging is a super quickly growing part of the AI ecosystem in healthcare is at the same time probably one of the more mature use cases, even though it's in the early innings. We're seeing that come up a lot, whether that's in AI applications, whether that's in foundation models, et cetera.

[00:04:30]    The second one we're seeing come up a lot is in some of that back office work where you might want to take things like unstructured doctor's notes and then be able to connect them to say insurance codes. And so the ability to take what is very manual processes and either totally automate them or give the folks who are working in those industries way more leverage, I think are trends that we're seeing happen and accelerate.

Brian Urban:
[00:05:00]    And I absolutely agree with you on the diagnostic side, digital labs, it's starting to pick up things that the human trained eye cannot and the speed in terms of sharing is unbelievable. That clearly, even if it is relatively new, it is the most matured use case to your point.

[00:05:30]    I think what I'm hoping for in the payment world, the revenue cycle management world, is that these huge EHR vendors like Cerner trying to catch up from behind Meditech and other smaller ones that it's applied to helping reduce remittance, it's helping apply to having better provider payer relations. I haven't seen that come to life. I've seen all kinds of talks this past year at HIMS, HFMA, et cetera.

Do you think it's going to have a real tangible application to the payment side of healthcare?

Bobby Samuels:
[00:06:00]    I do. And a lot of these questions to me and to us as sort of a question of when and not if, and a lot of the data questions that you see come up in the real world data side and a lot of those pain points, data cleanliness and fragmentation are still true on the AI side, but I think that's starting to change. We're hoping to try to change those types of things and so do think that hopefully the whole ecosystem can become much more efficient and ultimately that benefits everyone in the supply chain including patients.

[00:06:30]    We're seeing folks who on that end want to understand remittance data and they want to understand data on things related to that because then they can build that into their models and then help ensure that, hey, we know that this type of claim is likely to have this type of outcome. Let's then incorporate that into how we approach things.

Brian Urban:
[00:07:00]    I see. That actually is quite enlightening for me because I'm seeing it as an application of we take away the human side of who should be being reimbursed for what underneath the rate sheet or underneath the contract. But this is something that could maybe go a little bit further to your degree, not removing the people side of it, but just cleaning up the process overall. That's hopeful. I think that's really hopeful. That's been the biggest challenge across health plan healthcare relations for a very long time.

[00:07:30]    I am curious, Bobby, so the one really big challenge that I think a lot of people in healthcare are hoping AI can really meet and start to change is care coordination and also prevention too from particular disease that are continually on the rise, colorectal cancer, breast, prostate, there's a lot of different applications into the imaging that you were discussing. But from a care coordination side of once someone is diagnosed, if they're diagnosed early, what's the path of treatment therapeutics, care coordination downward there?

[00:08:00]    Do you think that's going to really depend on EHR is it's going to depend on the people workforce development side and training people through AI and healthcare? How does it actually get on the road and start driving?

Bobby Samuels:
[00:08:30]    One of the things that we are seeing too is taking some piece of data and then saying, based on this piece of data, here's a treatment plan that we think makes sense. And in the background, one of the things that we haven't touched on yet, but I think is so important for AI is this notion of multimodality. And you can imagine a world where you're bringing in someone's data and lab data, EHR data, all of this that you bring that to bear and that informs things like treatment plans that informs how doctors approach things in care coordination.

[00:09:00]    And so we're seeing that happen or folks maybe in the early innings of that more and more and I think that's deeply exciting. I do think to point, I don't think that we're near a place where there are going to be a lot of folks in the shade who are replaced. I think I see a lot of this as let's give folks more leverage and help understand patterns that may not be visible to the human eye.
[00:09:30]

[00:10:00]    One of the things, I try to go to the dentist every six months as one is supposed to. And so I went to the dentist seven months ago and I went to the dentist a month ago and seven months ago they did images. It was the sort of same normal thing. They put the thing in, they took a picture. Last time I went, they put a big machine in my mouth, they scan my mouth and they said, "All right, here's our baseline picture going forward. We're going to do this every six months or every year, and we will then get us a readout of based on this, here are all the little things that have changed." A doctor will review that and get that to me. And that kind of thing where you can see AI in person impact the care you are getting is really profound and that kind of thing is just going to accelerate.

Brian Urban:
[00:10:30]    That's a great example because I think a lot of people in healthcare, and this is great example of really what's considered a dental ancillary type of service underneath of a health plan coverage. People think about this big needle moving impact that AI is going to have. It's going to detect all things, cure all things, change all things, but really in the short term, it's about incremental, real, tangible progress.

[00:11:00]    And that's a great example because it's enabling that physician to be able to provide you with better care and whatever that means for your appointment there. That's a really good example and I think it makes it real for people that listen to our show is saying, AI is here, it's been here, it's going to accelerate a lot of processes, service models, and then there's some other bigger things as we progress that it will change. Am I on track with, I guess, the example that you were sharing?

Bobby Samuels:    I think what's hard about conceptualizing a lot of AI and the impact it's having ChatGBT was such a watershed moment where it went from zero to one basically overnight. And if you think about sort of those seminal moments in technology over the last couple of decades, you probably have that. You probably have the iPhone where things just came out and then it was this step change.

[00:11:30]    And the reality is the vast majority of AI in healthcare, healthcare is not an industry that is known for moving superfast, nor, I would argue, should it be. And it is one where you're going to see, to your point, all these incremental things over time where if you look over the course of seven months or seven years, there's going to be a massive change, but it's not going to be a ChatGPT type thing.

[00:12:00]    And so if folks have that in their minds as the benchmark, that's not what's going to happen. But it's going to be a million little changes, a million providers moving into how they go about the world and drug discovery companies thinking about how they're changing things. It's going to be a lot of meaningful, but smaller changes versus the ChatGPT overnight, how everyone thinks about the world has changed.

Brian Urban:
[00:12:30]    That is a very, very important statement that you just shared there because it is about how it will change over time, not this phenomenon that we've seen across the world in terms of a user level. It's very good that you shared that, Bobby. And when we think about it from a user perspective and then we put it into healthcare, immediately, you and I probably think about regulatory privacy, a lot of things that could impact the integrity of a patient provider relationship.

[00:13:00]    In your world, if we could go back for a moment in your Datavant brain from leading Privacy Hub for a while, what are the biggest concerns with AI application healthcare?

Bobby Samuels:
[00:13:30]    There are a lot. I mean, in general, HIPAA was written in like '94-'95, and so the Internet was just getting started. The idea that HIPAA could have foreseen AI is insane. And so in a lot of ways, HIPAA may say do this, and what we are trying to do and what we would encourage AI practitioners to do is sort of have an internal threshold and bar that goes well beyond what HIPAA might suggest.

[00:14:00]    We are thinking a lot about what are use cases that we feel like are appropriate, how do we think about de-identification? Because the context in which data is going to be used to power models that matters. And so I think the weight of what is being done on the de-identification side becomes all the more important when you're thinking about some of the downstream implications.

[00:14:30]    I do think it's also incumbent upon folks to be cognizant of how do we get to a point where models aren't spitting back training data? Because I think that is something that folks think about and rightly so. But I think more than anything, it just means that the stakes have gone up.

Brian Urban:    That's critically helpful because everyone's going to be adopting AI at a different pace in a different way for the most part, aside from some of the baseline use cases that we've discussed here. That will be an interesting view for us all to see what unfolds. I'd be remiss if we didn't talk about ...

Bobby Samuels:
[00:15:00]    The other thing I would throw out is like one of the, and this gets back to my earlier point around the type of data that really is impactful for AI that wasn't as important from a broad analysis perspective before is things like unstructured data. If you had, I think of the Reddit example, five years ago, Reddit data may have been used one-off as sort of an interesting academic project. Now it's a key part of how they think about their company valuation.

[00:15:30]    Similarly, data like imaging or doctor's notes or things like that is much more valuable today because of advances in machine learning and things like that. And so how one de-identifies, when you're talking about structured tables, the market is more mature in thinking about how to de-identify that than we are in some of the more novel types of data elements like unstructured data like imaging. And so I think that there is more prudence that's required in thinking about de-identifying some of those more novel data types that are newer entrants, so to speak, into the real world data world.

[00:16:00]
Brian Urban:    
That is really helpful. The data that feeds into the intelligence behind an AI model that will spit out a direction or a path for a user to have in healthcare when treating someone that is critical. I'm glad you brought up unstructured data or novel uses of unstructured data.

[00:16:30]    I think about credit bureaus. I have a very black and white view on these large, I think they should be more regulated and pushed to be better entities in the US, but they have a lot of unstructured data that shows behavioral side of the human condition that ones in healthcare could help in terms of social health needs and screenings and things like that.

[00:17:00]    The reason why I'm bringing up credit bureaus is because inherently most of their models are to a degree racist as we would look at them. And this is subjective from my perspective. And when we think about the equity inside AI models and how it could hold down large groups of individuals and how they're given access to certain things, how is that being thought of in AI today in healthcare? Are you seeing concerns about that right now from Protege or are you starting to think about what that could be in terms of how you address it?

Bobby Samuels:
[00:17:30]    It's something that we're thinking a lot about. It's something the market is thinking a lot about in addition to, and not only because from an FDA submission perspective, you have to demonstrate that you're really weighing that question. But yeah, I mean I think if AI doesn't work for everyone then, or even worse, penalizes certain groups, then a lot of the promises is neutralized or negated.

[00:18:00]

[00:18:30]    I think there are questions around, and I think there's been research both ways, but there are also issues if you include race in analysis, that can skew the results in different ways, too. And so I think there is a need for being cognizant of race, ethnicity, social determinants of health. I think it's probably need to do validation to ensure that you're getting results that are not skewed or biased one way or another, but it's certainly something that we're thinking about. It's something that we hear come up in a lot of different customer conversations and is really, really important to get right.

Brian Urban:    I wanted to get your perspective, even though Protege is very young in its work, it's something that is a very big challenge and I think in front of a lot of leaders desks today and how they approach being equitable in their AI algorithms for whatever they're deploying.

[00:19:00]    I would be remiss here if I didn't talk about Protege. My goodness, we're like 20 minutes in and we didn't talk about what Protege is doing today and where you're going. Break us down to today, here, September, what Protege is and what you're doing in the marketplace.

Bobby Samuels:
[00:19:30]    Yeah. Our business is built on the thesis that the biggest bottleneck for AI today is finding and using the right training data. And there are a bunch of reasons why that is. There are IP concerns and privacy concerns and security concerns and just general data fragmentation problems. We're building a platform for AI data exchange, starting in healthcare, but then looking to expand pretty much or pretty quickly beyond that to become what we want to be the sort of AI data layer.

[00:20:00]    As I mentioned, starting in healthcare, we started working on this in February. We've built the richest set of training data commercially available today in the US. Access to billions of unstructured clinical nodes, images, claims data, lab data, mortality data, et cetera. We are having our official launch later in the month and already good traction in conversations and looking to go after what we think is a really meaningful opportunity today that will grow really quickly.

Brian Urban:
[00:20:30]    That is really exciting and I'm very thankful to have your voice on our show here as you're leading up to your official launch and you're deepening in the marketplace. It sounds like you do a lot of not only research analysis across use cases, but also in the same light consultative approach to how you help design what a certain organization might be going to market with AI. It sounds like there's just so many different things that Protege going to be doing in the world of AI, so really excited for you Bobby.

Bobby Samuels:
[00:21:00]    Yeah, thank you. I mean, right now I'd say a lot of our focus is let's be the place that really creates that critical mass of data needed for AI. And I think to your point, this is so new for everybody. There is a consultative nature both at the buy side as well as folks who have a data asset and are thinking about licensing in one way or another. I think we do sort of see ourselves playing that role while building out what we hope is a scalable technology business at the same time.

[00:21:30]
Brian Urban:
[00:22:00]    
Some of the things you're just sharing here, Bobby, I saw this on LinkedIn the other day. This was coming from the former co-founder of Google. Eric was giving a speech to a small college class and saying, "This is the next wave of something that we can't quite predict how big it's going to be and how big it's going to change the way the modern world and developing worlds will interact with each other across payments, across healthcare, transportation, food, arts and sciences, everything in between."

When he was saying that he was giving a very clear vision of like, Hey, get ready. This is going to be 10 times bigger than what the iPhone was at the time in terms of communication and social interactions as well. Is that the same feeling that you are getting with AI and healthcare, or is it still hesitancy and finding the right baby steps to take?

[00:22:30]
Bobby Samuels:    
We are seeing more and more excitement, more and more momentum. We're still in the early days, but we think we're at inning one of a massive, massive movement. What he's saying very much aligns with what we see both within healthcare but then also beyond healthcare as well.

Brian Urban:
[00:23:00]    Really exciting because there's so much that can be done, but it seems like you're taking the approach of let's get a strong foundation, let's work collaboratively. Let's be sensitive to equitable AI algorithms. Let's be sensitive to privacy. We're being fundamentally sound in how we deploy things. And I love what your exchange marketplace will turn into. This is really exciting and you're just getting started, Bobby. This is the most fun thing.

[00:23:30]    Maybe a follow-up in a year from now, six months, we'll see what Protege has been doing and been up to. And thinking about the future, let's go out maybe five plus years from now, Bobby. How is Protege going to be influencing AI developments or the use of AI throughout data exchange and just in healthcare? What's Protege going to be in five years?

Bobby Samuels:
[00:24:00]    Ultimately, to build AI you need to have the right training data. You need to have the right data to start seeding your models, to build on your models, to move into new areas. Today, the process of getting that training data, unless you have that data internally is really cumbersome. And so what we're trying to do is make that process for folks building AI much faster, safer, more privacy compliant, and giving data holders the comfort to know that their data is being used in the thoughtful way.

[00:24:30]    What I hope in five years is that we're able to dramatically lower the amount of time needed to get AI into production doing so in a thoughtful, safe way. And folks come to us as sort of the go-to to begin launching and then building their companies.

Brian Urban:
[00:25:00]    I love that you're like an AI studio in which you are pulling in data to help speed up the process of how that data can be used in an AI algorithm and then tested and then put into use. It seems like this is a really cool innovation studio. What I guess the model is for Protege to start putting things out into the market. If I got it right.

Bobby Samuels:    Yeah, no, I think it definitely could go that way, for sure.

Brian Urban:    Yeah, this is very exciting, Bobby. Very thankful for you to be in our little show here today. And if you want to find out a little bit more about Protege, I believe your website is launched. I was on it earlier today.

Bobby Samuels:    Yes.

Brian Urban:    Any other information that you could tease about Protege coming up? Any events you're speaking at?

[00:25:30]
Bobby Samuels:    
The URL is withprotege.ai. We have some company fundraising news coming out in the next couple of weeks and then the healthcare launch in the end of the quarter, and then we'll just continue to roll out new verticals beyond that.

But yeah, I think more to come from us. We have a lot of good stuff coming out and yeah, Brian, thanks much for having me on and great to connect.

[00:26:00]
Brian Urban:    
Thank you, Bobby. CEO of Protege, Bobby Samuels. And for more exciting insights and excerpts, visit us at Finthrive.com.