The Stretch – How Generative AI is Making Benefits More Accessible and Empowering

Kevin Fyock:

Today, we're going to talk about the intersection of artificial intelligence and benefits. And rightfully so, we've titled our podcast, “How Genitive AI is Making Benefits More Accessible and Empowering.” AI is a topic that most employers and HR pros are thinking a lot about, but a majority of us are thinking about how AI impacts talent acquisition or learning and development.

But what's not commonly discussed is how AI is going to change the way we do and interact with employee benefits. So for today's podcast, I'm joined by someone who has thought a lot about artificial intelligence and thinking about how it's already impacting benefits. Dr. Ben Nguyen. Is lead product manager at Transcarent. So Ben, welcome. We're so excited that you're here.

Dr. Ben Nguyen:
Thanks, Kevin. I'm very excited to be here as well. I'll tell you a little bit about my background here. I've had quite the journey kind of getting to where I am now. I'm actually a physician by training. So I went to medical school at USC in Southern California and got my MD.

During that time, it just became so apparent there are so many big problems in health care that you can't really solve at the bedside, right? They're all systemic. And so, I got into the health care tech space years ago, and I went into product management because I really liked building things. I really liked working with really smart people to invent and build new things.

And I've always kind of had this specialization in AI for healthcare. The last couple of companies I've worked with, that's been the focus. So previously I was at a company that built AI products for radiology systems at health systems and hospitals. And now, you know, I'm at Transcarent where we're doing some really interesting things in AI in the healthcare benefits space.

also gotten to do some guest lecturing at USC, undergrad courses on AI every year. And more recently, I've had the privilege of speaking in front of Congress about AI and health care. So that was really interesting. I think it kind of highlights how much a part of the national consciousness this is becoming.

Back in November, Transcarent was one of a few companies that was asked to testify at a congressional hearing and educate our lawmakers about some of the things they need to consider when it comes to artificial intelligence, safety, and how it's impacting health care.

Kevin Fyock:
Yeah, that's really cool. Well, congratulations on that. That must have been an amazing opportunity to be in front of Congress. It's so interesting you talk about radiology, too because I think, for a layperson like myself, radiology is one of those areas that I feel like most folks go to as an example of how AI was used earlier on, so I'd love to revisit that just from a historical perspective if we could.

But the topic of AI and benefits, you know, is incredibly important, and it's funny, within the health and benefits space, Ben, I feel like there's always a topic each and every year that employers talk a lot about, right? So early 2010s, it was all about the Affordable Care Act, and then we jumped to specialty pharmacy, and now things like GLP-1s are all the rage, but it's hard to have a conversation without sort of lending itself to artificial intelligence in some way.

And because this term is so pervasive, really for the benefit of our listeners, maybe you could give a bit of a refresher on artificial intelligence. So at the risk of diving into a graduate-level AI crash course, maybe we could go to help the audience walk away with a baseline understanding of artificial intelligence.

Dr. Ben Nguyen:
No, no, that's great. I love talking about it at a basic level. I lectured undergrads about this all the time, you know, and I think it's just a really important topic for people to just be aware of whether you're in health care benefits or not because it's going to impact our lives, right? When we talk about AI, it's very broad, right?

And it's easy, this happens a lot when I'm talking to other people about it to be talking about different kinds of AI and the kinds of AI can be so different that you can talk past each other and not be talking about the same thing at all. So AI is a really broad term, and some of these concepts have been around for a long time.

But what we're going to talk about today is a certain kind of AI. It’s generative AI, right, is the class of AI we're going to talk about. Dig into that a little bit more later. And within that, we're specifically going to talk about large language models. And this is a kind of AI that can generate fluent-sounding, human-written language.

The most famous LLM, or large language model, product most people have heard of is ChatGPT, right? But ChatGPT is a product. What's powering it under the hood is that large language model.

Kevin Fyock:
I'm so glad you brought up ChatGPT, because that's the other term that, you know, we hear so much. And correct me if I'm wrong, Ben, but that sort of led the charge on the commercialization of these language models, right?

Dr. Ben Nguyen:
Yeah, it was a big breakthrough, right? So just to. Wind back time a little bit when I was at my previous company, we actually use some of the same technology that ChatGPT is built on. Built on an architecture called transformers, you don't have to go into how that works, but transformers kind of let you build these models.

And so, a lot of people around the industry for a while were actually building these small language models. They would do really simple things like they would read a radiology note and they would tell you, “Did this radiologist mention a tumor?” Right? That's a very simple task, right? ChatGPT, which is built by OpenAI, really put these models on the map because what they did was they took those and they fed tons and tons of data from the internet into the language models and got them to actually be able to generate lots and lots of fluent language, right?

So they took it a step further. That happened in late 2022, and once that was released to the public, the consumers, that really kicked off a new kind of renaissance in the AI world.

Kevin Fyock:
Yeah, no, it's fascinating. I mean, my family use it, my young children use it. So it’s just phenomenal technology. I think to many of us, though, Ben, you know, AI feels really new, but as I've done my own reading, that's not completely true. Right? So maybe you can walk us through maybe just high-level history of even sort of lended topic around radiology and using it years ago, but what are some of the recent developments within artificial intelligence?

Dr. Ben Nguyen:
You're right. So the term, the modern term of AI, is actually very old.

It actually goes back to the 1950s. That's when that term started to get popularized because that's when computers started to sort of become a real scalable technology, right? So people like Alan Turing, right, and people knew about, were responsible for sort of like putting forth a lot of these ideas, these concepts, about early artificial intelligence.

Now, from the 1950s, 60s, right, when people talk about that artificial intelligence, they're talking about computer programs where there's rules and those computer programs do things based on those rules. Today, you and I, we would just call that software, right? So that term has evolved, right? So, it started very broad.

Now, when you get into this more modern era of like, say the 2000s, 2010s, we get into this era of machine learning using predictions, right? So predictive models, and that's, that sounds probably familiar to a lot of folks. And that's because when the internet started to really take off and then mobile, the mobile era took off, we just started to get so much data from the internet, right?

We got so much data that was suddenly digital, and you could access that data, right? You could take it and you can build models to do things. So that was the next kind of evolution of artificial intelligence. And the models we were building at that time were things like, you know, on the commercial side, “What kind of shirt would Ben want to buy?” Right? That's like an example.

But in healthcare, what we would do with those models is you might use those models to predict somebody's risk profile, right? For maybe having a stroke. or, you know, going into sepsis in a hospital. In the radiology space, where I was working, you know, we used those models in different ways. One of them is we used visual models that were meant to predict the likelihood that there was maybe a lesion on a chest Xray or on a CT scan, right, that wasn't caught, right? So, these are like narrow, narrow models, and that was like kind of this new era.

Now, we've kind of leapt to this new age of this new renaissance with generative AI, and that's very different than the last couple of generations, right? And generative AI is pretty powerful, right? They are models that are naturally very good at understanding written human language, and they can generate fluent, human-like text in response to that. And they've gotten a lot more powerful even in the last two years. So, so it's very important.

Kevin Fyock:
This is fascinating. And I love the comments you've made around these machines helping humans and either a clinical or business use case. You know, I've heard the term a lot upscaling AI becoming the ultimate tool for upscaling and finding greater efficient. these. Do you fall in that camp Ben? Do you agree with that, that we're not going to be taken over by, you know, sentient robots, but in fact, we're going to be made better at our careers, jobs, and ultimately serving our members and clients and patients better?

Dr. Ben Nguyen:
Yes, I definitely think so. Right? And I mean, nobody can tell the future, but I've seen some of this view represented by certain economists. And I think if you kind of look at kind of the long tail of history, I think this is hopefully where we would go. So these systems are really, really powerful and some economists, right actually think that instead of losing jobs, what they're able to do is they're able to help people who are of maybe a lower skill level, right? A lower level of training. Maybe they're new to the workforce because they're younger. It helps those folks actually achieve higher proficiency in their work sooner, right?

So it brings up workers to the average. Many of these OLM systems, right, might not be super useful to the experts in a given field. But what it does is it actually can level the playing field for people who are new to that field or who aren't as well trained, right? And it helps create this equity.

Kevin Fyock:
Yeah, I love the equity play. And not too long ago, I posted an article on LinkedIn about this. Maybe you had read it as well, Ben, but this idea of upscaling and then this being one of the first examples of a technology that even minority populations are using at a higher prevalence than non-minority populations in. And to me, what an exciting development and opportunity from a diversity, equity, inclusion, and belonging perspectives.

Dr. Ben Nguyen:
Yeah, absolutely. You know, one of the most remarkable things about LLMs when you use one is their ability to transform the content into whatever form factor you want, right? You can use them to transform into other languages to make things simpler to understand, right? Like, and, and making things simpler, getting things to where people understand it, that's powerful, right? That's empowering.

Kevin Fyock:
Ah, fascinating. Okay, so I can keep nerding out on all these questions I have, but why don't we shift to employee health? So, you're with Transcarent. Obviously, you're an expert with an AI. You're also a physician. Tell us about Transcarent. So, how is Transcarent leaning into these language models?

Dr. Ben Nguyen:
So, Transcarent’s mission is to make healthcare very, very accessible, very easy for people, right? And as you can kind of tell from how we've been talking about these models, they are a great enabling technology for doing that. So, if you think about one of the chief problems that any employee has when they're trying to figure out their health benefits, right? They're going to go in and they're going to, you know, let's say I just need to see a specialist of some kind. Well, how much is it going to cost me? Who's in network? What is it that I need to tell them, right? Where do I go? Today, employees have to go gather all that information from everywhere, right? They might have to download a PDF of some kind. So they might have to download something. They might have to read something really long. They might have to call in to a phone treat, wait on line, you know, wait on the line before somebody answers that question.

Getting healthcare benefits is kind of this information gathering and understanding problem, right? And that is something that, done well, LLM products are really good at solving. So, one of the things we're doing is we're using large language models to solve for that problem. So, what we do is, you know, we're working with employers and we take all of their information, like, you know, we take the SPD, the summary plan description, that thing's like 100 pages, right? We'll take their enrollment, their open enrollment guide, which is another 100 pages. We use our tools, we simplify all that. And the employee, instead of having to read those things or email their HR team, we let them talk to an AI assistant, right? That's powered by many LLMs under the hood and a lot of other safety systems. But we let them actually just converse and naturally ask about their health benefits and get guidance on that, right?

And when you think about the paradigm shift that that is, I think it's kind of astounding because the entire kind of problem of getting people information in the digital age is like, how do you get people to go to a website and use this app and that a lot of people aren't comfortable with that, right?

Well, I don't think you need much training to talk, right? Or to just ask, right? So that's the power of using this kind of technology is you kind of remove that barrier of having to like figure out how to use this stuff.

Kevin Fyock:
Okay, so that's fascinating. So maybe for a layman, and I'm asking for myself and maybe to channel my inner fatherhood. I have two young kids and I feel like I use analogies all the time. So you talked about. large language models, you talk about sort of the machines helping to distill down a handful of, you know, pretty complex series of information. Is there an analogy out there to help lay folks really understand how AI can be used this way?

So is there a use case you can compare to even?

Dr. Ben Nguyen:
Yeah. So, I mean, there definitely are, right? I think like, if you think about the world of the internet in 1989, it probably kind of doesn't exist, right? You think about that, like, that was the year I was born, but I'm pretty sure nobody knew what that was, right?

You know, the internet, you know, had been kind of being built up in the late 80s and the 90s, but people couldn't access, access it, right? Then, a few things started to happen, right? So, people invented the web browser, and then they invented Google. What those inventions did was they took this massive complexity of all that information on the internet, and they put it in front of people in a way where it was very intuitive for those people to access it, right?

So now instead of having to understand how to navigate this really complex web of things, people just type something into a search box, right? And that's like, I would argue, probably to this day, how most of us interact with the internet. So, you can kind of think of that as an analogy a bit with how we use LLMs to simplify health benefits, is that you shouldn't have to go to all these places.

You shouldn't basically have to get a PhD in your health benefits in order to go and access your health benefits, right? You should be able to say, “Hey, what's it going to cost me to go to the specialist? Like, what do you think?” And get an answer, right? Like a normal person.

Kevin Fyock:
One, this is fascinating, but it's also validating because I feel like as folks have asked me, your clients have asked me around sort of the sort of use case for the AI models. I oftentimes hearken back to the early days of the internet. And I still remember being younger and folks saying the internet is so powerful. You'll be able to order a pizza instead of having to call, and to think about where we are now and sort of, there's so much potential. So, thanks for that analogy.

So, how do these capabilities then transfer to the benefits world? Like you talked about benefits guides and you talked about getting to care maybe in a more efficient setting. Could you expand upon that?

Dr. Ben Nguyen:
So LLMs can do a few things, you know, the technologies that we're using can do some things that are like really, really exciting. So the one I mentioned is one that's actually not often thought about when you LLMs, they're mostly thinking of chat bots, right? But that's not the only way you can apply them. But LLMs are really good at parsing information that's not structured. So, you know, a medical document, you could just feed in and have it pull out the relevant things for you. That's actually something that they're very good at.

There are a few things that could be very exciting. One is, we could see a world where. suddenlty the management of your care just gets a lot easier because you can just take a picture of your documents. And not only do you save a picture of your medical record, but that leads to the creation of a bunch of reminders.

It could lead to a world where, you know, you upload a physician's note and it is pended to the rest of your medical history, like, and it's easier to manage, right? So that's really powerful. One of the other things is being able to interact with your health care system in a way that does not require you to be an expert in it.

So being able to speak naturally to an AI assistant to not just understand your health benefits, but make a decision around when that's the best fit for you. Because these models, these technologies, allow people to ask things in whatever way they need, right? They can get the information they need at the level they need and help them make a decision.

And I think I would say the last really exciting thing that we're just kind of rounding the corner on now is actually being able to take action by just speaking about what you want. So we've really struggled in the world of healthcare tech to kind of connect different systems, right, and get people to get appointments and things.

The technology underlying the LLMs is very powerful in that, you know, we're, I think we're also going to see a future where instead of having to navigate and click through things, you're probably going to be able to say things like, Hey, can you just find me a orthopedic specialist two miles and make an appointment sometime in the next month and say something like that, walk away and have it be done for you in about three to a couple of days, right? Once they look at who's available, like those are things that are, I think can happen.

Kevin Fyock:
So Ben, that's great. And I think as we think about the application for these models. you know, in and outside of benefits. I think the one thing that I hear time and time again from our clients and different employers and folks in the industry that I speak with is this idea of AI safety and ethics and compliance.

I'd love your perspective on how you're thinking about this topic, especially as it relates to employee benefits.

Dr. Ben Nguyen:
Yeah, no, I'm glad we're talking about this, Kevin, because this is something that is just overlooked so much. I think, you know, if you talk to people in the industry, they'll tell you if they've really tried building these things, it's not easy to build an AI system, but it's about 10 times harder than building it to make it safe. So, it's a really important question.

Well, it starts with culture, right? So, you have to have a culture that prioritizes that safety. And most importantly, in your teams, you have to have subject matter experts, right?

So we work with clinicians. We work with benefits experts, not just in the testing phase, but they help us design some of these features, right? They have a hand, they have a voice. and a seat at the table when we are designing these AI features in order to ensure that we're within the safety guidelines.

Because, you know, when it comes to the employee space and the employee healthcare space, you can think about the things that could go wrong, right? An innocuous one might be the AI tells you the wrong deductible, but a not innocuous one would be that, you know, the AI tells you that you owe a bill that you don't owe or that you have, you know, a health condition that's very serious that you don't have, right? So, when you're in the domain space, if you have the right expertise around that domain space, then you know that you can guard against those things, right? You can build your systems to shape themselves to guard against those. And that's what we do.

So when we build our safety systems, right, these aren't systems you can pull off the shelf because AI safety systems are specific to the domain that they have to make safe. So, we build these things in house, we use the expertise of our people, and we, you know, seek to really deeply understand the things that can go wrong when it comes to healthcare benefits and using AI for it.

We put the guardrails in, we figure out like, this is a place where we should not be using AI at all in benefits. And this is okay. We have to also work with the people. It's not just AI. There are humans who are in the loop for many things, and we have to work with our experts to ensure that there are really well defined times and places to escalate to these humans.Like if somebody's having a medical emergency, right? So those are all the things you really have to consider when you're thinking about AI safety because it really is domain specific. Really got to deliberately design that into the process.

Kevin Fyock:
It's interesting, Ben, you talk about the flubs, right? And there are some in the non healthcare space that are a bit humorous.

I think my favorite is a soccer match where there was an AI agent that followed around what they thought was a ball, but it was a referee. So, some of those are funny and clearly we'll get there. And maybe that's a good segue to, you know, asking you a bit of a crystal ball question. Right. So, I mean, this was fascinating just to get to hear your perspectives on the use cases and the history.

But with these models continuing to evolve, I'd love your perspective, call it 10 years from now from an employee benefits perspective. How are we going to leverage these models? What does it look like when we come back here in 10 years? What will you be talking about?

Dr. Ben Nguyen:
Yeah, I think it's going to be an interesting world in 10 years because things have been exploding. I'll caveat this as saying I've been in the AI industry for years now. There are things that have happened in the last 18 months that I expected to happen in 10 years. Things are moving very quickly, but I think by and large, the pattern that I've always seen is that, you know, with benefits and healthcare, the consumer always kind of leads.

You kind of saw this in the mobile era, like people were using iPhones. And then people were in the hospitals, you know, doctors would say, why shouldn't we use mobile apps for things too? The consumer world leads the enterprise world, right? And rightfully so, right? Because there's a lot at stake. Now, I think what's going to happen is that in the consumer space, people are going to get used to expecting very smart experiences in their products and their consumer interactions.

So it might be that they expect to always be able to chat with a chatbot. It might be that they expect that, look, anything that I take a picture of should be cataloged and tagged for me for work because that's how it works in my consumer products, right? So, I think consumer expectations will rise, right?

And that trickles down into the expectations of people that people have around the work that they do. The ease of that work, right? They're going to expect to be able to use tools like this. They're going to expect for their products, for their benefits to be as smart as any product that they get off the shelf. You know, I think that's really kind of where we would be heading.

Kevin Fyock:
I love that. I mean, we've even seen that with consumer tools, not healthcare, right? So, The phones we use, the smartwatches we utilize, it sort of reset an expectation every time we see an update, every time we see a new piece of hardware or software.

So, Ben, I could talk to you for another 30 hours, let alone 30 minutes. So, this was awesome. I can't tell you how much we appreciate you coming to the show and lending your expertise. This has been a lot of fun, and we'll look forward to following your journey and Transcarent's journey in many ways, like really focusing on cutting-edge artificial intelligence topics. So, thank you.

And to our listeners. Thank you so much for tuning in. We hope you enjoyed How Genitive AI Is Making Benefits More Accessible and Empowering. And this is the second episode of “The Stretch,” a podcast dedicated to the ideas that are revolutionizing the world of workplace health and benefits.

If you enjoyed this episode, we encourage you to subscribe to the podcast and follow us throughout our season. We hope you'll join us next time and we'll see you soon. Thanks.