Clips
News
More from theCUBE's Coverage of Google Cloud at HIMSS25

Ted Slater

Managing Principal Scientific Informatics

EPAM Systems

Aashima Gupta

Director, Healthcare Solutions

Google Cloud

Why AI in healthcare hinges on a solid data strategy

Artificial intelligence is only as effective as the data behind it. Without data that’s findable, accessible, interoperable and reusable — known as FAIR data — healthcare organizations risk inefficiencies, inaccuracies and unreliable AI-generated insights.To unlock AI’s full potential, a well-structured data strategy built on FAIR principles must come first. AI alone can’t solve all data challenges; success hinges on a strong data foundation, according to Aashima Gupta (pictured, left), global director of healthcare solutions at Google Cloud.Epam Systems’ Ted Slater and Google Cloud’s Aashima Gupta talk with theCUBE’s Rebecca Knight about the role of FAIR data in AI adoption

play_circle_outline AI hype vs. solid data strategy for healthcare innovation.

play_circle_outline Leveraging AI agents for multi-step healthcare workflows.

play_circle_outline AI role in implementing and improving FAIR data principles.

play_circle_outline Exciting changes and transformations in healthcare innovation at HIMSS 2025.

Info
Transcript

Ted Slater, EPAM Systems & Aashima Gupta, Google Cloud

Ted Slater

Managing Principal Scientific Informatics EPAM Systems

Aashima Gupta

Director, Healthcare Solutions Google Cloud

Ted Slater, managing principal scientific informatics knowledge engineering at at EPAM Systems, and Aashima Gupta, global director of healthcare solutions at Google Cloud, join theCUBE at HIMSS 2025 to discuss the future of healthcare innovation with a focus on AI and FAIR data principles. The interview, conducted by theCUBE host Rebecca Knight, delves into the critical aspects of FAIR data — findable, accessible, interoperable and reusable — as essential components in unlocking AI's potential, according to Slater.

Key takeaways include the necessity... Read more

explore Keep Exploring

What factors are important to consider when developing an AI strategy within an enterprise environment? add

What are some examples of successful use cases in leveraging FAIR data in healthcare organizations? add

What factors contribute to the difficulty of implementing the FAIR principles and how can AI, specifically generative AI, help in this process? add

What potential benefits or implications does AI, especially in combination with knowledge graphs, have for society and technological advancement? add

bolt Powered by CUBE AI

Ted Slater, EPAM Systems & Aashima Gupta, Google Cloud

Rebecca Knight

>> Hello everyone, and welcome to theCUBE's coverage of HIMSS 2025 in Las Vegas. I'm your host, Rebecca Knight. We are here with Google and many of its ecosystem partners to talk about the future of healthcare innovation and agentic AI. We have got a great lineup of guests for you. And with that, I would like to welcome our first guest to the show, Aashima Gupta. She is the global director, healthcare provider, health plans, life sciences and strategy solutions at Google Cloud. Welcome, Aashima.

Aashima Gupta

>> Thank you for having me here.

Rebecca Knight

>> And Ted Slater, managing principal scientific Informatics at EPAM Systems. Thank you so much for coming on the show, Ted.

Ted Slater

>> Thank you very much for having me.

Rebecca Knight

>> So Ted, I'm going to start with you.

Ted Slater

>> Sure.

Rebecca Knight

>> You say that there can be no AI strategy without a solid FAIR data strategy. So in fact, you co-authored a paper if published in the journal Nature on this very topic. Why don't you start by telling our viewers what is FAIR data and why you believe it is so critical to unlocking the potential of AI?

Ted Slater

>> Sure. So FAIR data is a set of principles. So we came up in some years ago now, but FAIR is an acronym, F-A-I-R. It stands for findable, accessible, interoperable and reusable. And what those items mean is your data need to be findable. If you can't find them, then you're never going to be able to use them. There are certain things that you can do to make sure you can find those data and we'll talk about those. The next bit is accessibility. Once you find them, if you can't access them either by machine or if you're a person, then you're still in trouble. Interoperability is super important and if you walk around here at HIMSS, you'll see a lot of organizations are interested in interoperability of their data and making sure that that's possible. And finally, the ultimate goal of FAIR is reusability. And what that means is you've assembled the data for something that you want to use it for, but there may be things that you haven't even thought of that you want to have your data ready for, and that's what reusability is about. Making sure you can use the data for things that you haven't thought of yet.

Rebecca Knight

>> Well, we're going to dig into all of those things in the next 15 minutes, but Aashima, there is a growing misconception that AI can solve all data problems, in fact, maybe even all problems and stop. What would you say to leaders who assume that AI is the thing that can fix all their data issues unless they have a solid strategy in place?

Aashima Gupta

>> Rebecca, you're touching on a great point. There's a lot of AI hype, there's a lot of AI washing. All night all the data projects have become AI projects, AI projects have become GenAI projects, and GenAI have become agentic AI projects. But going back to enterprise has diverse starting points, data matters, data strategy matters, and I would say there's no GenAI or Gentic AI strategy if you don't have the data strategy. There are different interoperability, semantics, syntactic, all that matters. And you'll see different use cases deploy a different data strategy in general as well but you need to have that foundation.

Rebecca Knight

>> Ted, going back to you, AI and FAIR data strategies can be hard to put into practice because a lot of organizations ask the question, what's the ROI? What's in it for me? What's the return on investment here? What do you say to them?

Ted Slater

>> I'll tell you what I say to that. I say that's just the wrong question. The right question is how much is it costing you to not have FAIR data? Again, the ultimate goal of FAIR is reusability. And if you can't use your data for the things you need them for now if things are popping up for which you need to retool everything and make your data anew, that's just not going to work for you. It costs too much money to make that happen. So that's the answer.

Rebecca Knight

>> So what are some practical steps that organizations can take to integrate FAIR data practices?

Ted Slater

>> This is a great question. There's a couple of things to talk about. It turns out one of the things that gets expensive is waiting too long to verify your data, so there's-

Rebecca Knight

>> Verify your data, I like that. Okay.

Ted Slater

>> Yeah. That's the word's. It's verify. So the longer you wait, the more expensive it is. So try to start early. In the community, we call this born FAIR. The other thing that we're concerned about is making sure that we're focusing on being FAIR enough. And what that means is almost no one has data that are 100% FAIR if that even means anything. So don't try to get to 100% FAIR. Just make sure you're doing FAIR to the extent that you can support what you're doing with your own operations and stop there. If you need to do more someday, then you can do it and you'll be in a better position than you would've been before. But you don't need to go all the way to do that. So that's the thing to think about. Those two things, starting early and then making sure that you're maintaining what you need to do as you go forward without doing too much, because then it's too expensive.

Aashima Gupta

>> And I would add to what Ted is saying and which is oftentimes people are building this data strategy for years. So build it and they'll come. While that is important, it's also important to balance. Show the value much faster when you're taking the FAIR approach, because if that's five-year strategy and you're not spun out any use cases, people will lose that vision. So I would say always balance that near-term use cases built on that longer-term data strategy. But you need to show the value earlier on to get the stakeholder.

Ted Slater

>> Can I be a little provocative about that?

Rebecca Knight

>> Please do.

Ted Slater

>> Okay. So I'm a computer scientist as well as a molecular biologist. So use cases matter to me in the context of FAIR, I'm not a huge fan. Use cases are good to help guide what you're trying to do, but when you're talking about reusability, and remember reusability is about making sure your data are useful. Even if you don't know what they're using them for, that's not a use case. So not everything is guided by use cases and you need to be careful about how that works.

Rebecca Knight

>> I wonder if AI itself could help play a role in this process. I love what you're saying about FAIR enough. It also makes it a little less daunting for organizations too. But can AI itself play a role here?

Aashima Gupta

>> Especially now with agentic AI, we are seeing a shift. It used to be get all your data, map it, transform it, harmonize it, normalize it, put it in cloud, and then build AI or data use cases. With agentic, it's much more local data processing data stays in place. Again, not all use cases are built for that, but especially for agentic AI, now we have agent space, that can connect to your data in place. So that means data can be in different enterprise environment and you can still build agentic AI experiences.

Rebecca Knight

>> So Aashima, even though Ted has just said we shouldn't pay any attention to use cases, I'd love to talk to you about some use cases because the Google ecosystem has really helped driving this execution in strategy. Can you tell our viewers about a specific example of a life sciences organization or any other kind of healthcare organization that is successfully leveraging FAIR data here?

Aashima Gupta

>> So I would say I think use cases do matter a lot so we can defer. I think if you don't, to me, even roadmap matters. If you don't know where you're going, you might not get there. So what is your technology strategy? What use cases you would like to first? So one example is we are working with MEDITECH, they're an EHR company, and what they have done is they built search and summarization of that data saving 7 1/2 minutes per patient for the visit preparation. Of course, they had to normalize the data, they had to standardize prior, but they were able to light up that use case much faster. Now, that's not the be all and all of their data strategy they're going to continue on, but they're seeing the value much faster.

Rebecca Knight

>> One of the things you've talked a lot here on the show about, Ted, is the problem of interoperability. How can FAIR data principles and AI work together to improve data sharing across systems?

Ted Slater

>> This is a great question and has much to do with making sure that we can use AI to help implement those FAIR principles. Interoperability has a lot to do with domain expertise, whether the domain is scientific or anything else. So order to build a model of the world that represents the concepts and the relationships between those concepts in whatever domain you're in, you have to have some domain expertise and some pretty special skills to do it. That's one of the reasons why it's hard to implement the FAIR principles. That's one of the places where things like generative AI can really help because you can use generative AI to start from something like journal articles, or patient data or anything you can think of and use the generative AI tools to pick out concepts and relationships and properties from those data or from the content, and create an ontology for you. There's a good chance it'll have to be tweaked in the end, but that's one of the ways that AI can really help you move forward in this space.

Rebecca Knight

>> So let's dig into that a little bit. So tell me about how a practitioner might use that information that he or she is getting from that journal article and put it into practice in terms of delivering care or coming up with some other kind of life-saving device or medicine?

Ted Slater

>> Sure. So the way that it typically works is it can be as simple as you sit down in front of your computer and you upload your favorite patient data or even an image file or journal article or what have you. Then you can do some clever prompt engineering to make sure that you're guiding the large language model into the right semantic space to help you create an ontology. And then you just say things like, "Please generate for me the concepts you find in these data, the properties, those concepts can have and the relationships those concepts can have between each other." And it will do it and it will do it pretty well. And then the next thing you can say is, "This is great. I would like to see this in RDF," which is a semantic web standard for representing ontologies and similar kinds of things. And again, the large language model will take that list of classes and properties and relationships and turn that into proper RDF. So now you've got a proper ontology. And the next thing you can do with it is do things like mapping the concepts in that ontology like a patient or something like that to a concept in a third-party publicly available kind of ontology like SNOMED for example. And once you've got that kind of relationship, now you know that when you see patient in your data, if you're not actually using SNOMED's ID for what a patient is, you at least have a map to that. Now your data are much more interoperable than they otherwise would've been. So it's as easy as doing that. That's how it goes.

Rebecca Knight

>> Excellent. One of the things you've talked about Aashima is AI agents and they are becoming more prevalent and they're gaining a lot of traction in healthcare. What are some of the most promising use cases that you see in terms of how AI agents can be used and what role does their data play in helping them realize their promise?

Aashima Gupta

>> As we move into agent enterprise, what we are seeing is the difference between, I said gene AI and agent tech AI. What makes agents special is their reasoning capabilities, their deep research capabilities, their sophisticated multimodal understanding, and they are able to execute a task. So as an industry, we are going to see... Actually, it is not just healthcare. We will see a move from simple chatbot that we all are using writing an email, summarize this doc, all the way to these agents who are executing the task. They are planning the steps and executing on a goal. And in that you can think of claims operations, you can think of revenue cycle. So whatever is a multi-step workflow, that's where AI agents will move from a simple chatbot to a single agent to multi-agent orchestration. And all of that is required. If agent has to do the job, it needs to connect to your data to be able to service that part of the workflow. So again, going back, all this builds on having a robust data strategy, but it's a paradigm shift. Agents can actually connect to your data sources and where they are as well.

Rebecca Knight

>> They need to be FAIR enough, as both have said.

Ted Slater

>> Can I make a quick point about that? So one of the things that we found when we were working with large language models and those kinds of things is turns out large language models hallucinate. I think everybody figured that out. And it turns out that that's a baked-in property of transformer networks, which are behind large language models. So that's not going to go away. And what we found was it's really nice to make simple agents that know how to go do things in the world that perhaps a bench scientist or someone working in a clinic would not necessarily know how to do, like write Python or access an API at some database. But these agents can do those things. You can use an LLM to interpret a natural language prompt to go and marshal an agent to go out in the world, do what it needs to do, and return data that are not hallucinated because there's no LLM doing those things. That's a really nice thing to do. And the other way to think about it is this is very, very natural for people. All day long we think with this neural network, but when we need to look up our friend's phone number or something like that, we go to our phones. Our phones have explicit semantics, things are represented explicitly in there and we can look them up directly. That's not the way it is in a large language model. There's no place I can point in a large-language model to find my phone number or anybody else's. So that's kind of a nice way to think about it is explicit semantics and large-language models working together in concert, much more powerful than either one by itself.

Rebecca Knight

>> Empowering scientists to do much more.

Ted Slater

>> For sure.

Aashima Gupta

>> Yeah. I would just add to Ted's point, you touched upon hallucination, how do you reduce that? You ground the responses on your data corpus. So we have search as an example, but for that, that means the model is connected to your data. So that is grounded on that, which is citable referenceable, and that is a context and then goes to LLM, say, based on this context, and it is called in-context learning. So it's a combination of search and LLM going together, search referencing the enterprise data, grounding it, then going to LLM and making a much better response from the hallucination perspective.

Rebecca Knight

>> Very exciting. Ted, I'm going to give you the last word. Here we are here at HIMSS, which is the epicenter of healthcare innovation. When you're walking around this show floor, which is heaving, I have to say, what do you see as the most exciting changes, transformations in this space? You've been in this industry for a while now. What do you see as the most promising and most exciting?

Ted Slater

>> Obviously, I have to mention AI as something like that. I mean, that's super exciting and I don't think we've even started to figure out what that's going to do for us. So I think that's great. You can go a bit too far with it, but if you do a Google search and you look for example, AI and knowledge graphs, like what we were talking about before, you'll come up with an awful lot of information about that combination of explicit semantics and large language models. I see a lot of that happening here. That's a great thing for us to see. I'm really excited about it.

Rebecca Knight

>> Great. Ted, Aashima, thank you both so much for coming on the show.

Ted Slater

>> What a pleasure. Thanks for having us.

Aashima Gupta

>> Yeah. Pleasure. Thank you. Thank you, Ted.

Rebecca Knight

>> Stay tuned for more of theCUBE's live coverage from HIMSS 2025. I'm your host, Rebecca Knight.