AI in Hospitals: Building the AI-Native Hospital

Chris Hutchins: Welcome back to The Signal Room. I'm Chris Hutchins. Quick thing before we start. I wrote a book. It's called Beneath the Signal. And it's about the human work behind trusted data and responsible AI in healthcare. Search Beneath the Signal on Amazon now and pick up your copy. Okay, here's our episode. Today we're exploring what it takes to build an AI-native hospital. And what is that, you say? We'll get to that. What changes when AI becomes part of how a hospital coordinates its work? What does that actually mean for the people delivering care? Joining me today is David Niewolny from NVIDIA's healthcare and medical business development team. David, welcome to The Signal Room.

David Niewolny: Thanks, Chris. Excited to be here today.

Chris Hutchins: I've been excited about this conversation. Why healthcare? Why technology? Seems like an interesting space, but you probably are gluttoned for punishment because it's a tough, tough spot when you put the two together.

David Niewolny: The great news, it's definitely been a good long-term career choice because as we know, nothing in healthcare happens exceptionally fast. But I mean if you go back, I've always been a technologist at heart, loved technology, and I'd say about five years into my career, really looked at, you know, what is that why? And it's frankly using technology to help people in a way that healthcare touches all of us. And it always feels like it's one of those areas that, you know, frankly, in terms of technology adoption, ends up in some ways getting neglected. And it kind of made that my personal mission to try to take whatever that latest greatest technology that we end up seeing pervasive in a number of other industries. And how do we bring those use cases into impacting something that is truly personal to each and every one of us? And that is healthcare. It Right. I mean to give you a little bit more background on that, you know, it really started with wireless technology and you know and the kind of start of the internet and how do you actually begin capturing data from all of these medical devices, you know, with this general premise that at someday in the future you would begin to derive some sort of intelligence out of that. And Chris, I think you highlighted little did I know that would be a 20-year journey to get Right. Here. But I think it's you know frankly safe to say that we have Finally, in a world where today we've kind of built all of these pieces. It's like we have the communication and secure technology to get data from medical devices, electronic medical records, and others. So we've kind of gone through this process of digitizing healthcare. And really, this next phase now is taking that and deriving true, meaningful intelligence. And I think as we start talking about AI native health systems. What we're going to talk about today is taking that intelligence and frankly making it seamless and really supporting Right. All of the clinicians, hospital administrators, everyone within that healthcare system to essentially give them superpowers. I mean, it's designed to augment a lot of the care that they're already providing, to make them much more efficient, much more effective, which at the end of the day improves access to care by allowing care for more people. At the same time, has the significant ability to lower costs, which always ends up being a huge topic when you start talking about healthcare.

Chris Hutchins: For certain. But let's start with the real basics. When you say an AI native hospital, what does that mean?

David Niewolny: We think of an AI native hospital. This is a hospital that is truly using the intelligence that it is creating. That comes from patient data, that comes from operational data, that, as we talked about, has been digitized. But at this point, it is not necessarily creating a benefit to the nurses, the doctors, the surgeons, the hospital administrators. Think of it almost as a database. They have data. But still you have this human component where whether it be a clinician having to analyze radiology images, whether it be a hospital administrator trying to plan their schedule for the next day, that data today is being acted on by humans, which requires a whole lot of humans that are essentially employed by that hospital system to potentially, well, actually provide care. At the end of the day for patients. That's where you go to a hospital for. But as you talked about, we've introduced by digitizing a lot of this a whole lot of new workloads and workflows that are not directly associated with providing patient care. So if you look at kind of the end goal of what we're trying to do here is really allow everyone in that hospital focus on providing better care to more patients at the hospital with all of this newly digitized information, as opposed to Adding to their workload, which to this point, Yeah. Arguably I'd say it has. And I thought you brought up a pretty good, you know, a couple of pretty good examples. I mean, a lot of the clinical documentation, but like I said, it goes across the entire health system. I mean, you talk to hospital administrators, and frankly, just the fact of planning schedules for the next three days with frankly a lack of true information. Yes, you might have a spreadsheet. But is it really using that to make recommendations? I mean, if you look at a hospital again today, how much of that information is documented by using whiteboards, clipboards? Right. You know, we talk about digitization, but there's still a lot of manual work that happens in a hospital agent every day.

Chris Hutchins: But l let's kind of talk about some some of the practical things that come to mind now when we're talking about the challenge with this overwhelming amount of data, some of its necessary, some of its noise. What are some of the things that you think about that a hospital should be doing now kind of In preparation for maybe some pretty innovative things that are gonna be game changing in terms of h what this impact looks like as you start to gain some traction.

David Niewolny: And maybe to take a step back before we dive into kind of what a should a hospital be thinking about. I mean, just kind of defining that idea of AI native. And I'd say, you know, if you've seen in very short order, I mean, again, NVIDIA's been involved in healthcare and AI, you know, for the Yeah. Better part of 15 years. Almost all of that work was initially anchored in medical imaging because a lot of that data was essentially readily readily available, high compute loads, a lot of computer vision. And you know, I think famously it was said that you know, radiologists would essentially become extinct. And I think, you know, we've seen that play out, and there's now more radiologists than ever, because now the efficiency that's being created has created a need for more medical imaging, which is arguably one of the most effective diagnosis tools. Which then drives the need for more radiologists. And you look at the way AI has evolved from what is initially was, you know, perception, then we moved into generative very quickly after that. And now Right. I think we're in probably the most important area, which is essentially agentic AI, and then moving into physical AI, which physical AI is truly that embodiment of these agentic workloads, which now really makes I'd say AI employable. I mean, these are now essentially tools that each and every one of these clinicians, hospital administrators, surgeons, radiologists can begin using to make themselves far, far more efficient. So when we talk about AI native, we're really thinking in hospital that is adopting technologies that are both agentic and then also in the idea of physical. And when you think of physical, start thinking about the idea of physical robots or robots that we'll say roboticizing medical devices. I think a great example is, you know, in the world of you know X-ray, you still today need an X-ray tech to set up that X-ray. And for those of you who've had an X-ray recently, it's essentially a giant robot. You walk into a room, X-ray tech sets that up. You know, we look to a future where essentially you can walk in the room, talk to a digital agent. That digital agent can guide you in terms of how you should be positioned. And that medical device can then actually analyze your position, set itself up, take the image, analyze the image to make sure that it's a quality image that the radiologist can actually use before you even step Right. Out. You know, I didn't even realize that, you know, one of the the I'd say main causes for patients having to come back is image quality. You know, you might not have the actual right image. And now kind of building these AI skills from think that you know robot that's actually setting itself up to the agentic solutions on the back end that are actually analyzing that image for the quality, making sure that again, your overall doctor's visit goes as smooth as possible. And then if you think about taking that a step further, in many cases you need to spend you know a couple days waiting for the radiologist to then analyze that image. And all of that can be done essentially in a single workflow in the time of your overall doctor's visit. So you can really kind of see, I mean, through one clear example how that sort of efficiency is being brought into place. And that's just one of many examples that we could we could talk about.

Chris Hutchins: And I love it 'cause it kind of gets into this the whole fear aspect a little bit that people talk about, you know, what's gonna be replaced and what's not. You've been doing working in this space for a l for a long time, so you've seen a number of different advancements come in and go. And I'd be curious if you can name any significant role that actually just disappeared rather than it actually got expanded and God enabled it at a much higher level and advanced the ability to understand and diagnose and you know, a variety of things. But it seems more more likely that it's upping our game than it is actually just replacing something.

David Niewolny: And I cannot think of a single example. And again, if you look at AI as a whole, there are going to be areas that can become automated. But I do think when you start looking at healthcare, one of those areas that is the most personal aspect of care that I think, you know, any of us you know, we're all familiar with, but we'll ever be a part of, we'll always have you know, human intervention. I mean, there is absolutely no way that human intervention is going away in that particular case, but it does come back to providing those, you know, efficiencies. I mean, at the end of the day, these AI native hospitals, AI models, AI applications, they're designed to give time back to doctors, nurses, patients, families. This is not something that is designed to frankly replace any of that. And, you know, the story that probably isn't told enough is just in general that access to care. I mean, these hospitals, if you think about it, AI native, our end goal should be how can you see double, triple, quadruple the amount of patients that you're seeing today? And we do know with a lot of these nationalized healthcare systems that there's many cases where patients based on need and priority are left waiting. And you know, that's a situation that I do think we're starting to come with this convergence that is truly you know, frankly, hitting a wall where you have staffing shortages in many of these areas. Yeah. That's All of these hospitals right. Are overloaded. They're operating on exceptionally thin margins and expected to provide an exceptionally high quality of care. Great news for all of us is I think technology is coming in at the right time to Yes. Really help augment all of these clinicians and the work that they're the work that they're doing.

Chris Hutchins: No, I couldn't agree more. I mean, there's a couple of glaring examples that I've seen in re in recent years. One, it's like I had been around I grew up around a cancer center when was a kid and I hadn't been in one for probably twenty years. And I went into one when I was working in New York and they sh they took me into the radiation area and showed me the gamma knife and explained it to me. Just when I started to understand what they were doing with that technology, just they're able to map tumors in such a precise way. They literally have hundreds of thousands of these little like electronic gadgets they're I don't know how what the right word is, but they move very, very precisely so that they're targeting a tumor and they're protecting and shielding with this technology cells that are completely healthy. I'd never even heard of such a thing until they were explaining it to me. I'm like, that is far beyond anything I've ever heard before. So I mean it's a definitely it's up upping the ability to be accurate in a specific area like that. Even in a more personal way, when you talk about personalized medicine, I've g talked to some guys that are doing some great work out at Stanford where they're providing real-time coaching to the c the care team that's taking care of a patient. And they're picking up things in the patient's cues to understand how they're feeling, you know, whether they're anxious or whatever. And the entire thing is getting really, really great reviews from the clinical teams and from the patients because they feel like it's becoming much more personalized and it's precise. And the ones that need more time it takes into account that they need more time. And they actually adjust schedules as well. I mean, it's just crazy how much is possible already.

David Niewolny: I think those, I mean, two really key words you mentioned. I think, you know, personalized and precise. And I mean just kind of coming into some Yeah. Other examples of, you know, what could exist in an AI native hospital. You know, we're working across the board with a number of different providers of surgical technology. And just the idea of putting together preoperative surgical plans. Again, a lot of that today is done Very, very manually, basically taking data from electronic medical records, patient notes. Also, a key piece to that is all of the medical images to actually begin kind of putting together a plan for that surgery. And now using a lot of these AI and agentic tools, you can essentially now pull data from all of these sources, use reasoning models to begin actually doing the same clinical reasoning that a surgeon would do. And the nice thing is, you can actually begin training these models off of what the world's best surgeon would do, which then up-levels the skill set Right. Across the entire surgical profession. So you can put together these kind of pre-procedure plans, and then you also look at a lot of the work that we're doing in AI and simulation. You can then begin simulating these procedures where a surgeon can actually Go through that procedure a number of times prior to ever operating on a patient. So you essentially get kind of a you know world's best pre-procedural plan. You get a surgeon that now has the tools through simulation to begin essentially practicing the surgery ahead of time. I mean, the quality of the outcome and frankly the speed and efficiency at which that surgeon can operate has. No way to go but up with much more positive outcomes. So it's again a lot of really, really cool applications we're working with across the board.

Chris Hutchins: I love that part of it because I think we've historically seen people kind of forget that medicine i the way it's actually referred to, th it's a practice. There's an expectation that it's gonna be perfect or whatever. And we know it's a hu it's human beings, so there you there's no way it's gonna be perfect. But what you're talking about is introducing the ability to break really simulate and do something in a totally on a whole totally different scale.

David Niewolny: That's one of the things, I mean, you'd be I'd just say surprised in terms of some of the advances that have been made since the introduction of robotic surgery. I think you've seen Yeah. The growth of robotic surgery Great going example. Up significantly. I think now we're saying, you know, roughly twenty percent of surgeries are done robotically, and we definitely see that increasing. And in those particular cases, you now link, you know, essentially what the endoscopic cameras are seeing with the kinematics data. Of what the surgeon's doing, and you're beginning to get a very good idea of outcomes. And I think there was a recent study that I saw at the Surgical Robotics Conference done by Andrew Hung that actually linked the kinematics to the outcomes of the surgery. So you could basically see, based on how the surgeon was moving their hands during that surgery, there was a direct correlation to the patient outcomes. Which definitely now proves there's a trainable data set to begin giving guidance to these surgeons. Yeah. And if you think about it, I mean, we've definitely indexed heavily on some of the surgical applications, but that sort of application can apply anywhere in the hospital. I think, you know, one of the other areas we're seeing in these ideas of AI-native hospitals is, you know, generally around the idea of spatial intelligence. How do you begin kind of putting eyes and ears in the patient room, in the surgical suite, in the radiology suite, and frankly throughout the entire hospital to begin giving guidance and using that now very contextual data to really, you know, provide much better plans for the entire organization?

Chris Hutchins: Right. Now you're touching on some of the things that actually got me motivated to w to take one of the jobs I had a few years back. I went to New York and the attraction was was for me to work in one of the biggest health systems in the country that had by far the most heavily concentrated population centers of people groups from all over the world. And they didn't have a robust set of platforms to be able to har harvest that data and r really to be able to produce the kind of impactful change that it could have. And you know, we're talking about something that's compl a complete game changer way beyond that even. Imagine having the ability to have the one of the top ten surgeons in the world that's seen more cases than anybody alive be the one who's directing this the service. That kind of stuff is possible now. And it's getting to the point where now it doesn't even matter that you might be in a rural area that doesn't have the the technology infrastructure that you would be ideal for someone to be able to do that procedure on site. But I mean it it's leveling the playing field in a in a way that really the Hippocratic Oath I think is being applied by technology companies now even more so than they probably think about. But it rul that's really what what's going on at scale right now, seeing it and a at a level I haven't before. So I'm really excited about to see where it's going.

David Niewolny: Yeah, it is fantastic to really see these hospitals, I mean, truly start going down this journey of transforming into an intelligent physical system, you know, one that Yeah. It can listen, it can see, it can coordinate, it can actually capture personalized data from the patients. And in those cases, you know, the clinicians, surgeons, radiologists, hospital administrators remain in control, but they now have not just access to data. They're actually getting guidance that has been reasoned through by Yeah. Essentially the same sort of model that a leader in their profession would take, which as Yes. I mentioned, I mean it has that fantastic ability to really just uplevel the skill set of all clinicians.

Chris Hutchins: I yeah, and we we're increasingly in a in a s a world that's shrinking, and I don't mean that, you know, to be funny. It's just that Mm-hmm. It yes, it's a small world and w with the technology and the data that we're seeing being able to be aggregated and utilized, we're making things much, much easier. I say it like it's not that big of a deal, but i it is. But the complexities that it takes to be able to get to what is the best practice for a particular procedure. A lot of that historically depends on the area and the geography and the culture even in the region. There could be the weather patterns. There's lot of things that impact that stuff. And this concerns that I've always had is I don't want us to get stuck if we're in a big metropolitan area. Don't think we've actually solved the human model for anything. We haven't. I mean, population centers are different. They are different for a variety of reasons. Belief system has th has things to do with it. You know religion has things to do with it, geographic, geography, all these different things are are are things that have to be taken into account. But having the ability to go at it to scale and make and connect like we're talking about here. To me, that that's the part that gets me really excited because we are literally going to hopefully take down some of the barriers that people are concerned about when it comes to having equity, when it comes to what kind of care and access do people have access to.

David Niewolny: For sure. And just talking about, you know, I'd just say that social acceptance. I think it's one of the initiatives that, you know, we've taken on as a healthcare practice. And I think you've seen just across the board from NVIDIA, we're huge proponents of open models and open model architectures. I think, you know, there's generally concern when you think about AI as some sort of black box. You know, how is that model trained? What data was it trained on? What are the model weights? You know, how is it actually coming up with this information? And I think, you know, that is one of the most important pieces, is specifically in the world of healthcare, which we talked about being very personal, this idea of open model and open model architectures is an initiative that we are full force behind because we think it's exceptionally important to have all of these AI models and their associated applications essentially be explainable. You need to have these explainable. I mean, I think one of the great pieces about healthcare and med tech, you know, a lot of people do give little bit grief about the regulatory agencies and not allowing progress as fast as it should. But I guess I would argue on the ancillary and say their goal is to make sure that all of these technologies are safe and effective. And you know, that burden falls on the med tech organizations and the technology companies. That are building those to ensure that these are explainable, safe, effective, and the regulatory agencies are essentially going to do their job. And it's one of our, I think, key roles in the ecosystem goes along with that education and frankly helping explain this technology, how it's making an impact, how it was built, how Yes. It's going to be deployed. And then, yes, it's going to go through the same sort of clinical trials that any sort of medical device. Would go through, which again gives me confidence that yes, I think, you know, healthcare and med tech has moved slower, some of that due to regulatory, but not all of it. A lot of it just is a mindset shift. A lot of these, you know, historically, you know, med tech and health tech companies have wanted to build everything from the ground up. They've seen themselves as innovators. And there's no doubt they are innovators. But similar to a lot of the discussion, we're Having today about essentially giving these superpowers to clinicians, a lot of what NVIDIA is doing is frankly accelerating the development of all of these new technologies, both software and hardware and robotics, so that these products can now get to market faster and they can actually have and begin those clinical trials earlier. So we can start seeing this technology at a much faster pace.

Chris Hutchins: So what I l love to have you talk a little bit about is some of the ways that you're seeing and you're thinking about what role should it be taking that maybe it's not yet, or maybe some places where you're seeing it maybe surprise some people because it's able to take on something that has been largely human and really cumbersome. But that you're seeing, you know, really promising results in.

David Niewolny: I can't say there's anything anymore that's truly surprising to me. You know, you start, you know, immersing yourself in the technology and seeing the capability. I mean, it is, you know, pretty amazing. But, you know, just to highlight, you know, a couple use cases, some of them seem more simplistic, but will make, you know, I think a really big impact. And then I'll maybe take us down a journey of I think, you know, where I see us going in kind of five and 10 years, but I'd say probably one of the most significant clinical impacts right out of the gate has been a problem that, you know, frankly has been tried to be solved for the better part of, you know, 10 years prior to today, which was clinical documentation. You know, everybody talks about pajama time from radiologists for all of these, you know, primary care doctors. You know, the fact that they have to go home and document and they're documenting based on memory. Of what they did that day because they were essentially back-to-back-to-back visits. You know, looking at the way the technology has now evolved, you can do absolutely fantastic speech-to-text transcription and now reason through summarization to provide fantastic clinical notes. And that was something that, you know, frankly, had been worked on, but I think we all know that specifically for healthcare, you don't want something that is 70% there. You want something that truly documents that visit in a way that you can refer back to it, you know, with the trust that it documented everything you needed, gave you the proper summarization, and you know, ideally they're even making recommendations based on that. So that's one that I think we see today. Through, you know, some of the work that NVIDIA has done and collaborated with the likes of Abridge and others in the industry in terms of beginning to collaborate with them to build these foundation models that you know aren't just large language models, they're actually specifically trained on clinical and medical language.

Chris Hutchins: So you're getting Yeah, on the documentation, that one I love that the you're the getting the summarization. I love that the patient interaction is also part of what gets captured and that is a gap that exists and has been there for the since we started it, long, long time ago. And the True. Value of that I think is amazing.

David Niewolny: Yeah. So that's one that I'd say, you know, we're definitely seeing today. And frankly, just in terms of the team, really proud of the way we've kind of accelerated that industry. And, you know, it's one of those where you work with a leader and really speaks to the way NVIDIA collaborates is, you know, find that organization with a shared vision of the future with a really hard problem that we can jointly bring our efforts. NVIDIA from a technology perspective, you know, in this case, the likes of Abridge. From a perspective of really understanding that clinical unmet need and the problem, and then collaborating to create an end solution, which at the end of the day opened up a market for a number of different organizations that now give you know a lot of these healthcare providers choice in that specific area. So again, I think really fantastic innovation and one that I think we're all going to be reaping benefits from, if not today, in the very near future.

Chris Hutchins: Yeah, it me i at the very least if we're having a little bit more of a personal conversation when we're there so the doc's not actually having to stare at the screen and key things b while while we're there. I mean it it's I mean that just seemed to happen almost it felt like overnight. And w we just we've gotta be able to do much, much better and give time back to that encounter. Obviously let let these guys do what they went to school for. I mean, they didn't they didn't study computer science to be a doctor.

David Niewolny: But No, no, and We they definitely we didn't put think it sent when down they that were going path to be for a while. When they were going to be a doctor, they were gonna spend so much of their time actually documenting what they did versus actually truly, you know, as you talked about, practicing medicine. The one area I did wanna kind of start thinking about, you know, longer term, which, you know, frankly is really, really exciting and really I think paints a picture of this AI native hospital. And I'm gonna kinda anchor it around you know, what is know a couple things. It's the highest revenue generator in a hospital, but it's also where the most amount of complications could occur, and that's the operating room. So there's a lot of work that we have ongoing in the operating room. But if you think about you know an operating room of the future and we kind of started with this idea of I'm going to start with just almost the medical device. You know, robots are going become more and more common. When I talk about robots, I'm going to talk about a surgical robot. You know, you have You know, all of the orthopedic companies, you have all of the soft tissue robotic companies, all now essentially moving from this world where we used to be doing open surgery. We then moved into endoscopic surgery, where we saw a huge reduction in the amount of readmissions, infection, complications, and now taking this down the path of robotic. And now you can start taking that down the path that. We essentially have all seen before, and even NVIDIA being very close to as a leader in some of the autonomous vehicle technology. Similar to autonomous vehicles, you have different levels of autonomy. Level one being, you know, very basic autonomy, level five being full self-driving. And going back to healthcare being so personal, I don't see us going to full self-driving in the area of. Surgical robotics, but wouldn't it be great if your surgeon essentially had the equivalent of lane change warnings, automatic braking, those type of things within the procedure. And it's frankly just going to make those procedures much more safe and effective. And now, if you think of that surgical robot now communicating with the system with cameras outside the room. Microphones in that room capturing what's happening in the procedure, you can now begin orchestrating the entire hospital. If there are tools that are not available, someone can speak and potentially an AMR or another robot, not necessarily humanoid, could actually go get those tools, bring them back into the operating room, and it can do all of that just through the visual context that it's getting. So it's almost like You have this AI layer in an AI native hospital that essentially becomes an orchestration layer where all of these different agents, all of these different pieces of software, begin communicating using reasoning, just as each of those professionals would have used reasoning to begin making that procedure much more safe, effective, and efficient. And I definitely see, you know, I talked about the operating room as an example. And I think we talked earlier just about a radiology example using the X-ray, but that is really the way we see this moving, where you have this entire hospital almost acting as a robot. And that hospital has access to all of the information, all of these different software AI tools that are all communicating in a cohesive manner to help make this hospital operate as efficiently as possible.

Chris Hutchins: Right. It f it kinda gets to the you Mm-hmm. Know some of the stuff you hear when you talk about this to just y an average family member who, you know, doesn't have a lot of exposure to it other than just f f probably just being impatient a few times in their life, hopefully not many. But the whole idea it's you think of something that seemed radical at one point was calling a calling a call center. And no one's answering the phone, it's robotic. There's this resistance that is still pretty solid for the most part out there. When people call, they want to talk to a human. If you want to go to the doctor, you want to see a human being. Talk a little bit about you know how you see that and how you guys think about it and w how you're designing and working with not only the technologies but w with your partners to to try to solve for some of the things that you're going after. I mean, they're not easy to do. But there's that human piece of it that we've gotta try to figure out how to get keep people comfortable and figure out what are the trade offs that they will make or that or that there just won't.

David Niewolny: I'm surprised yeah, I didn't bring this up earlier, but I think the I don't want to call it a call center, but any of these kind of phone conversations, and I think a great example would be and again, indexing on some of these surgical examples. I think for every surgery that happens, you know, you end up having a conversation with your clinician to essentially kind of do that pre-procedure planning. You ask, you know, some questions, you get some interactions, you actually have the ability to Ask questions. And that is one of the most amazing things I'd say about some of the technology advances we've had, specifically in generative and agentic AI, is these tools are now absolutely fantastic. I mean, in some ways it may even be a more personal conversation, depending on your doctor's bedside manner, than even talking to your clinician. And I do think it'll be very short order that. Some of these you know bad memories of some of the automated recordings that essentially had you know essentially pre-planned words that needed to come out based on what somebody said that were a constant frustration. And you know we're not in that day and age anymore. I know a lot of companies are still using them and they do provide, unfortunately, some frustrations. But as all of these companies begin updating to more of these, you know, agentic and generative solutions, you can have a conversation that is as good as the conversation you have with your doctor, and I'd argue in some cases even better. And it also, again, plays back into saving that clinician time. It's something that they have to do. But they don't necessarily need to do because now they're essentially gonna have, you know, their expert going off and having those having those conversations, preparing that patient for surgery, scheduling those appointments. So all of those, I'd say, you know, from the advances we've had really in the last three years, I don't want to consider it a completely solved problem. These models can always continually get better, but The interactions that I've seen with some of the companies that we have worked with specifically in this space have been, you know, frankly, really amazing. And I think, you know, one to highlight is targeting essentially post-procedure care. But we did a close collaboration with a company called Hippocratic AI. And think of this as a digital agent that you can now essentially converse with on a computing device, iPad, phone, and with that you get Essentially someone that looks exactly like an actual person, but it's a simulation. And with that, have a conversation and ask any question you may have about your post-procedure care. And they will also give you the right information in terms of when you should actually, you know, consult your doctor. But going back to another fantastic use case that truly shows this idea of offloading some of the work from the clinician. That today allows them now to focus on more patients as opposed to some of the work that they were doing.

Chris Hutchins: Right. No, you you've talked about some some things in the in access. You you've mentioned explainability a couple of times too, and I want to kind of del double click on that a little bit. But I'd love to hear how how you guys are thinking about that piece of the development work that you're doing is the explainability. Because I mean obviously you're working with some brilliant people and you're gonna s you're gonna do some great stuff, but how do you explain it?

David Niewolny: I guess kind of coming back to that explainability discussion that we had on the open models, it's you know, one, almost in every case, you know, when you're looking at building an AI model in an AI application, you need to be very clear about what is the problem you're trying to solve. With that, making sure you're aggregating the correct data set to solve that problem. And that kind of goes into the explainable AI piece is why did you choose the data that you choose? How Was that data labeled? And to your point, you would love to have data being labeled by the world leader in their specific profession, because that just makes that information that much better. And that's actually what you're then training the model on. So being able to explain, you know, here is the data set, here is, you know, who and how that data was trained. And then, you know, one area that I'll introduce that we're still beginning the process Of educating regular regulators on is the use of synthetic and simulated data. Because again, I don't think many people know this, but you look at, you know, the reason that these large language models that the frontier model or frontier labs are are putting out, they are trained on billions of hours of data. I mean, these are giant, giant data pools, and that's really what makes these models continually get better and better and better. If you look at the pool of data when you start begin looking at healthcare, it's much, much smaller. So it's almost like you can start with the large data sets, but then you do need to augment that because again, going back to the a surgical example I'm very familiar with, how many hours of cases are you going to need to capture to really train on some of these unique corner cases? That you may Right. See once every 10 years, but you still need to know and be aware, going back to your whole point of you know, these doctors, surgeons, clinicians are a wealth of information, and the more experience they have, the better they are. And in many of these cases now, we can, using world foundation models, begin creating essentially synthetic data of these unique cases to add to that training set, which then gives you a much much more robust AI model, which then when put into an application and a workflow, creates much more positive outcomes. So you know, from that explainability piece, I mean just to summarize, you need to be able to explain the data you're capturing, how that data was labeled, how you trained it, how you actually augmented it with additional data, why you augmented it with additional data, which is essentially to cover those, you know, corner cases or rare cases. And then I with that, I think there's always going to be a burden of proof on now show outcomes. How is this actually impacting outcomes? And that's actually another very interesting area where you can begin testing a lot of these models within simulated environments. So it's almost like now you're using the simulated data to augment to build the model, build the application, and then you can run that application in simulation before this ever. Ends up going to clinical trials and touching a patient. So you have a much better idea of the inputs, the outputs, how this is actually going to impact, you know, the patient, the clinician experience before Mm-hmm. Anything even goes to clinical trials. And as I mentioned, I mean that burden of proof, the regulatory agencies are there for a reason and they want to make sure that this technology is safe and effective. But I think with the AI tools that we built, from open models to simulated data sets, all of that really helps accelerate all of these products to market.

Chris Hutchins: Right. You know, you you're kind of touching s a little bit on one of the things I wanted to make sure that we don't forget to talk about is the digital twin concept. That one excites me for a lot of reasons. So you're touching on it just now, around you know, the ability to do simulation. But I think about what could become possible from a research and development standpoint for you know medical research. You know the f ev every industry could could can use technology for in with digital twins. And I've heard it being used in the automotive industry, transportation sectors and anything like that. But we would talk a little bit about how what you see is the the some of the cool opportunities that are starting to emerge on in that space. I that one seems like it's got a wide open area where it can the sky's the limit, I guess, is a way I'd put it.

David Niewolny: I mean, it's pretty amazing, and this does actually span, I think, all of the healthcare industry. I mean, there's a ton of really amazing work being done today in AI and simulation just in the area of drug discovery, molecular simulation. So looking at, you know, accelerating a lot of these pharmaceutical and drug discovery cycles, it has a huge opportunity there. Specifically within you know the idea of AI native hospitals. I think one of the concepts that we're seeing with, you know, any major healthcare organization today that is looking at building a new hospital, how do you actually build that entire hospital in simulation first? Just like we've seen, you know, you mentioned a couple other examples. It's like, you know, we have some very clear examples within NVIDIA where we worked with, you know, BMW to build out one of their entire manufacturing facilities. And this was down to every last component. And they were actually able to run simulations, find out, you know, what do they need for staffing? How are all of these robots going to work together? What's the most efficient way to set that up prior to ever breaking ground in the facility? And we're now seeing that same sort of concept being put in place in healthcare. You design your facility and you can begin, you know, essentially adjusting to ensure the absolutely ideal workflow for that entire facility. And, you know, now to make it really, really interesting, I mean, we all know hospitals get built once and they stick around for quite a while. A lot of these organizations are now beginning to collaborate with us to figure out, you know, in this, you know, new world where a lot of these tasks that aren't necessarily patient-facing that may be automated with things like humanoids and other AMR type robots that are going to be operating within the hospital. How would you actually design a hospital for that? You know, are there separate hallways that essentially, you know, robots are transporting material to and from, you know, either you know, warehouses back into, you know, patient rooms, storage areas, things like that. I mean, a lot of this, and you know, again, those of you or those of us who who have been in a hospital and went into some of the back operations, I can tell you that, you know, the you know facilities, closets and you know, warehouses where they're keeping a lot of these, you know, medical supplies is vastly different than an Amazon warehouse. I mean, everything is much more manual. And that's just a another area where by doing these digital twins and almost looking at that facility as a manufacturing facility, there is a huge opportunity. And that is something that we're seeing across the board.

Chris Hutchins: If y someone listening's in a in an executive role in a hospital system right now. They're kind of trying to figure out which things that they're are really worth their time and what should they be preparing for. Cause it seems like you could jump into something today and tomorrow it's going to get leap fried by something that's going to be much, much better. Somebody's in that kind of a situation, what are a couple of things that you would encourage them to be looking at that from your standpoint, they're not going to change dramatically in terms of what you need to do to be ready for the next thing?

David Niewolny: Well I'd say I mean a couple of things. I'll kind of start with the some of the advice that, you know, we specifically give, you know, hospital administrators as they're beginning kind of this path down AI native hospitals. I think, you know, and it actually goes to the way NVIDIA began using AI itself, is you know, you are gonna have pilots, but don't start with a pilot that is not going to deliver huge impact. Look at what is the single biggest problem impacting your organization or facility today. And that is the challenge that you should actually go after using AI. And I will say hospitals in general are buyers of technology, not builders. And where we end up working with a lot of Sorry. These organizations is essentially partnering them up with the right software organizations. That we're working very closely with pushing advances for this specific use case. So the advice ends up being, it ends up being a collaboration between, I'd say in many cases, you know, a medtech health tech company, NVIDIA, and the healthcare organization, working to solve arguably what they see as the biggest challenge to them, either providing patient care, cost, whatever their big problem may be. And then with that, you have kind of this circular feedback loop where requirements are going to the healthcare or medtech company. All of those requirements translate into technical challenges that the NVIDIA team, with our you know, largely world-leading experts in each of these domain areas, end up backstopping that healthcare and med tech company, solving their technology problems. And we essentially divide those into kind of the clinician or client facing versus underlying technology and we're really trying to remove that undifferentiated heavy lifting for the entire industry. And we really Right. See those type of collaborations are the ones that are exceptionally successful. And I think you'll see more of those kind of coming out from us in you know months and years to come.

Chris Hutchins: A little bit well, I'm I'll be I'll be kind. Y y you're right. I think in healthcare f we have at times tried to bite off more than we actually could do. And the first thing that we have to remember is the real the real burden that we carry, and that is taking care of the people that are coming through the door, keeping the lights on in the place. I think we've overestimated our ability to tackle really complex problems when it comes to technology, and we've underestimated the value or the time commitment it actually takes to become proficient at it. If you want to scale something, it takes a lot more than trying. And most of us in the healthcare sector don't know anything about scale because the majority of the hospitals are not sitting on a massive geographic area. It's a pretty confined space. There's a hand there's just a handful that really deal with that kind of scale. And even in that case, they don't have overlapping markets. So they're still not having to run into some of the barriers that you guys are very, very well equipped to do from a technology standpoint. So couldn't agree with you more. You know, you gotta have the right partners to work on these things together. Th th this has been phenomenal. I feel like I could talk to you for a couple more hours and not exhaust the kind of things I want to know about because you guys are doing some really great stuff. I guess f maybe just one one last question. Thinking about, you know, w where you are in your own career and what you've seen up till now. Fast forward ten years, a any what would shock you at this point from an advancement standpoint? Because I think I don't I don't know what would shock me at this point. I mean things are just crazy the way that they're moving.

David Niewolny: I'd say, I mean, if you're truly looking at what would would shock me would be kind of going outside of the realm that I mentioned is all of this essentially being governed by humans. Where I truly today don't see a world where, you know, the humans are essentially removed from the loop. But you can see research coming out of different organizations that are showing. Fully autonomous robotic surgeries. I think there was just a recent study out of one of the academic medical systems in California that was essentially you know essentially proving that you know you could actually do a full surgical procedure with humanoids. You know, you have Elon saying that you know the future of all surgery is going to be done with Fully autonomous robots. In 10 years, that would be to me a little shocking and surprising. It's like the technology and the foundational components, I think, will be there. We're going to continue pushing down that. But this ends up being just more of a personal opinion from my standpoint and kind of where we have our organization directing a lot of our efforts. It really goes around that idea of augmentation. And we still in you know 100% of the use cases that we're working with see humans essentially in the loop just being made much more efficient as opposed to you know completely transferring you know to me the full ownership of that patient experience and patient outcome to a robot in the future. So that would essentially to me be a surprising component if we if we went down that route.

Chris Hutchins: That th that that's this has been incredible for me. Thank you so much, David, for coming on and talking about this stuff. And I guess very for the last thing, is if there's someone who wants to reach out to you, is the best player to get you on LinkedIn or w what be your preference. I'll certainly put things in the show notes.

David Niewolny: Yeah, please don't hesitate to connect on LinkedIn. Tend to share as much as I can about, you know, new and exciting innovations in the space. And if there is a way that you see a potential collaboration, please don't hesitate to drop.

Chris Hutchins: Awesome. Well, th thanks again, David. I can't thank you enough for coming on the show. I mean, this has been amazing for me. It's like a kid going to the candy store and I'm not having to select just one thing. I feel like I came out with a whole whole bag. So th thank you so much for that. And for listeners, this has been a fun episode for me. And if you've enjoyed it, please make sure that you ch you like it, put in the comments, tell your friends to subscribe. I can't wait to see you next time. For now that's gonna be it for this episode of The Signal Room. Take care.