EP 594: Data Dreams & Digital Delusions: The role of AI in health tech

Topics Covered in This Episode:

1. Trillion-Dollar Data Center Investments for AI
2. Generative AI Transforming Health Tech Use Cases
3. AI and Physician Burnout Solutions
4. Organizational Challenges: AI Literacy and Education
5. Data Quality, Cleanliness, and Health Tech Outcomes
6. Transparency and Accountability in Health Data Pipelines
7. Importance of RAG (Retrieval Augmented Generation)
8. AI Hallucinations and Patient Safety Risks
9. Open Source AI Models and Health Data Privacy
10. Future Impacts of Large Language Model Investments

Keywords:

Generative AI in health tech, artificial intelligence in healthcare, large language models, data center investment, AI hallucinations, data quality, retrieval augmented generation (RAG), physician burnout, AI-powered clinical notes, medical imaging AI, HR workforce optimization, health tech digital transformation, healthcare data privacy, healthcare data security, AI model training, AI education, data literacy, operational efficiency in healthcare, sustainable data centers, environmental impact of data centers, transparency in AI, data trustworthiness, real-time data extraction, cross-checking AI outputs, patient outcomes, healthcare regulations, high-stakes AI industries, open source AI models, AI supply chain management, healthcare claims denial, prior authorization automation, grounding AI models, human-in-the-loop in AI, data accuracy, precision in healthcare AI, medical decision making AI, data-driven healthcare, responsible AI use, digital health, retail health, AI for finance and accounting in health, healthcare payers and providers, AI in marketing for healthcare, resilience in healthcare systems, AI data pipeline transparency.

Podcast Transcript

Jordan Wilson [00:00:46]:
The largest companies in the world are spending trillions of dollars building out data centers for artificial intelligence. You have single companies investing hundreds of billions of dollars, hoping that more data and hope maybe even better data will make AI exponentially better. But is that a reality or is that a dilution, and especially in high stakes industries, like health. So today, it's gonna be a fun conversation. I can't wait to have this. We're gonna be talking about data dreams in digital delusions, the role of AI in health tech. This is gonna be a good one. Trust me.

Jordan Wilson [00:01:32]:
Strap in. It's gonna be fun. Alright. This is your first time. Welcome. This is everyday AI. My name is Jordan Wilson, and welcome. This thing, it's your daily live stream podcast and free daily newsletter helping everyday business leaders like you and me not just keep up with everything happening in the world of AI, but how we can leverage all this information to actually grow our companies and our careers.

Jordan Wilson [00:01:53]:
If that's you, you're like, hey. That's what I'm trying to do. Cool. Starts here with the unedited, unscripted livestream podcast. But if you wanna take it to the next level, make sure to go to youreverydayai.com. Sign up for the free daily newsletter. We're gonna be recapping the highlights from today's conversation. So if you're out jogging on the treadmill and you run out of breath and you're like, wait.

Jordan Wilson [00:02:13]:
What did they just say? It's all gonna be in the newsletter. So make sure you go, check that out as well as today's news is gonna be in there as well. Alright. Enough chit chat. Let's talk data, not just data because we know we all need it for large language models, for generative AI, for our companies to leverage that technology. But what about just these wild investments in these data centers? And is it ultimately gonna get rid of hallucinations? And what does that mean for these high stakes industries? Alright. You don't have to listen to me, chitchat by myself. I'm excited for today's guest, so please help me welcome to the show.

Jordan Wilson [00:02:50]:
There we have her, Smriti Kurabanandan. Smriti, thank you for joining the Everyday AI Show.

Smriti Kirubanandan [00:02:56]:
Thank you for having me. Happy to speak.

Jordan Wilson [00:02:58]:
Tell us a little bit about your background. So you are a health tech executive, but tell us a little bit, kind of on your expertise.

Smriti Kirubanandan [00:03:05]:
So Smriti, Simi, two names, based in Los Angeles. Been here for a couple of decades, but background by education is primarily in robotics and public health. Also recently completing my master's in data science because I figured that's where the world is going. We all are geeking out on data, might as well jump on the boat, but really spent my couple of my decades working in start ups, product companies, designing UI, going to market, working on the growth side of these products, and then eventually got into the consulting world, with obviously the, you know, top fortune companies, working on digital transformations, especially for health care payers and providers, and also, you know, the blend of health where it meets every industry. So retail health, health tech, digital health, but it's quite fascinating to me, especially now the world we are living in with AI and GenAI, the implications of that, the usage of that is quite quite large and quite expansive. But, obviously, today, we'll talk about, you know, the two sides of the coin, which I'm excited to talk about is the investments in data center, but also as an individual even divorcing me from what I do, are we truly, you know, getting the right information at the right time? Yeah.

Jordan Wilson [00:04:12]:
So, obviously, health tech, is nothing new. Right? And artificial intelligence in health tech is also not new. But, you know, bring everyone maybe up to speed before we dive into the details. How has generative AI, changed the health tech scene?

Smriti Kirubanandan [00:04:32]:
See, I think in in many ways. Right? So one is using GenAI in certain use cases, maybe, for example, physicians using GenAI could be prescribing for clinical notes. Obviously, a very positive change given our physician burnout. Right? Couple of years back, we had 800,000 physicians. Today, I'm gonna say we have 600,000 physicians. So we're dropping quite a bit. Right? But systems where they're using GenAI to help them write clinical notes, verify medical imaging, is obviously helping them accelerate the job so they can see more patients with greater quality than go through a major burnout. Right? So that's just like one use case.

Smriti Kirubanandan [00:05:10]:
But we are seeing implications of AI and GenAI across the enterprise. It could be for HR workforce optimization. It could be as simple as, you know, note taking, which I shared before. It could be, you know, in marketing where you're creating images and documentation. It could be legal. So the implications are enterprise wide, and it's quite helpful if if done the right way. So that's kind of where we're seeing a lot of momentum and, you know, kind of changes being implemented.

Jordan Wilson [00:05:39]:
Alright. Let's maybe skip to the end a little bit here. When we talk about these just insanely large investments in, you know, infrastructure around AI, You know, you have individual companies, multiple of them investing hundreds, hundreds of billions of dollars, into these projects. Why? And is this ultimately in the health tech space going to mean safer, better outcomes, for patients, if it all works out how big tech companies want it to work out?

Smriti Kirubanandan [00:06:16]:
You know, great great question. I feel like we'll have, we'll need hours or years to actually get to the bottom of this. But, see, I think some of this is a business imperative. Right? This is a business decision which each company is making. In order to be ahead in terms of the GenAI, just as a competition in the market, you, by just general, need access to data. Right? So data centers is a natural investment all these companies are making. But given the current administration, the regulations, obviously, trying to move the workforce to The US, have the job market to be really strong here. Obviously, the investments in The US have been large, but that's also globally.

Smriti Kirubanandan [00:06:50]:
Like, I found this news that there's a big investment just made in Ghana for, you know, Africans and that population to really come up to speed on AI and data centers. Right? So the investments are being made overall for the right intent and the right reasons. But that comes to the other side of the equation is that how much of this data is truly clean, how much of that is truly modeled. And then on the other side, users like you and me and just everyone else, even data scientists, how many of them are actually prompting right? How many of these answers are right? So there's a little bit of, you know, disconnect, I would say, between what's being trained, what's being prompted, what's being coming out. And, obviously, in health care, the the weightage of these outcomes are really, really high stakes.

Jordan Wilson [00:07:35]:
Yeah. You bring up some great points. And, you know, I won't, name drop or shame drop, I guess, anyone by name, but I've been taken aback by even larger companies making, just giant investments, $7.08, 9 figure investments that still don't push literacy. They don't push AI education. Right? But they're investing sometimes hundreds of millions or billions of dollars. How have you seen that play out in the health tech, side? Because I I would assume it's it's no different. Right? It's one of the, I think, one of the the more, you know, appealing maybe, sectors that has yet to fully get cracked by AI. But how do you see it shaking out on that side?

Smriti Kirubanandan [00:08:23]:
So let me answer this two ways. Right? One is the personal side. When someone is investing in data center, I wrote a piece just maybe last year on just the, the sustainable impact of data centers. Right? There is a high cost on environmental pollution, on just mental health, physical health, and all those things that happens, the creation of data centers. So one call to action is obviously to make a responsible ecosystem investment, not just in terms of data and what the business outcomes are, but also what does that do for the particular region society in terms of health. Right? So that's key. That's a public health call out. The second one is how is this truly paying playing out in health care where people are using a shared audio that physicians sorry.

Smriti Kirubanandan [00:09:05]:
Physicians are using it. Clinicians are using it. Healthcare management is using it. We're using it across a supply chain, but we wanna see what is happening in inventory for transparency, resiliency. We're using it in f and a, which is finance and accounting to truly create, you know, payment transparency across different models, give people access to, you know, the payers, insurances. So there's quite a bit of use cases on where GenAI is being used. But, obviously, the three stacks where I see where leadership is using it is one, cost takeout, operational efficiency, obviously, reducing burnout for physician physicians and clinicians. But overall, I think some of this is being done with the right intent of augmenting and doing the right things.

Smriti Kirubanandan [00:09:48]:
Right? But then teeing back to our conversation about dilutions, a little bit of hallucinations, how much of these models and answer that we're retrieving is accurate. Right? So some of the things that I do, advise clients and, you know, work is how do they implement direct, which is retrieval augmented generation, to make sure that the data that they're extracting is from the right datasets and is being cross checked live and is not just, you know, from the abundance of stored data that is not real time. So it's very important that the extraction that's being done is real time, real time answers, real time analytics, real time web sources, and even that, it's, you know, before we jump on the call, you and I had a discussion and example about organic fruits and vegetables. When we go to the store, it's tagged organic. Right? But do we know is the soil organic? Is the fertilizer organic? Was it grown organic? We have no clue. Right? It's a similar, similar example, but the the weightage and the high stakes of health care is really high because, you know, it's you're dealing with lives. So it's important for us to really go back to what is the source, what is the process, you know, what is the end outcome. So, you know, my ask is, obviously, investments are important.

Smriti Kirubanandan [00:11:03]:
Right? But when they're making these investments, make the investment to your point on data education. What does the process look like? How much are you investing in rack, in the guardrails of the framework and security? And then the, obviously, the end outcomes, which is, you know, you and I or even a patient viewing the data, what are not just the, information outcomes, but the mental and physical outcomes of reading that data? You know, I think I think it's a it's a heavy pull.

Jordan Wilson [00:11:31]:
Yeah. So using your organic fruit analogy there, how should maybe a health care executive when they're looking at, you know, maybe a maybe they've had a successful, pilot of GenAI in a smaller, scenario and they're looking at a wider rollout. How should they be examining that fruit, that large language model to make sure it's organic? How should they be looking at that large language model to make sure the data is maybe more real than delusion?

Smriti Kirubanandan [00:12:04]:
So, let me show an example. I was I was with a friend having glass of wine many months back, and, you know, he shared that there you know, honesty is great, Simi. Right? Honesty is saying, Simi, I ask you, where did you go for dinner? And you say I went to x for dinner, and the it it was with friends. Right? That's honesty. But he said what's really true and authentic is transparency. If you said I went to Javier's and I met Jordan, Eve, and Alice for dinner, that's transparency. Right? So at that point of transparency, you build trust, you build credibility, and you own accountability. Right? So so that's where the the moving of the needle.

Smriti Kirubanandan [00:12:41]:
Right? I think organizations by large are honest. But if they're transparent, then that's when they're able to show consumers and clients and everybody else, what are the investments of data centers? What does the investment look like? Like, break it down to us. Right? Have some kind of, transparency on where what what that's happening. What data sets are they truly using? Like, create, like, a transparent like, a pipeline on the training, what's being trained, how often it's being trained, and then the entire process pipeline. So if I go in and I'm like, okay. These are the datasets that's being pulled from. That's how efficient those datasets are. That's how accurate those datasets are.

Smriti Kirubanandan [00:13:18]:
Now you're building trust. And then the entire pipeline of how is it reasoning. Right? Because as you know, there is a chain of thought when a gen GenAI is reasoning. How actively and accurately and how fast is the reasoning happening? I think seeing that, like, creating this very glass model of GenAI, I think would be incredible. Right? And then the prism light, if I look at it, it's like the different outcomes it produces. Then the accountability goes to a person on how they wanna use the data, how do they wanna trust it, and how quickly do do they wanna keep prompting to train it further. Right? So I think that that would be my ask, creating this last model of GenAI.

Jordan Wilson [00:13:55]:
So you had just mentioned, a little bit about RAG, Retrieval Augmented Generation. Right? And I don't know. At least for me, and and maybe it's just because we're getting, over identified, but it seems like people are maybe talking about or focusing less on rag, right, which is probably more bad than good, but that's beside the point. It it it seems like people are just hoping that, you know, the scaling laws of large language models and larger context windows, right, they're gonna get, like, get rid of the need for for rag. Right. Can you talk a little bit about, like, is that a good or bad idea? And then specifically talk about, maybe the importance of RAG on the help tech side.

Smriti Kirubanandan [00:14:38]:
So, I mean, just for people listening. Right? I think RAG is when you you prompt something, and then the system is actually actively checking from live resources as you type in, and it's not pulling data from something stored, you know, ten years back or twenty years back. So it's more real time. It's accurate data. It's live, which is more trusted. So, you know, I I personally, Jordan, believe that everything in life needs a checkpoint. Right? Every process needs a checkpoint, and it's healthy for especially a data driven system to have a checkpoint, which is what RAG is in a very simple way. Right? It's a checkpoint to make sure it's accurate, it's live, and everything else.

Smriti Kirubanandan [00:15:15]:
So regardless of how models are being trained, unless that's a modern version of, you know, RAG, I think that'd be great. But I think it's very important for us to get grounded. That's what RAG does. Right? The grounding prevents hallucinations. Without the grounding, we're all just taking information and becoming these bodies of misinformation because perception is a reality. What we read is true. Like, how much time do you and I have to be cross checking models and figuring out if the answer's right. Right? So I think the the responsibility and the imperative goes on the leaders to create that rag as a grounding, as an important pillar to be able to do what they're doing, especially in health care.

Smriti Kirubanandan [00:15:54]:
Because implications for, say, even a physician health care management to make a decision, let me say an example of prior authorization. Right? Because claims denial is one of the key issues where patients don't get the right care at the right time. But if the rag is there, it grounds them a reality on the claim, approves it, disapproves it, whatever. It's a very trusted source and forms as a sense of truth. Without that, you know, the claims can be denied. It could be on a base of wrong datasets, wrong prompting. So it's almost like how do you it's kind of a mediator that holds both parties accountable and holds like a source of truth. So to me, I think I I think it is important.

Jordan Wilson [00:16:35]:
So, you know, as we talk about data investment, and I think that everyone wants better data. Right? They might not ultimately know what that means or how to get it, but, you know what? And I'm not asking you to look in your your crystal ball here. But, presumably, right, you have, as an example, the, Stargate project, a $500,000,000,000 investment in data centers for AI. Right? And we already look at the level, in empower and complexity of these models, like, prior to these, you know, $100,000,000,000 investments. What might this mean for the future of AI, its applications, and the data quality. I know, you you know, I'm not gonna

Smriti Kirubanandan [00:17:18]:
Yeah.

Jordan Wilson [00:17:18]:
You know, put this put this on a wall somewhere and and come back in five years and see if you were wrong or right. But, you know, I think we always have to be thinking ahead. So how do we think ahead about that, scenario?

Google Gemini [00:17:34]:
This podcast is supported by Google. Hey, everyone. David here, one of the product leads for Google Gemini. If you dream it and describe it, v o3 in Gemini can help you bring it to life as a video, now with incredible sound effects, background noise, and even dialogue. Try it with the Google AI Pro plan or get the highest access with the Ultra plan. Sign up at gemini.google to get started and show us what you create.

Smriti Kirubanandan [00:18:06]:
So I I love the question. One, I think we should connect in fibers and see where the world is gone. But, I think, see, I believe that data is very important just overall in life to make the right decisions. Right? It empowers people to make the right decision. So the availability of this amount of data, I think the other problem we have is, like, it's too much data now. We we don't we don't know. Right? It's kind of like it's like kid in a candy store. Now the kid has unlimited access to data, but too much sugar leads to diabetes and other issues.

Smriti Kirubanandan [00:18:34]:
Right? So that that's where I struggle with, quite honestly, as to how much data is good data, and then how do you truly make sure this is responsible in execution of data. So I think to that point, I think we have to make responsible investments and also figure out, is there a way where we can control the the release of data and datasets, And then what does that look like? As long as we're still in the process of refining the datasets, making sure the information is accurate, because we still don't know. I mean I mean, you know, you and I discussed, are there bad actors creating fake datasets to confuse the system? Possibly. Right? So until we get to the point of, you know, a trusted way of really managing these datasets and extracting them, I think it's a very longitudinal process. But the good part of this is I'm hoping that especially in health care, all this data helps, you know, research, it helps make the right decisions, improves physician burnout, obviously improves the job market, really helps people out. I think that is obviously tons and tons of hope and the goodness it also brings. Right? But with great power comes great responsibility. I just don't want people to forget that.

Smriti Kirubanandan [00:19:46]:
Yeah.

Jordan Wilson [00:19:47]:
Yeah. And, you know, at least, you know, I'm always looking when I'm thinking about the future of AI. Right? What's after the next model, so to speak, and what might that mean for certain sectors? And, you know, my my very little understanding of the, you know, health tech, scene. Right? I was, you know, I've I've got to talk to super smart people before, you know, the head of the AMA and and some other great people. But it seemed like earlier on, you know, in the first, you know, year or three of the GenAI phase, right, so many bigger health organizations just didn't get on board with AI, you know, just because of data security, data privacy, you know, PHI, all of these things. But now as we look forward, okay, we are at a point today where you have open source models from OpenAI, right, with their new GPT OSFs. You have new variants of, you know, JEMMA's, three. So you have models that are, open source.

Jordan Wilson [00:20:46]:
You can download them. You can run them on prem, no Internet, anything else, right, that are more powerful than what we had eighteen months ago. How might this change? Might we see a big surgeons of, you know, kind of open source usage, you you know, in in the health tech side?

Smriti Kirubanandan [00:21:04]:
I mean, another great question. I think, see, the the one thing you mentioned is data privacy is very, very key, and the responsible use of data is also important. So, I mean, that could be a, you know, extreme search where everyone's using it and implementing it and, you know, using a different use cases. So I I don't know if I know the right answer to where, you know, the trajectory of this goes in health care. But because health care, unlike other industries, touches people's lives. Right? This is not retail. I mean, obviously, they they touch people's life, but not, you know, the actual lives. I think going slow to go fast is it is important, and, hence, I do see why health care is behind in terms of the the implementations and using it because it does it does impact life.

Smriti Kirubanandan [00:21:48]:
Right? One wrong outcome or a decision can change or, you know, save lives. So I think, that's where I'm always a little caution about promoting or sharing. Be like, this is great. Let's all use it in health care because we don't know what we don't know. Right?

Jordan Wilson [00:22:01]:
Yeah. That's that's a great point. And even, kind of related to that, right, there's obviously with today's models, if, you know, I'll rewind and stop asking you questions five years down the road. But when we look at today's models, hallucinations are still very much a part of everyone's day to day. How do hallucinations, especially, hallucinations that maybe get unchecked or kind of go to production on the health tech side, How does that impact, you you know, research, you know, patients' quality of life, and just the overall direction of health companies, and how can they better deal with hallucinations when they can be increasingly difficult to spot?

Smriti Kirubanandan [00:22:45]:
So, I mean, this goes back to having the transparency in the pipeline. Right? Where is the data being extracted from? That that is the first step. Right? The second one is understanding, is the data that's being extracted from the prompt, is it live? Is it accurate? Is it from a web source? Like, are those sources credible? That's why RAC comes to place. The grounding of that is key. But the implications and the high stakes are high. Right? So say someone puts in an m an MRI and x-ray saying, does this does this person have, say, cancer? And say the AI system is hallucinating and says, yes. The potential of this person having cancer is eighty percent, and it's wrong. Think about think about the mental, the physical, the financial burden on the patient, on the provider, and let's talk about the other side of just the payers and providers fighting about utilization and people use losing Medicaid and Medicare.

Smriti Kirubanandan [00:23:36]:
So, you know, bringing this kind of a hallucinated decision and outcome in a very complex, turbulent time can have severe implications on just the patient. Right? That's why I keep saying, like, this is this is very serious when it comes to patient outcomes because we do have to take it seriously. So in order to avoid these hallucinations, that's why I believe we need to go slow to go fast. Right? We need to have the transparency to check the dataset, to have the rag, to double check even models. Right? Like, I don't I don't want, I I you know, it's okay to go an extra mile to make sure it's right. Check with three different models to see if you're getting the same answer. Right? And then and then make an informed decision. But you should also keep the human in the loop.

Smriti Kirubanandan [00:24:17]:
Right? We are all there because we are educated. We're there for a reason, especially physicians. So I think having that ecosystem of cross checking, having the rack, having human in the loop, I think it's it's still still will always be important.

Jordan Wilson [00:24:31]:
Alright. So we've covered a lot in today's episode. We've talked a little bit about the current and future role of AI in health tech. We've talked about, these large data investments and how they ultimately ultimately may impact, data quality in, future large language models. But, you know, as we wrap up, what is your one most important, takeaway or piece of advice, for business leaders in the health tech space still grappling maybe between the data dreams and the digital delusions?

Smriti Kirubanandan [00:25:04]:
I think one key, maybe couple, will be obviously transparency over honesty. Second one, you know, is truly creating, being responsible about the data that's being released, but also constantly training and making sure it's accurate. I think it's just, you know, accuracy and precision is gonna take over, you know, most of these concerns, especially in health care. So I think those those two will be my key key important pressing points for data.

Jordan Wilson [00:25:31]:
Alright. We covered a ton, and I think that, this was a conversation that was very much worth having, because, I think the, innovations, in the health tech scene is going to ex be exploding, in 2025 and beyond. So this was a great, I think, conversation to have. So, Smirdi, thank you so much for your time and for coming on the Everyday AI Show. We really appreciate it.

Smriti Kirubanandan [00:25:56]:
Thank you for having me, John. Appreciate it.

Jordan Wilson [00:25:58]:
Alright. There was a lot we covered. If you missed anything, it's gonna be in our newsletter. So if you haven't already, please make sure you go to youreverydayai.com. If this was helpful, tell someone about it. Right? We bring on some of the world's leading experts so you can get your questions answered. Thanks for tuning in. Hope to see you tomorrow and everyday for more everyday AI.

Jordan Wilson [00:26:19]:
Thanks, y'all.