This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Today on This Week Health.

human beings can make mistakes and be forgiven.

Technology gets to screw up once, maybe twice. And then, There's a halt to it.

Welcome to Newsday A this week Health Newsroom Show. My name is Bill Russell. I'm a former C I O for a 16 hospital system and creator of this week health, A set of channels dedicated to keeping health IT staff current and engaged. For five years we've been making podcasts that amplify great thinking to propel healthcare forward.

Special thanks to our Newsday show partners and we have a lot of 'em this year, which I am really excited about. Cedar Sinai Accelerator. Clearsense, CrowdStrike,. Digital scientists, Optimum Healthcare IT, Pure Storage, SureTest, Tausight,, Lumeon and VMware. We appreciate them investing in our mission to develop the next generation of health leaders.

Now onto the show.

(Main)   here we are. It's Newsday. And am joined by Charles Boise, one of my favorite technologists. And we're going to cover some news, AWS. decided to step up and compete in the Gen AI space. I'm actually at a conference where we are talking about generative AI and actually all sorts of AI technology.

So that's kind of fun. And it's always fun to see what and where Charles is at. Charles, welcome back to the show. Hey, good to be with you, Bill. We man there's a ton of stories out there. I think I'm going to start with the AWS. Anthropic story. And so Amazon invests 4 billion into AI firm Anthropic competing with Google, Microsoft, and the AI race, attains minority ownership, promotes generative AI development.

By the way, 4 billion for a minority ownership is pretty amazing. Promotes gen AI development, propels the Anthropic competition with chat to BT's developer, OpenAI. Anthropic selects AWS as primary cloud provider, plans customer first access to select model customization, fine tuning features, investment pushes Amazon alongside Microsoft, Google, AI investments.

AWS customers can build on Anthropic models with Amazon Bedrock agreements. makes Amazon's Anthropx custom chip source for training. So Amazon has their own chips, deploying unique AI systems advancing against NVIDIA. And this is from National CIO Review and was submitted to our new... Actually, if you go to our website, there's a new news tab to take a look at.

It's early access, but it's essentially a... Industry curated series of news stories from across all the various so the news stories we talk about and you're wondering, Oh, how do I find that? Just go to thisweekhealth. com slash news and you're going to see all those stories categorized. With that being said.

Charles, this was inevitable, wasn't it? AWS had to step up. And

Bill,

we hope that there's others. This is not a competitive thing, although from a business perspective, it's considered competitive. But for myself as a technologist, the more that are in doing this, the better the technology is going to be, and the faster it's going to mature and evolve.

Can you imagine if... Two or three others put 4 billion a next stage EMR, what that product would, might be. So I see this thing as a as fantastic in that again, it'll advance the technologies and

make better products. we talk about healthcare IT, but we have to remember that this investment is not only for healthcare IT, but, it's across the board.

So let's look at all the different strategies. Microsoft has partnered heavily with OpenAI, and everyone's familiar with ChatGPT. They continue to make progress. They're creating corporate accounts where you can sign up in a corporate way. You can do Azure. You can do OpenAI through Azure.

You can train your own models that way. You can put a BAA around it. So that's their whole model. You have AWS with Anthropic and you can access their tools and whatnot through Bedrock and that's the direction they're going. And that'll work itself out over the next couple of months. You have Google who arguably It's the foundation for most of this technology. Like they wrote it, they put papers out, the transformer, all that stuff. But in traditional Google fashion, it couldn't figure out how to, I don't know, commercialize it, put it out there. They stepped in very slowly. Well, they're not stepping in slowly anymore.

They're we're starting to see. A bunch of tools come out from that, MedPalm and other things, especially in the healthcare space. And then the one I find funny is so they all have sort of commercial models and they're going to make billions from it and that kind of stuff. And I think Zuckerberg figured out he's not going to be able to compete.

So he's playing the spoiler and he just wants to burn it all down. Lambda 2 out there is open source and he's like look everything we come up with we're putting it out there for everybody to make sure that my competitors don't make any money from this. So those seem to be the four main plays and then as you say there's other players out there right now that are saying look this is the start of the gold rush.

So you know get in there develop something get it out there and there's an opportunity for you to establish a new company. Yeah, and what, one of the things

that, we're working on is, it's really important, especially on the healthcare side, is the the ability to orchestrate and project those prompts against the proper language model and or across multiple language models and ensuring that prompt is the intent of that prompt is accurate.

So, There's a lot of stuff going on, kind of in the weeds, if you will, technically, and that's why I'm really happy to see as many people as possible, be involved in this because we'll end up with better technologies going forward. And, Zuckerberg takes all of that and puts it out open source.

Yeah it's interesting. I've seen some so, there's really three ways people are going to do AI within healthcare. And I think a majority of them are going to essentially, they're going to sit around and talk about governance. They're going to talk about a bunch of stuff and then they're just going to sort of happen into AI.

What's going to happen is companies like Clearsense and others are going to start to integrate it. And actually you already have AI built into your platform in a lot of ways. But the whole chat GPT and whatnot, It's going to start showing up in the EHR because we've seen announcements from from Epic.

We've seen announcements from Cerner. I'm sorry, from Oracle. Seen announcements from Meditech. So, all the major EHR providers are going to start bringing it in. You're going to see other players, obviously you're doing it. We're seeing some security players bring AI in. We're seeing AI platforms like Notable, which is the conference I'm at this week is noteworthy.

And we're looking at AI. And you're going to see these things sort of just eke into healthcare just because it's just part of the products that they already own. And all of a sudden they're going to have it. I think that's gonna be the majority of it. And then you're going to see other people who stand up an Azure instance and they're going to start training it themselves on various aspects of their health system.

And then you're going to have some people who are going to be really out there. And I'm not sure that there's going to be two aspects of the people who really go out on the frontier. And one is going to be those people who are trying to protect their data. The people in healthcare who are like, oh, this data is worth something.

It's so valuable. And we don't want to give it to this AI company to train their model and not get compensated. And those people are going to go in that direction. That's going to fail. Sorry, that's just my opinion, but I've seen some interesting things out there. I've seen. Some nerd physicians or nerd clinicians, which is the category I put you in, that have started to do some tests and they anonymize some of the, and they're seeing patients, they anonymize some of that stuff, and they're creating, they're using FHIR and whatnot to pull this stuff out, generate prompts, send it up to an LLM, a large language model of some kind, and then see how it functions in the ER.

And, would it come back with the same diagnosis I did? And we're seeing some of these experiments start to happen. Kevin Malloy, MD, has a couple of posts, if you follow him on LinkedIn, and it's really fascinating. He's like, he's sitting there going, look yeah, there's some hallucinations out there.

And, every now and then it comes back with a diagnosis that no doctor in their right mind would, look at abdominal pain and come back with this. But in a lot of cases for a model that wasn't trained on healthcare specifically, it's pretty doggone accurate.

Yeah. you're absolutely right. And Bill, the work that, you and I did together. It's 10 years now where we use this technology and it was adjunctive. And I think that's the important concept here. It's adjunctive to the clinician. It's there to present, to to, cause a cognitive trigger, if you will.

So, in the last half an hour, This is what's occurred from a vital science perspective. Here's the labs, 8, 000 patients like this. This is kind of the trajectory. It's a good time to make a decision. I think if we keep that mindset that this is adjunctive and not replacement I think we're going to be in really good shape.

And the reality of the generative AI It's got everybody's attention. Generative AI is not the end all be all to AI machine learning in healthcare. There's other aspects of it. We, most folks are starting to learn, about predictive and other applications of this.

But again, adjunctive to the human, in this case, clinician. So that they can be made aware and make some decisions and whatnot. So I think that's the trajectory that we're on. And, I'm happy to see it. Those that prescribe, do not want to be prescribed too. So

let's keep that in mind.

When you say the word adjunctive. We've been talking on the show about a co pilot design construct. That's what you're talking about. Yeah, angel on your shoulder,

you know that, practitioner that, we've all admired from our careers is right behind us, guiding us

along the way.

I did a Today Show on this, last week. I think by the time this airs, it'll be last week. And it's probably the closest I've ever come to actually doing a commercial for a company, because I'm at the Notable Conference. And so they've ingested the, obviously, we all ingest the discrete data elements.

Simple, no problem. We, we ingest that. But they've layered the various AI technologies. You go from OCR to NLP to the large language models to, machine learning and whatnot. And they've taken all the unstructured data. And they've made sense of the unstructured data.

And then they just came out with a thing called assistant, notable assistant. And I, I want to get your thoughts on this because. I think this was the promise of doing meaningful use. This was the promise of doing the EHR and digitizing everything. And it was, Hey, we're going to go through this pain because there's a lot of pain.

And there was a lot of pain getting to this point, but at some point we're going to be able to query the medical record. Not only the structured data, but the unstructured data, and it's going to respond. And we're going to be able, as a patient, we're going to be able to say, Hey, I have these symptoms. I need to see a doctor.

And it's going to present a list of doctors based on my insurance and with appointments. And the process is going to be frictionless. And I'm just going to click on that appointment and boom, it comes back. Well, Assistant is that Google, that the promise of the Google front end. You put it on your website and it's just one box.

And you can start asking it questions in like a hundred different languages. And it responds in the language you ask it. And you can query things like, hey, I need this. And there's an unauthenticated set of things it can answer. Like, where's the nearest this? And what doctors take my insurance?

And that kind of stuff. And then there's an authenticated experience where you can essentially say, hey, what medications am I on? when can I get a refill? And because it's ingested all that, and it's made sense of it, it actually comes back with meaningful responses. This was the promise of digitizing the medical record.

And I think we're on the cusp of really seeing some magic happen with these AI technologies. as I talk about that construct, I'm curious what your thoughts are. Sure,

from the patient perspective, there's some key elements that also need to go into that. That is all the quality measures, that information.

I don't remember, I don't know if you remember, Bill, but back in the day, we built a website called Will they kill me? And basically you listed the symptoms, you listed your location, and you got a percentage of likelihood of having a good outcome depending on where you went. And those, that was all developed off of publicly available quality measures and whatnot.

So if you have that information within that application, then you can, as a consumer, We're patients, yes, but we're consumers as well. You can make the proper decisions.

Charles, is that still out there? Cause that would, that's fascinating.

No. I'm surprised I even, it came out of my mouth.

The answer is no, I'll never give that URL out ever, but but it's something, we played around with, quite some time ago.

  📍 📍  We'll get back to our show in just a minute. We have an excellent webinar coming up for you in November. We had an excellent conversation about AI in September with three academic medical centers around the topic of artificial intelligence.

It really was exceptional, and we released it on our podcast channel so that we could share it with a wider audience. I wanted to explore that topic a little bit more, and I asked a couple of additional health systems to join us to explore the use of generative AI and other forms of artificial intelligence to see if we can identify some pragmatic approaches to how health systems are looking at taking advantage of this technology.

The webinar is on November 2nd, 1pm Eastern Time, 10am Pacific Time. You can reserve your spot on ThisWeekHealth. com and one of the things we love is that you can submit your questions in advance and we can make sure that we, answer those questions and keep the webinar relevant to the things that you're looking to talk about.

So, please join us November 2nd, 1 p. m. Eastern Time, 10 a. m. Pacific Time. Now, back to our show. 📍  

Because it's true. We trust a brand, so we'll trust Mayo or Cleveland Clinic or Cedars Sinai. We'll trust the brand, but the reality is their doctors rank in terms of their quality and their outcomes and that kind of stuff.

And and in some cases. When they sign on with a medical group, they don't get to say, Hey, we'll take two thirds of your medical group. You get all of the medical group. That would, that's why that's a really interesting concept as a consumer, I want to know if I'm seeing a doctor, that's a good doctor there.

Good bill.

I kind of call it the chaos of healthcare and depending on what organization you go to. What time of day you go to, because that's the staff that's on, on, and where the room that you're going, all those different variables really play an effect on what your outcome is going to look like.

So, if I have a, fractured arm, what facility should I go? It's Wednesday. It's eight o'clock in the morning. Where should I go? And if you think about internally you as a healthcare organization know all your outcomes. You know what's going on at any particular time and whatnot.

So, yeah, at some point in time that will, again, allow us as patient consumers to make appropriate

choices. All right, Charles, let me take you in another article. Sam Altman says he intends to replace normal people with AI. And this is from Futurism.

Describe normal.

Sam Altman views artificial general intelligence.

It's important to note we're not there yet, AGI. Replacing median human workers. Experts caution against the dehumanizing connotation and oversimplification of human intelligence. Altman's definition of equitably benefiting from AI involves reducing majority of humans. To replaceable median figures. The reason I bring that up is it's gonna be harder and harder to deny that we're going to be able to take AI into areas like RevCycle.

And you and I have been in these rooms where there's cubicles as far as the eye can see doing coding and submissions and those kinds of things. going to get harder and harder to make the case that this is not going to displace humans. It's going to displace at least hours of work.

Yeah, and also the process around ensuring that the bill that gets dropped is 100 percent accurate or as close as it can be. Again, This is adjunctive initially. Will it replace those folks over time? We'll see, but there's still some, there's still, there's something to be said for, local knowledge and, user

experience.

The reason there's mountains of cubicles is because the process is really hard. It's like they have to go in and research the insurance, and they have to make sure the card is right. When a submission goes in wrong and comes back, they have to research the fields, go into the EMR. Look through some of the unstructured data.

If AI has collected all that stuff, essentially what you're going to get is, first of all, you're going to get a lot of submissions that aren't going to have those errors to start with. you're going to cut down labor right there. And then you're going to have things like, you take a picture of the insurance card on either side.

And essentially you have that, which we do today, you capture that information. But here's where it goes to the next step. You use RPA, and RPA actually logs into the carrier site and identifies what's needed for a prior authorization. What's needed for, this and that. So you have all this information.

So the system's intelligence is greater. And so even on denied claims, it can auto generate a response. now again, I believe in the co pilot design construct, but even if it's a co pilot design construct, that mountain of cubicles will become a manageable group of cubicles of people that are just, looking over and going, yeah, approve.

No, that's not right. Approve, not right, kind of stuff. And, essentially an hour to work a claim will go down, or 30 claims a day will all of a sudden become 120 claims a day. I agree with

that, but there's one thing that needs to come into play. That's the payers. And the providers have got to come together and not be butting heads to, to make this work as data comes from the provider to the payer, and then back from the payer to the provider.

That will be key in these systems maturing and coming into place and actually doing what

you just described. Well, aren't we going to have AI wars? We're going to have the payer's going to have their AI, the provider's going to have their AI, and essentially they're going to start talking.

No. Hopefully they're using the co pilot design contract. Otherwise, you can end up in an endless loop here. But but at the end of the day, Yeah, denied

days go out to years, not months,

right? Yeah. Yeah, well, it's interesting because, first iterations of denials based on computer algorithms has not gone well for the payers Cygnus being sued I think by California because they denied like, thousands of claims in minutes and it was a lot more than a thousand.

It was a ton of claims in minutes. The algorithm just goes through. Well, that's what algorithms do, right? It looks at the various things and it goes no, and they see that as frivolous and whatever. But if the algorithm's written correctly, it may have denied all those claims correctly.

Now I don't know the aspects, but so you have those denials coming back very quickly. But now what you're going to have, which we haven't had before, is you're going to have health systems, instead of labor intensive, high cost response to those denials, you're going to have computer generated, computer automated, AI driven, and human assisted responses.

That are literally in hours instead of, days and the queues building up and that kind of stuff. So, these cycles are going to get shorter. I think it's going to benefit the patient, to be honest with you. I think it's good stuff. I agree. Charles, what else is going on in your world? It's always fascinating to hear what you're up to. Yeah, I think the,

The big thing that I'm working on that I described earlier is understanding, especially in healthcare, the intent of the prompt, whether it's generated by a patient and or a clinician.

And I think the multimodal aspects of, how that is generated, whether it's voice like you and I, it's, American Sign Language. It's keyboard, and we're even working on from a lip reading or CGI type of, perspective. How can we enable, patients and clinicians of all sorts to not only get that information in, but feedback.

To ensure that once that intent is passed on, that intent is accurate and, or the prompt is accurate, and it's meaningful, and what follows is is what the the user is expecting, and then again I this is personal. I don't believe that we're going to have I don't even use the word large language model.

I use language model. I don't think there's going to be one all encompassing language model.

I just don't. I'm hearing a lot of smaller targeted, specifically trained models that are orchestrated by some... Orchestration.

So that's the other side of it. So taking that prompt and then doing the orchestration to which language model do we hit, one combination or et cetera, and I just love it because we're in healthcare, the interoperability aspects of.

Language models. Because we haven't gotten it right with anything else we've done, but maybe we'll be able to be successful in this aspect

of it. Yeah, it's going to be interesting. we look at these models, I'm going to want to talk to you more about what this means for data harmonization. Essentially definitions, metadata, and that kind of stuff, because in theory, some of these models could be trained to answer appropriately, length of stay and that kind of stuff, based on the context, and it could come back with, the appropriate stuff, the form of, When it responds, it gives us references like this is where I found it and that kind of stuff.

Yeah, and I think, Bill, the other thing is,

Now we've introduced a vertical database. So we have to educate on what that's all about versus, a graph database. We're pretty familiar with NoSQL and and relational databases and whatnot, so, whether the differentiation, that's a lot of what I'm doing now is from an educational perspective, what can you do in a graph environment that you can't do and a vertical environment and so forth. And how, what are the differences? What are the use cases? And so forth, because I think it's important that we're all, educated, and we really understand these technologies, at a certain level. And, I'll go back, human beings can make mistakes and be forgiven.

Technology gets to screw up once, maybe twice. And then, There's a halt to it all. So, it's really important in healthcare, as we progress responsibly and ethically and make sure that we're always doing the right things and we're not, putting, anybody

in jeopardy.

Charles, I'll give you the last word. So, thank you very much for coming on the show. It's always great to catch up with you. And to I don't know, the world is changing in front of us, so it's good to catch up and see hopefully you're, I see you as my lens into the future.

Yep. Oh I

like that. So, we'll end with Andreessen's quote the, the definition between a hallucination and a vision. Other people can see the vision,

hallucination, nobody else sees except for you. That's, And that's what people think. You're crazy. 'cause it's like, oh, you see it? No, I don't see it. I don't see it. And if when computers do that yeah, it's I we are utilizing chat GPT fairly extensively. And we're looking at some other large language models and the hallucinations are funny.

It's like, it's fun stuff. Yeah, it really, it's like, if I ask you to do a bio of Charles Boise, will come back. And some of the things that we'll put in there, I'm like. I don't think he's ever won a beauty pageant. I could be wrong, but maybe.

I'm pretty impressed with the bio that Chad GPT creates on all of us.

It's much more impressive than anything I

would have ever come up with. Flowery language and whatnot. Charles, thanks. Thanks again for your time. All right. Thanks.

  📍 And that is the news. If I were a CIO today, I think what I would do is I'd have every team member listening to a show just like this one, and trying to have conversations with them after the show about what they've learned.

and what we can apply to our health system. If you wanna support this week Health, one of the ways you can do that is you can recommend our channels to a peer or to one of your staff members. We have two channels this week, health Newsroom, and this week Health Conference. You can check them out anywhere you listen to podcasts, which is a lot of places apple, Google, , overcast, Spotify, you name it, you could find it there. You could also find us on. And of course you could go to our website this week, health.com, and we want to thank our new state partners again, a lot of 'em, and we appreciate their participation in this show.

Cedar Sinai Accelerator Clearsense, CrowdStrike, digital Scientists, optimum, Pure Storage, Suretest, tausight, Lumeon, and VMware who have invested in our mission to develop the next generation of health leaders. Thanks for listening. That's all for now.