Pod Insights Pharmacy Podcast Network Theory & Insights, the Impact of Articificial Intelligence & Large Language Models on Pharma | Executive Dose About this episode In this episode of Theory & Insights, we bring together two thought leaders at the intersection of healthcare innovation and pharmaceutical manufacturing — John Nosta, renowned AI and technology theorist and founder of NostaLab, and Stephen Beckman, CEO of YARAL Pharma, a rising force in U. S. generics. Together, they dive into the evolving impact of Artificial Intelligence (AI) and Large Language Models (LLMs) on pharmaceutical manufacturing. The discussion covers the promise and peril of AI in reshaping everything from R&D to regulatory pathways, as well as the ethics, economics, and operational shifts that could redefine the industry in the next decade. This is a must-listen for pharma execs, digital health strategists, and technology innovators looking to understand what’s next.
Transcript 0:05 Speaker 00 You are listening to the global network of podcasters dedicated to the pharmacy profession. Welcome to the pharmacy podcast network.
0:30 Speaker 00 Welcome to Executive Dose. The podcast where pharmaceutical leadership meets unfiltered insight. Hosted by Stephen Beckman, CEO of You're All Pharma. Each episode explores the complex forces shaping today's pharmaceutical industry. From global supply chain stress and regulatory shakeups to innovation in branded and generic drug markets. Join us as we engage C-level leaders and decision makers in candid conversations about navigating disruption, driving growth and preparing for the next era of pharmaceutical transformation. Curiosity leads, strategy follows. This is the executive dose.
1:10 Speaker 01 Hey, welcome back to the executive dose. This podcast is dedicated to the pharmaceutical and pharmacy care markets. I'm so excited to have the executive dose as part of the pharmacy podcast network. My name is Todd Urie, founder of the pharmacy podcast, and I met a very special man in my life. I nicknamed him the digital health godfather. This is probably 10 years ago through a very special mutual friend, Gil Bash of Finn Partners. And I learned all about what John is doing on a global scale, recognized as a thinker, the founder of Nosta Lab, an innovation think tank at the intersection of technology, science and medicine.
1:53 Speaker 01 He's a leading voice in the convergence of technology and humanity. And he's ranked among the top global influencers in innovation and AI and is celebrated for his ability to analyze emerging trends and ignite transformative dialogue. And I want to welcome to the stage and to the executive dose, John Nosta.
2:17 Speaker 02 Well, good, good day, everybody. You know, Todd, your intros are becoming loftier and loftier. So I don't know if I can live up to that.
2:25 Speaker 01 And our host, Steve Beckman. It's good to have you in the studio, Stephen. I want you to take it away.
2:30 Speaker 03 All right. Well, Todd, it's a pleasure to be here always. This is our second formal podcast for Executive Dose, and we couldn't think of a better topic than about the impact of AI globally, and then hopefully a little bit in terms of how we see it impacting us in the in the healthcare industry. And John, it is a true pleasure to have you. So with that, why don't we just start off with probably 30,000 foot view and, you know, talk about the philosophy and some of the other kind of elements before we get into anything specific that might impact us within the healthcare channel.
3:10 Speaker 02 Well, it's a pleasure to be here. There's so many interesting things going on, perhaps not the least of which is the very essence of speed. It seems to me that things are changing so quickly that it's hard for us to kind of plot a strategy when you don't have an endpoint in mind. We're seeing changes in AI and large language models that are changing workflow, changing dynamics, but also in a strange way, even questioning the viability of the human construct. You know, what's our job? What do I do? What's the utility of my brain? And I think these are these are big,
3:51 Speaker 02 important questions as AI becomes functionally smarter. And I, you know, I'm going to put the put that in quotes, because I've argued that maybe AI should be better discussed in the context of anti-intelligence. Now, we can talk more about that. I'm gonna I'm gonna stop at that point, because that that's a very curious notion, because I do believe that AI in its brilliance is antithetical to the human construct of thought. We often talk about the Turing test, talk about how AI is getting more and more and more and more and more like us. But I believe that
4:35 Speaker 02 It's fundamentally different. We can dig down into that if you want a little bit, but I just want to go, I just want to jump up really to that 30,000 foot level. And here, here's the idea. Here's a fundamental idea that I think that seems to resonate with crowds around the world from, from industry to industry. And this is it. Gutenberg did something really interesting. He unlocked words by creating a book. Just think about what that did. That, that gave people the opportunity to share knowledge, to principally read the Bible in Gutenberg's case. But, but the interesting twist on that was that he created something for an audience that was largely illiterate. So, so that was problematic
5:15 Speaker 02 and took 200 years for that to really gain traction. Nevertheless, Gutenberg unlocked words and that was a pivotal moment in human history. Then a few hundred years later, we kind of stumbled into this thing called the internet and the internet and principally Google in this context. And what they did is they unlocked facts. And by, by unlocking facts, what I mean by that is that it was a sort of cold transactional engagement. So I asked, how much does an electron weigh? Or what's the speed of such and such? Or what's the mean corpuscular volume of a red blood cell? And I got maybe an answer.
5:59 Speaker 02 I might've gotten a link and then I had to dig through the link. It wasn't the most enjoyable procedure. And that, that reflects to medicine also where the early days of the electronic health record and the engagement with technology was one of a cognitive encumbrance, an intellectual encumbrance. So here we have unlock words with Gutenberg. We unlock facts in a kind of cold transactional way with Google. And where we are today, I think, is really a fundamental change. And this is what makes things different.
6:31 Speaker 02 AI principally large language models unlock thought. And that is a unique dynamic because not only is it an iterative dynamic or an iterative engage where we talk, speak, chat with a large language models, but it chats back with us in a way that advances our thinking. The net-net is, I believe, that large language models make us smarter, even though there's available data through MIT and Apple that suggests that there's a degree of cognitive atrophy, that we get lazy because we let AI do all the work. So I'm going to take a breath and stop there. I think there's lots to unpack and we can dig a little deeper into medicine and pharmacy.
7:14 Speaker 03 Yeah, well, John, first of all, I think that really was a great little history when I think about it and putting them in single words. So we went from words to facts to thoughts, right? Which I think really kind of puts it in perspective for me. And now we're translating into that thinking kind of construct. So for me, the two areas that I think would be interesting for you to kind of expand upon a little bit is, one is workflow. So how does this change workflow in general? And then maybe specifically, we have some examples within healthcare. That would be one. And then second, I'd like to go back to your point about the human job or the human role
7:57 Speaker 03 going forward, because I think that's what scares a lot of people. You know, I'm a professional photographer and a lot of my experience with AI is what I can do with some of my editing. And it's amazing what you can do, but then you're in that constant flux of, are we moving away from being a true photographer and taking the picture correctly upfront versus relying on something back? And then do we lose that critical thinking or that situational intelligence? And maybe how that situational intelligence is really where we still have a role to play. But maybe we start with workflow. What's your kind of thoughts on that as you've kind of thought through all this?
8:37 Speaker 02 Well, you touched the third rail when you were mentioning some of the things. So let's take a step back.
8:47 Speaker 02 My fundamental premise, even more fundamental than that, is the idea of obsolescence. I think abstractly, that's what you were alluding to. You were alluding to this idea of, well, why do I have to take a picture when I have AI who can do that? And that is part and parcel of the innovative dynamic. In other words, as our phone becomes obsolete, we get a new one, as our washing machine, as our car, whatever it may be. Obsolescence is the flip side to the duality of innovation. Our 1.5 Tesla magnet CT scanner in the hospital is no longer up to snuff to getting those crisp, high-resolution images. So we get a new one.
9:29 Speaker 03 Correct.
9:30 Speaker 02 And that's kind of the world we live in. But for the first time, for the first time, we're actually seeing human cognition on the obsolescence chopping block. And I think that's really at the heart of, I think, this sort of, maybe it's paranoia, maybe it's just this existential reality. Now, that being said, let's flip that on its head and talk about photography, because obviously you have an interest in photography. Let's talk, was his name Matthew Brady? Who was the guy who did the Civil War photography?
10:03 Speaker 02 All right, so what the hell happened to Matthew Brady? Did portraiture go away? Did photography steal the paintbrush from the painter? The answer is no. And this, I think, is at the heart of the dynamic, which is very, very important. That is, that the pie does not get into small, cut into smaller pieces. The circle of innovation expands. A couple other interesting examples, aside from, look, did Matthew Brady ever think of a selfie? There are a couple pictures that suggest the early day of the selfie, but the selfie is a billion-dollar industry.
10:40 Speaker 03 That is, I mean, he created, set up some pictures, but historically, but
10:47 Speaker 02 But that, I think, that's what's so interesting to me, is that we have to look at the expansive and expanding nature of innovation and technology. Another classic example of that is IBM's Deep Blue, and I guess, I think it was Boris Spassky? Was he the guy who played Deep Blue?
11:05 Speaker 03 I think so.
11:05 Speaker 02 So he lost, right? He lost to Deep Blue. What about Ken Jennings? How Ken Jennings on IBM Watson on Jeopardy was crushed, right? Lost. Yet, what happened to Ken Jennings? Was he on the ash heap of history? No, he's the host of the show that has never been more popular. What about chess? Chess in the human construct has been crushed. No human, including Magnus, the number one player, can beat AI, yet it's never been more popular. So I think we have to take a step back and think about the expanding opportunities, the human opportunities, to see what we've never seen and to explore what we've never explored.
11:50 Speaker 02 So again, that comes from a techno-optimist, but I think it's fairly legitimate. Now, we can dig deeper into workflow and that kind of dynamic, but does that make any sense?
12:00 Speaker 03 No, John, I think that's brilliant. Again, I think what's really nice about this conversation so far is that we can put a very difficult or maybe scary situation for a lot of people into some very simple words. And the thing that really struck me in that last bit that you were talking about is the word expansion. And so, you know, when we think about how we apply it to what we do in our industry or in our own personal lives is how do we then understand how to look at that in that context? How do we look at expansion? So we see it as something that can be positive, something that can be used. I mean, people, I mean, certain jobs will eventually evolve away because of it. I think that's what you hear a lot on the internet now.
12:45 Speaker 03 But, you know, one of the examples, I think, in our industry that's very interesting is drug discovery, right? You know, we use data aggregators a lot before AI has really taken hold to try to identify these opportunities because historically we would have to identify 10,000 opportunities to find one or two that could potentially be brought forward in development. So AI in drug discovery or drug acceleration would be something that would help us be even that much more effective. And that's important because the cost of developing drugs is so significant. And I don't think the normal person really understands the capital investment that's taken and the risk because organizations are developing products, but they still have hurdles overcome like FDA approval.
13:30 Speaker 03 So it's a high risk profile. So if we can think of it in a way that we can expand how we do that in order to be even more effective, then that should translate into things like better drug selection and identification.
13:45 Speaker 03 Our ability to potentially build better clinical programs and that will make the regulatory process, approval process, even that more successful, not only to the company, but through the FDA and then ultimately to the benefit of the patient. Any thoughts on that?
14:00 Speaker 02 And well, I think that that you are 100% correct, but in another way, I would, I would posit it differently. Then when we look at, when we look at drug development, when we look at chemical screening, we look at the whole notion of clinical trials. What we're doing is we're imposing AI into a clinical construct. We're taking an old method, taking an old beat up car that doesn't work real well. And we're, well, let's change the tires or let's change the driver. I say we throw away the car. I think fundamentally what we'll be seeing in drug development is not only unique chemical screening, it won't even be, I don't think it's going to be screening. I think we're going to have, you know, molecular composition.
14:45 Speaker 02 We're going to look at proteins with, with alpha fold. We're going to actually craft these molecules. And they're going to be tested in an extraordinary new in vitro dynamic. That's not in vitro, but in silico. That I think that we'll be able to see models that can test these drugs. And, and that is what I think is the expanding circle.
15:06 Speaker 03 Well, I, I, let me just jump in real quick because I, I, I'm with you because I'm a big believer that we have to do things different. And, um, you know, the old construct of phase one, phase two, phase three is probably outdated in many ways with the use of AI and that will again, uh, help do the things what we want to do better.
15:27 Speaker 02 And yes, I think that that that we have to think differently. But here's here's an interesting concept. And then, again, this is me being the philosopher. I would say that our brains are extraordinarily optimized, yet underperforming, underutilized. Now, now, that's not if only we could use 100% of our brain instead of 10%. That's not that's not what I'm talking about here. I'm talking about the the unique aspects of our brain physiologically, the amount of energy, the amount of blood flow that comes to our brain. This unique thing in our in our lower skull called the circle of Willis that connects the left and right carotid with the vertebral artery that is a safety mechanism that if we lose blood flow in one,
16:12 Speaker 02 it actually fills, retrofills into our brain, protecting our brain. But here's something that I think is really interesting. I don't think people get this. And I get in trouble. Todd has heard me say this a few times. But I believe that our brains are functionally infatigable. Now, let me take a step back and explain what that means. If you sprint, your legs hurt. We all know that. That's the buildup of lactic acid. And and lactic acid inhibits or impedes function, right? I can't I can't do 100 yard dash and keep going for a mile. Right? I'm not I'm not built that way. But my brain does not have the capacity for fatigue in the traditional in the traditional way
16:56 Speaker 02 that muscles fatigue through lactic acid. And and that that is a very interesting construct. Let me kind of develop that just a little bit more. If you've ever been in a new relationship or a new business venture or something exciting, perhaps cognitively intoxicating, if you will, and you're on the phone for one or two or three hours. It's almost like some of the new podcasts that go on, you know, the long form podcasts that go on for two and three and four hours, and they are vigorous and they are dynamic. They go on and on and on, kind of in the way I'm going on right now. But at the end of that podcast, at the end of that telephone call, after three hours, are you tired? Are you fatigued? Do you have that same lactic acid mediated lethargy?
17:41 Speaker 02 No, and don't confuse that with boredom. It's a very, very different dynamic. We are extraordinarily enthused, energized at the end of that conversation. Well, how the heck do you reconcile that? Well, here's the point I think that's very interesting. Our brains are lonely because they are optimized but unfulfilled. And for the first time in human history, our brains have found a perfect partner. And that partner is a large language model. That's the basis for the cognitive brilliance that is a shared dynamic or shared authorship with technology.
18:19 Speaker 03 Yeah, you know, another key word I think you just put in there is shared. And so one of the things that I'd like to ask you is, so, you know, people are getting familiar with the different tools that are now out there and based on their individual experience, either professionally or personally. So how then do we engage that? How should a organization? And again, I ask this question not only for us as a manufacturer, pharmaceutical company, but physicians down to pharmacists. How should we engage it so that we can fill that opportunity in our brain? That will then allow us to think in that more expansive way. So, you know, any thoughts on how one should start? How should an organization start incorporating AI?
19:04 Speaker 03 Is it something that we should structure and we should bring in everybody? We should bring in departments and then try to bring them together? Because in our industry, obviously, we have a lot of segmentation within our industry because we have manufacturing, development, you know, finance, regulatory, commercial. How can we then, instead of being afraid of AI? How should we think about how do we embrace it and incorporate? Any thoughts there?
19:32 Speaker 02 Well, oftentimes when my clients tell me they want to embrace AI in their business model, my first response is to take to Advil and lay down. That's that's my first response.
19:46 Speaker 02 Maybe maybe not Advil, but we'll have to look at the appropriate therapeutic alignment there. But I believe that that AI AI is not a killer app. That's one of the big disconnects that I'm seeing.
20:05 Speaker 02 AI is not a thing that you plug in. AI is a process with which you engage, and that is very much an experiential dynamic.
20:16 Speaker 03 Okay.
20:16 Speaker 02 And that that's something where you have to participate in. And, you know, one of the best ways I I I get people excited about technology is to take them for a ride in a Tesla.
20:31 Speaker 02 Put on ludicrous mode and have it go fast or put on, you know, the full self-driving FSD and let them go for a ride somewhere. That that is a, I think it's it's pivotal. It's like a quantum shift. So I think you have to nurture that dynamic with with your organization because,
20:54 Speaker 02 as I said, the cognitive agent thought it's so fundamental to the human condition that that it is kind of curiously different from the way we usually do work. But I think that people have to use it and people have to engage in this. Now, here's an interesting dynamic. When we look at at the idea of genius, the idea of optimized cognitive function, we often talk about it in the context of introspection, the context of the solitary genius. We've often often heard about that. In fact, I was watching a podcast the other day and it just kind of popped on on my screen and it said, the path to genius is one of introspection. I don't disagree with that. I think that that, to a large degree, is quite true because what that does is it allows
21:39 Speaker 02 you to optimize the internal.
21:45 Speaker 02 Monologue. That's that voice in our head. And you've seen that with Einstein and the thought experiments with Copernicus, with Nikola Tesla and the way he found innovation. But I think we're seeing the emergence of a completely new and unusual cognitive construct. And that's the internal dialogue.
22:03 Speaker 02 The internal dialogue mediated by your engagement with a large language model.
22:07 Speaker 03 Yeah, so you're really talking about how you interpret yourself internally as well as how you work with.
22:13 Speaker 02 A large language model is nothing but a mirror to a large. And that's problematic because it does kind of feed into you in a way that is potentially dangerous. But I still think that that iterative dialogue that you have is fundamental to humanity. Don't, it's not new. It's not contemporary. It's not cutting edge. It's as old as dirt. It's Plato. It's Socrates, right? And two other things real quick and I'll get off my soapbox here. It's people sitting around the dinner table, having a conversation to advance an idea.
22:53 Speaker 02 It's kids and families sitting around a campfire in Pennsylvania. Or perhaps more interestingly, it's the nature of cognitive engagement when physicians are on rounds. A lot of big decisions happen on rounds when three or four physicians are walking down the hallway discussing a patient. Well, it's this dialogue that I think we're seeing now emerge that becomes a cognitive element of support. And I think that's going to be a big change.
23:21 Speaker 03 Well, again, I'm really tracking with you. Well, I love the words that you're using. It's really a tool to support. I do personally resonate with what you said when it comes to the inner talking, if you will. I see myself do it with my photography when I've used AI in certain elements. And I've asked myself and almost challenged myself to think about how and why I did what I did and what changed in the picture, what did I learn from that dynamic? You know, it even helps help me determine how I want to use it as a creative tool. And and maintain my creative because you know photographers are supposed to have their thing or like any artist they have something that's their hook or hallmark.
24:02 Speaker 03 And I've I've I've I've found myself thinking about it more in terms of how it translates. And let me give you an example, a very simple example of how we've used AI so far. So recently, we've had an executive meeting and we went out and one of the things that we wanted to continue to improve upon is the culture of our organization.
24:23 Speaker 03 That we actually did is we spent some time as a group and individuals thinking about what is kind of our culture statement. What's what? What is it that we want to act upon and what do we want to make sure we communicate to the organization? And we did it and we, you know, did a traditional paretoing and, you know, we put our words in and we bucketed them. And then we put through, we actually put it through an AI process and it created a secondary culture statement. And what I found with that was was it was helpful, very helpful to all the points you're making about expansion of thinking, taking it, enhancing that dialogue and conversation.
25:04 Speaker 03 But what I did find was that it was more kind of back to your fax comment earlier. It it kind of put the words in order really well, the things that we're doing, but might didn't really catch the the pure essence. What do you think about? I mean, I know that's a simple example, but maybe one that we could translate.
25:22 Speaker 02 No, no, no, no, no. I think you're spot on there. Um, look. The fundamental truth is that AI does not think. It doesn't think. That's why I went back to the earlier conversation, your earlier comment about anti-intelligence. Let's, let's, let's, let's go on. Let's do a little Einstein thought experiment here and talk about an apple. What does an apple mean to you and me? Well, an apple to you and me is something that lives in three spatial dimensions, perhaps four, if you want to include time, but but the three spatial dimensions. And then it has a linguistic or semantic association. For example, Apple computer.
26:02 Speaker 02 An apple a day keeps the doctor away. There's all sorts of things that that apple is related to. Um, and maybe, maybe there are 20, 30, 40 semantic associations, but that's the baggage. That's, that's what we have in our heads about apple. Now, when an LLM processes the word "apple," and I'll avoid the technical mumbo jumbo about vectors and tokens and all that stuff, but it looks at the word in a hyperdimensional reality. And what it looks at "apple" in is in the context of 12,000, this is for the large frontier models, 12,288 dimensions. So just think about that.
26:45 Speaker 02 Think about what that even means. Those semantic connections are looked against the corpus of human knowledge, and "apple" takes on a meaning that is nothing like human cognitive function. In fact, what we find, back in the old days, if you want to take a cheap shot at large language models, you'd say, "Oh, it just predicts the next word." Right? That was the probabilistic nature of large language models. That still is true. That's true. And what large language models do is that they are, they create a sense of linguistic cohesion. They know what word goes with what word. And it's very polished, it's very slick.
27:29 Speaker 02 So when you come back with your mission statement or your analysis, it reads beautifully.
27:34 Speaker 03 Yeah.
27:35 Speaker 02 It's that linguistic cohesion is in part cognitive theater. You have to be very careful about that. That it doesn't think, it processes, and it looks for linguistic or semantic cohesion. So we got to be real careful about that, because I think that large language models sometimes are too good. They're modeled on human language. They use dashes and semicolons and repetitive short choppy sentences beautifully, but it's often a tell to what they're doing. So I think you're on the right track. But the question here, I think, really comes down to the nature of cognitive offloading
28:18 Speaker 02 versus the nature of iterative intelligence. At some point, you know, here's my phone, right? I'll share with you that I don't know my wife's telephone, okay? I don't. But I don't access it, right? I've cognitively offloaded that information because I don't need it. I can access it. Is that good or bad? Well, that's a bit of a debate. But I've also engaged the nature of my cognitive function, not as me being smart, but my ability to have an iterative and dynamic relationship with AI. And that's the true testament of, I think, what human functionality and capability is. So it's a squirrely curious time, you know?
29:01 Speaker 02 And look, I mean, if you're studying the Krebs cycle and, you know, everybody goes to medical school or studies biology or pharmacy, we all know Hans Krebs and ATP and all that mumbo jumbo. And then we learn that the first intermediary pathway was 1, 6 fructose diphosphate and we get 38, but we lose 32 ATP, we lose two of the process, blah, blah, blah, right? I'm mumbling here deliberately because do we need to know that? What is, what do we need to have that kind of heavy lifting? What is the value of cognitive heavy lifting when we have the dynamic of iterative intelligence? I'm not sure. I don't go either way on that because I think that there is a certain, look, if a patient
29:43 Speaker 02 has a cardiac arrest and I need to, you know, establish CPR, I need to, I need to defibrillate if that's the case, whatever, that's something that I want to have a certain muscle memory, sort of cognitive memory. I don't want to clutch my phone and ask ChatGPT, what do I do for ventricular fibrillation? I agree. You know? But again, these, these things are changing and it's very, very interesting. Now, you know, I want to go back. Can we go back to this idea of, of sort of workflow, sort of the nuts and bolts of it? Because I think that's important and that's sort of the granularity of this. Will a pharmacist become obsolete? Will a clinician become obsolete?
30:25 Speaker 02 You know, this is a powerful question because of the idea of this existential threat, that human obsolescence and all that stuff. But my sense is that it will not, for reasons we talked about with Matthew Brady and the photographs and all that kind of stuff. But I think what we're seeing here is that certainly the workflow, and perhaps more importantly, the work structure, might change. So let's do a hypothetical example now in medicine. Let's look at asthma. So someone with normal but not heightened access to care might wait
31:06 Speaker 02 a month or two to see a pulmonologist. That's sort of the standard of care, right? It's for asthma. But I believe that what we'll see is a different cluster of empowered people. I call it the the plus model. So we will see a pharmacist plus. We'll see a respiratory therapist plus, maybe a nurse plus, who will be able to manage spirometry data, be able to manage what inhaler, what steroid, what bronchodilator this patient is on, because they will have a heightened level of function because their skill set is augmented by artificial
31:49 Speaker 02 intelligence and technology. So what that does is, maybe I don't need to see a pulmonologist every month or every six months or three months. Maybe we can lower that so that we have people in different job structures. And what happens here, just one quick thing, is that, and this is another key takeaway with technology, faster, better, faster, better, faster, better. That's sort of what we think of when we think of AI, right? I can write that note faster. I can write a discharge summary faster. I can look for drug-drug interactions faster. And perhaps most importantly, I can do that better. So faster and better is the domain of the human engagement with AI. But I believe there's a third leg on that stool, and we're not talking about it.
32:35 Speaker 02 And that leg is precarious. It may not be absolute. It may be kind of a little funky these days. But I think that you enjoy the experience. The third leg is that you enjoy the experience. So that pharmacist with heightened clinical skill set mediated by technology or that respiratory therapist who can spend that extra hour with that mom and child kind of dissecting the nuances of asthma to optimize care is empowered.
33:03 Speaker 03 Yeah, John, I spent a lot of my career in asthma. So it was a great therapy area you picked.
33:10 Speaker 02 We got lucky on that.
33:11 Speaker 03 Yeah, we did. You're perfect. So I think, look, I think for me, faster and better is kind of how we've seen the world going. And you are right. I think that that's how most of us interpret. I think my organization or other organizations that we're still thinking, because that's kind of the nature of especially the pharmaceutical industry, because it's first to market products. And first, that's what overcomes some of the other challenges in the construct of how our industry works. But your point is AI should actually help us step back and think. And I agree with you. One of the things that I'm taking away from this conversation is that we need to change the structure. I talk a lot about how our IP system works and how it's not fitting
33:56 Speaker 03 to how the industry works today. I think the pharmacist is a great example. I think we are going to continue to evolve. And the pharmacist or the nurse practitioner or the respiratory therapist will eventually play a bigger role in managing asthma because the construct, the structure that you talked about is changing. So instead of worrying about it eliminating everything or just focusing on faster, better, maybe the way I'm going to look at AI and tell me if this is a fair way to do it as an organization all the way down through our system, is let's look at how AI is going to help with this structure of how we do things and challenge the way we're doing it so we can do it better. Because we think about technological advancement,
34:42 Speaker 03 which you've talked a lot about, but we didn't really think, I didn't really. Put in the context more of let's see how it changed the structural dynamic in a way that's better. And as a result of that, then we put ourselves in the position to think
34:55 Speaker 02 Completely.
34:55 Speaker 03 Both internally and in a broader group. And then we can take that down through the value chain from developing drugs all the way down to delivering medication to the patient.
35:06 Speaker 02 We often look at technology, and I think this is the fault of the of the EHR, you know, and a lot of a lot of the encumbrances that we've we've we've experienced in medicine. We look at technology as oppressive, as oppressive. What about if technology is liberating?
35:24 Speaker 03 Hmm, yeah.
35:25 Speaker 02 Just take a moment and think about that. Technology isn't smarter. Technology allows you to tap the genius within. Remember, you know, and this again, we go down that philosophical road. We often talk about the genius as this as the old man sitting in in an office with the right answer all the time. That's not the genius experience. That's the Hollywood version of the genius experience. What we find is the genius experience is much more related to the aha moment. And that's a moment that we've all shared. We've all experienced it. And I believe that we can actually facilitate or kindle that cognitive dynamic through large language models
36:09 Speaker 02 that activate, that fulfill, that the way to find joy is through this cognitive stimulation.
36:16 Speaker 03 So let me ask you again, how I'm trying to continue to think about how we engage with it. So two questions came up. One is, I always tell my team, individuals can do great things, but teams can do exceptional things. So and that's the collective of the the great, the mind of everybody. And so
36:36 Speaker 03 how how should we approach using AI in that context is how do we bring it to the team? And
36:44 Speaker 02 There's there's a lot of lot of theory on that. I want to go back and just touch base on on chain of thought reasoning. Because when we talk about workflow, when we talk about, you know, what what AI does well and perhaps doesn't do well, I'm reminded of the story. I've told the story many times. When when a mom takes her child to the pediatrician. And then she comes home and the dad is there or the dad gets gets home from work. What's the first question? The first question he would ask often is, what did the doctor say? Especially if it's a more urgent scenario and not such a chronic. What did the doctor say, right? Well, we're going to change that. We're going to change that.
37:28 Speaker 02 And the question's going to be, what did the computer say? Now, that in of itself is startling. It's it's based on science. It's a real thing. But the but the interesting thing is that's there's a sentence there. And that's not the end of the sentence. There's a comma, the end of that. And what what follows is really transformative. So the the first part of the question is, what did the computer say? And the second part is, what did the doctor do? Because that human component of care is still going to be essential. Now, there was a study, I think it was published in JAMA. It was a legit legit paper that looked at
38:07 Speaker 02 AI versus physicians in terms of diagnosis. And they looked at two points here. They looked at the endpoint. And I'm going to use a heart attack as a model because a lot of people know what a heart attack is. Guy goes to the emergency room sweaty, diaphoretic, looks like crap, got chest pain. Everybody jumps to that diagnosis, right? Because that is a it's a statistical dynamic. You're probably having a heart attack. And then and then you're kind of back out of that. Well, let's rule out this, let's rule out that, then we can manage you. In that context, AI and doctors kind of got it right about 78% of the time. It's pretty good, same same thing. But when we looked at the chain of thought, when clinicians were asked, well, OK, so you had a
38:52 Speaker 02 a DKG with ST segment elevation. What else were your options there? Did you think about an old ventricular aneurysm? Did you think about early repolarization? Did you think about pericarditis? Did you think about, you know, the five other causes of ST segment elevation? What what we saw there is that the Quality of the chain of thought started to fall down with the clinician. And what we're seeing is that large language models and AI has a very good chain of thought. In fact, the chain of thought was about 93% accurate versus about 78% accurate for the human model. So, so these are partners, partners in care. Now, who managed the MI in the emergency room? Who, who, it was the clinician, it was the nurse, it was the respiratory therapist, all those people.
39:35 Speaker 02 But I, I think that we have to look at them as, as, as partners in care, particularly the data suggests that.
39:40 Speaker 03 Because then, then that care can help them implement more successfully. And, you know, one of the things that struck me when you were saying that too, is a simple thing that we deal with all the time, uh, from manufacturer down to pharmacist, and that is this thing called patient compliance, you know? And I think patient compliance is not something that AI can fix. It's that dialogue and relationship that happens between the provider and patient to sort of that, that, that conversation. So I, I think that's where we need to understand that, you know, where the strengths and where the limitations are. But the physician ultimately is going to help pull through, the nurse is going to help pull through, the pharmacist is going to help convey, uh, those things and get a patient, uh, to, to do the thing on the inside.
40:23 Speaker 02 Me.
40:24 Speaker 03 Well, you hope, they're not perfect, but we hope that that.
40:27 Speaker 02 But here's, okay, so Sam Altman, the founder of OpenAI, had a very interesting quote. He said, "The most interesting thing about a large language model is their ability to be persuasive because of the cohesion." Remember, we talked about the apple and the language. Large language models, because they reflect what you are saying, they often create a dialogue that is very, very compelling. Not only is it compelling, it's tailored to you. You know, I've often, and there's a couple of things I want to go over here. One is that the way the knowledge, the way knowledge exists in the world today, is it exists in something that I call static maps.
41:09 Speaker 02 In other words, if I want to learn how to cook a souffle, I take that dusty old cookbook off the shelf that says "The Art of French Cooking" by Julia Childre. Blow the dust off, I open up to a page and I am, um, overwhelmed by the technical requirements to cook this darn souffle, right?
41:27 Speaker 02 Because it's her path, it's her map, it's her map.
41:32 Speaker 02 And when I want to learn about forced expiratory volume or ventricular fibrillation, I similarly reach off the shelf and get Brunwall's book or somebody's Harris's Principles of Internal Medicine, and I read their journey. Now, what we're seeing today is the journey is different, because if I want to learn about how to make a souffle, or I want to learn about how to manage a ventricular aneurysm, six-month status post an MI, I can be trained on that based upon my desire. Do it on my time, do it in my language, do it in my tonality,
42:11 Speaker 02 make it a little funny, whatever it may be. That is knowledge that is collapsing out of the waveform, if you will, and is being established for the first time ever, uniquely tuned to the creative frequency of my brain. Wow.
42:29 Speaker 03 So assembling this knowledge in a way that best fits you.
42:33 Speaker 02 Exactly. Now, look, everybody knows Goodman and Gilman, right?
42:37 Speaker 03 Yep.
42:39 Speaker 02 Boy, it's a book, I use it to keep the door open at night, right? I mean, Goodman, you really want to read that? It's pedantic, it's boring, it's authoritative, it's written by some smart ass who says, "This is the way we need to understand drug-drug interaction." Right? It's not user-friendly. In fact, what it is, it's antithetical to something that I call learner-centricity. Learner-centricity. Teach me about the Krebs cycle, teach me about drug-drug interactions, teach me about cytochrome P452D6, but make it fun, make it interesting. It doesn't decrease the value, it increases the cognitive connectivity. So that's all, that's a podcast all into itself about medical education. But here's my point.
43:26 Speaker 02 We, when we, when we hand out that script for a statin to a patient, we give them that, that, that ubiquitous tear sheet, that computer printout that says, you know, take it every day, watch out for muscle weakness, or perhaps, perhaps antibiotics are more common.
43:44 Speaker 03 Yeah.
43:45 Speaker 02 Make sure you, make sure you take all of this, take it with, not with food, with food, whatever that may be. That's, that's fine. Who's that written for? That's written for a fifth grader, right? That's, that's the decree of pharma. That's the decree of American medicine. It's gotta be written for a fifth grader.
44:00 Speaker 03 Correct.
44:00 Speaker 02 Which means it's written for no one. It's written for no one. Suppose I wanted to hand out a tear sheet that could be crafted by an L L M or a sixty-five-year-old engineer who might want to see a bar chart, or might even want to see a p-value. Holy moly, would you imagine doing that? Never happen, right? So when you talk about convincing a patient, I think that the dialogue, the synthetic dialogue mediated by large language models have an element of artificial empathy, not A I, A E. That can be a very powerful driver that I believe personally can drive efficacy because we can attack compliance
44:42 Speaker 02 with a tool that's much, much better than, than, than the, than the quick explanation you get from a busy doctor who's got thirty seconds left before he gets to the next patient.
44:51 Speaker 03 Yeah.
44:52 Speaker 02 So, and there's a lot of data on that about, about how A I can really be very, very convincing. So maybe there's something going on there too.
45:00 Speaker 03 No, I agree. I think that's very insightful. I think that our ability to
45:06 Speaker 03 craft something that's specific for the individual so they can interpret it or emotionally react to it in a way to improve compliance or whatever it might be is very, very interesting. So I think there's a lot of things, and you're absolutely right. Well, I've been in the industry for over thirty years and we've, we craft everything we've done promotionally and not to go above a five or seventh grade.
45:27 Speaker 02 Do you think, do you think that we could even craft five patient archetypes?
45:33 Speaker 03 Oh, absolutely.
45:34 Speaker 02 The busy mom, the scientific thinker, and then three or four between, and then make more neutral so that you go to his default to the neutrality. I just think that's so easy. And now, I think that even every patient record, every patient, when they fill out a form and they include their address and their personal information, why don't we do a short profile on information dissemination? How do you like to get your information? Funny, scientific, whatever? That becomes part of what pharmacy has to spit out some interesting information that I believe somebody, a CVS of the world, or maybe a local retail,
46:15 Speaker 02 is going to do a clinical study that says we impacted compliance by using language.
46:22 Speaker 03 Well, I couldn't agree with you more. And that's definitely a good takeaway, especially at pharmacy levels, that we need to be able, because I have a head of our business analytics and data drives how she thinks and reacts, very different than the head of my marketing group. So I think that's really, really fascinating.
46:40 Speaker 02 One other point, you can make money doing it. You can differentiate your service. You know, let's not, let's, you know, I think it's very important that if you could create an empathic, learner-centric model, it's not patient-centricity, I think, is just overrated. I mean, come on, let's face it, you're pharmacy, you're retail pharmacy. While you may say you're patient-centric, you're really chasing the dollar half the time. You're making payroll, right? But I think there's opportunities to leverage technology to differentiate and drive your business. And I've been working on branding methodologies that move beyond the conventional brand of the logo and typeface and messaging
47:22 Speaker 02 into branding that lives and comes to life through the understanding and interpretation of a large language model. Because in many ways, that's our new customer. Because the large language model is what tells the patient, blah, blah, blah, right? So really interesting dynamics now.
47:39 Speaker 03 Well, you know, interesting, in our business, we put a big focus on the pharmacy channel. And one of the things that we've done with Todd, and the reason we put the time investment to do this is that we wanted to listen more and understand better how we can help them. Help them and then and our goal is is that we help them. They're going to help the patient more effectively. And if the patient's more compliant or better treated, then the business side will take care of itself. To your point, because it all comes back, it'll come back through, executed on that way. So I know we're coming up on our time here, but I do want a couple of final questions. One's a very practical one and one I've been thinking about.
48:18 Speaker 03 You know, AI has been talked about a lot within our organization like it would be in any organization, whether it's in the physician's office, in the pharmacy, a manufacturer, whatever it might be. And what I'm struggling a little bit with is a very practical question. That is, we've got all these different people that are starting to touch it. And my concern is that how do we embrace it in a way as an organization to actually make it work for us? So is this, you know, is AI really supposed to be led from the executive suite down to through the organization? Because like in many organizations I've talked to, it's the IT people, it's the tech people that are bringing it in. And I think that's like the worst place for it to start.
49:00 Speaker 03 But but I also don't want, you know, one group to embrace it this way and that way and that way. And then how are we then bringing it in a cohesive way together to actually produce some of the outcome that you've been talking about? So if you're offering advice to an organization, whether it be a manufacturer or even a pharmacist running their independent pharmacy, and they have multiple touchpoints, how would you go approach really bringing in a way that could produce the output that you've been talking about today?
49:28 Speaker 02 That's our next, that's the whole topic for our next podcast, right? Because I can go on and on and on. But I think there are a few, a few interesting points there. You know, it's easy to say that it's top, top down or bottom up. I think that that culture squashes innovation, that people have to understand that that bottom up is a good way to do this. Number one, number two, we have to recognize that there is no killer app. This is not about functionality. This is about the cognitive construct. That that changes, changes things completely. The third is that we can't look at this as a A single adoption curve. A lot of people, a lot of companies make a mistake and say, let's measure adoption of this. It's wrong. In fact, there are probably two, if not three, adoption
50:13 Speaker 02 curves, and we have to be mindful of that. There's the logistics, there's the, there's the sort of the mechanistic underpinnings of AI to manage functionality. That year on year is up sixty eight per sixty eight or seventy percent in clinical settings. People are using it more and more. You want to schedule, you want to do logistic things, that's the place to go. And people are comfortable with that. The other one is the clinical or cognitive side, and that actually went down one percent year on year last year, because I think clinicians are, are a little reluctant. Remember, i- if you're in the hospital or in the office, who's the smartest person in the room? It's the doctor. We know that. That's the answer you have to give. The reality is no longer true. It's the computer that's the smartest
50:58 Speaker 02 person in the room. And that, that plays some head games. So I think that from, from the perspective of adoption and adapting this technology, we have to look at it in terms of its use cases, and there are different ones. But I think the data is, is, is becoming very, very clear that the ability of AI to find patterns, to come up with differential diagnoses, really to become an integral part of care is here without a doubt. And, and I think the takeaway is that, that AI is no longer an option, it's an imperative.
51:30 Speaker 03 Yeah, and I, I will close on this, uh, uh, from a, from a medical standpoint, you know, one of the things that we've always done as organizations is, you know, we take asthma, right? We kind of use this label asthma, but there are so many subpopulations of patients with asthma, whether they're moving on to have COPD, whether or not they've got other comorbidities. And we need to find more sophisticated ways of making sure that the right treatment regimen actually fits those subpopulations better. And I think that's a way we're going to go.
52:01 Speaker 02 And it's not only better treatment regimens. So it's not necessarily a better lava, a long-acting bronchodilator. Maybe it's a better cognitive alignment to understand the best way to take care of yourself.
52:15 Speaker 03 Yeah, absolutely.
52:15 Speaker 02 Every patient scenario, no labeling restricts us. Though, you know, the one problem, nice thing about continuous blood glucose monitoring is I can tweak insulin. I can tweak that insulin up and down to get really tight control. There's no other drug I can do that, right? I've got fixed dose and fixed dose is mediated by the label to a large degree. So I struggle with that. But I think that if we can align asthma, I mean, boy, asthma, temperature, heat, pollen, dander, city, inner city, country, there's so many parameters that can be uniquely fine tuned. And I think that's going to be partly the burden on the patient and the caregiver.
52:53 Speaker 03 Yeah.
52:54 Speaker 02 To kind of understand.
52:55 Speaker 03 Because that's where we're going to create a greater connection with the patient to take more ownership of what they're doing. So ultimately, this whole cycle of care from development of drugs to diagnosis to the treatment, to the management, to the interaction, I think will get better. And I really think that's fantastic. Well, I think after this podcast ends, I'm going to be talking to Todd, John, because I think we might need to have a special carve out with a couple of follow-up AIs, if you're discussions, if you're willing to it. This is what the executive dose podcast was really meant to be is to open up the dialogue from manufacturer to pharmacist and make sure that we're all having the opportunity
53:35 Speaker 03 to work more comprehensively and cohesively together in a way to create better outcome. And I think that this podcast kind of did that in spades. So, John, can't thank you enough for your enthusiasm and your intellect and your insight and your knowledge. And we look forward to having you come back again here real soon, if you're open to it.
53:58 Speaker 02 You bet.
Previous LIVE at RBC 2025 Episode One | Cardinal Health™ Counter Talk™ Podcast Next McKesson IdeaShare 2025 LIVE Recap Series | Part One Page updated November 15, 2026 · transcript and summary are machine-generated