The Power and Peril of Generative AI in Biopharma and Healthcare

Life Sciences DNA podcast is sponsored by Agilisium Labs, a collaborative space where Agilisium works with its clients ranging from early stage biotechs to pharmaceutical giants to co -develop and incubate POCs, products and solutions that improve patient outcomes and accelerate the development of therapies to the market. To learn how Agilisium Labs can use the power of its generative AI for life sciences analytics to help you turn your visionary ideas into realities, visit them at labs.agilisium .com.  

You're tuned to Life Sciences DNA with Dr. Amar Drawid.

Amar, we've got Christian Hein on the show today. Who's Christian? Christian is an executive advisor and board member in HealthTech and AI. We've worked together at Novartis where he served as the vice president and global head of digital transformation and innovation execution. Prior to that, he spent nearly 10 years with Amgen where he headed EU digital technology hubs for Amgen innovation. He has consulted with a variety of biotech and healthcare companies

and helped guide them on their digital strategies. And what are you hoping to discuss with Christian today? So I want to focus on the use of generative AI and AI in general and the promise in biotech and healthcare that it offers, how it's deployed today, what are the challenges we're learning about, and then what it's going to take to be successful with this technology. You know, we've seen a big push around digital health and using those technologies to improve things like care and patient monitoring, medication compliance. And one of the things this did was open up the potential to capture all kinds of real world data. Does AI now enable us to take all of these disparate kinds of data streams and provide actionable insights? Yeah, I think it should. And that's very exciting. And I'm hoping that now we can...

move from the abstract to some concrete examples from Christian about what's being done today and what can be done in the near future. Before we start, I wanted to remind our audience that if they want to stay up on the latest episodes to hit the subscribe button and if they enjoy the show to hit the like button and share their feedbacks in the comments section. I also wanted to remind the audience that they can listen to these episodes on their preferred podcast platforms as well.

With that, why don't we welcome Christian to the show and hear what he has to say.

Christian, thanks for joining us today. We're going to talk today about digital health, the promise to transform healthcare with that, and also how the integration of generative AI has the potential to turbocharge the type of data that we can gather and the insights that we can generate from that. And you played a prominent role in big pharma, big biotech settings in this area. So can you tell us a bit about that, about when we talk about digital health technologies, what does that mean and what are the kind of interesting things that happen in this area? So, I'm going to thank you first of all for having me. It's really a great pleasure to be here. Digital health, you ask five people, you get six different answers. I've seen some half decent definition, particularly when you take the kind of diamond digital medicines definition where basically digital health is the broadest. You go to the center, have like a digital medicines definition in there.

And then even further in that, you know, onion, if you if you peeled on even further, you would go to digital therapeutics. Digital health, in a broader sense means I'm using digital technology somewhere to improve the health. That could be stuff like, you know, your Fitbit, that could be stuff like a lot of very basic technology we're using, as long as they're applied for healthcare settings. So it is a very broad definition.

I believe what obviously you and I are most interested in is the part where it becomes somewhat evidence based which means somebody has gone through the hard work of collecting some form of clinical data around to show that this thing actually does what it was supposed to do and not just, you know, based on, you know, wishful thinking. But but to me, digital health is a relatively broad term. But I think you and I should really go down to the more specific terms of anything that is evidence -based, that is not just consumer lifestyle products, as you can see, much more that are out there. Okay, gotcha. And can you tell us, maybe give us an example or two about when we talk about digital health, let's say, in the clinical setting, how does that work? So you talked about Fitbit, but then in terms of getting the data from the real world data and then processing that. Can you tell us a bit more about that? Yeah. So there we go down the entire medical device route. Obviously, I mean, even things like a Fitbit or even if you're just carrying around your iPhone, I mean, these things now are in Apple Watch. These things now have pretty good inbuilt sensors that can sense a lot of different things. I mean, some of the even the...

Apple watch stuff is medical grade, so can be used to, for example, you see how much blood oxygen you have and measure other stuff that is actually clinically relevant. The hard part right now using this in clinical trials quite often is the data capture and actually then making sense of the data because right now we have a little bit of a data overload. But if you define your clinical trials well, and you're using regulated device grade tools to measure real -world evidence, then it becomes really interesting because then I'm still hoping that at some point in the future, we're going to get rid of all those outdated endpoints that are just being used because they've always been used like six minute walk test. I mean, seriously, somebody walking for six minutes is a test of - I mean, it has its place. What with motion sensors, we just have so much more of an insight that we can generate than these very old measures that are just being used because they've always been used. So I have a really big belief in that, you know, digitally enhanced endpoints will play a really important role in clinical trial design. Today, we start to see this emerging in stuff like multiple sclerosis and other areas. So that's an interesting shift, right? So in clinical trials before with the pharma companies, there wasn't that much regarding digital health or devices, but that now is coming up quite a bit and is being integrated into the traditional clinical trials, right? Yes, absolutely. And also, you know, the hot topic of decentralized clinical trials, right, of making a little bit of the clinical trials a little bit more patient centric, a bit more patient friendly, and not making them come to a clinical research center, which can be miles or hundreds of miles away for some of them, right?

So obviously the ability of doing decentralized remote clinical trials is also enhanced by these technologies. That's great. And now these days we are hearing a lot about generative AI. That's all the buzz here. So I wanted to ask you more about how generative AI is used, but can you first tell us about what's happening in the world of generative AI and how that is different from the traditional AI?

Yeah, I love that we're now having a traditional AI. I mean, let's face it, right. I mean, the so -called traditional AI machine learning. Okay, it's been around for 40, 50 years if you go back into academia, but seriously, nobody cared until probably five to 10 years ago when machine learning just because of computing power and bigger computers was finally able to do something. So, I mean, what you can do with traditional machine learning is quite often predictive. You can look at trends.

You look at a lot of big data, you can predict what from those trends with cool stuff like neural networks is going to be coming out. Here I'm using a lot of Linus language, right? We don't want this to be a totally techy nerdy podcast, I believe. So this is what the traditional machine learning could do. What generative AI is now doing, is it can really just absorb so much data and reshuffle that information to generate something that hasn't been there before. All this is based on some work that's been done by the colleagues at Google called Transformers. We're not talking about the action figures, but actually, interestingly, quite complex mathematical models. But again, let's not go into mathematical models. What that means for you and me is that in reality, we can...

start speaking to a computer the way you and I are speaking now. And I believe that really is the revolution and they are part of the revolution. We can go through some other stuff that GenAI can do, but it is really giving us a completely new interface to interact with technology. Interestingly enough, at Novartis, we tested voice interfaces when they first came out with the likes of Siri and Alexa. We even tried in the clinical setting to use these things. Turns out, that they were just stupid. They could understand the words, but they couldn't understand the meaning, which means you had to hard code every single part of the dialogue. And that is just totally impossible. You will always end up with, I don't get what you're saying. And the fun part is now with GenAI, first of all, a computer most of the time understands and if you just correct them a little bit, they will definitely understand you. So it is giving us a totally different interface to interact with computers.

It is also giving us a totally different interface to interact with information because you just ask questions, you actually get answers. I mean, that was the point of Google all along, right? You're not Googling to get 500 links, you're Googling because you have a problem you're trying to solve. And so, GenAI can now do this. So that is part number one. Part number two of GenAI that I really find intriguing is its ability to just go through tons of data and structure it in a meaningful way. So, you can just let's say you have tons of PDFs. So let's face it right now, most patient records are data dumps of PDFs. So unstructured text data or unstructured image data. GenAI can now go through that and actually extract all the relevant information and make that meaningful so that we can do the advanced analytics that machine learning, etc. requires. So there is a really interesting part of using GenAI to extract data from data that was previously not very meaningful to either people or machines. Plus number three, you can actually then use it to come up with new ideas. And I really believe that is when we go to concrete use cases for pharma, we'll certainly find some like that. But those to me is really what sets Generative AI apart. And that is obviously since ChatGPT 3 .5 came out and was all of a sudden the fastest adopted technology ever seen...

has made us all reconsider how we use technology, I would say. Yeah. And we have been developing a lot of solutions for generative AI. And it's pretty interesting how much generative AI, I mean, these models can actually do without teaching them anything, right? And of course, we have to tailor them to the business solutions, the business questions, but it's amazing. I mean, like when you compare that especially to the traditional NLP, the natural language processing where you have to define every little thing and how the text gets converted into the query. Here, you have so much more freedom with generative AI, which is very powerful. So, what do you think in terms of, like, generative AI? It's being adopted now in the life sciences. Can you give us maybe some examples? Maybe let's start with something related to R&D. Can you think about something that is you know, up and coming and can illustrate how generative AI is going to change research and development. Yep. So, I mean, there's obviously the sky's the limit in terms of your creativity. We've seen a couple of use cases that are being explored. And let's not forget all of this is a one -year, 18 -month -old technology, right? So it's not that we have a lot of track record. We are back with the first iPhone. Remember that, right? So we will see exponential growth of these technologies over the next 10 years, but already in the first year, at least the ideas that you can think of, you can, as I mentioned, use this to just look at very complex data sets, which means, for example, you're trying to figure out, okay, what's my next indication I'm actually trying to develop against? And so you can just ask, to your favorite model of choice, okay, tell me, you know, compare these five different indications and tell me why one is more promising than the other and that will absorb all the kind of literature, all the kind of data, all the kind of papers that you actually have people doing this, you know, full time in pharma, just really figuring out, okay, in our R & D strategy, where do we go next? Once you have that, you know, GenAI becomes really interesting, particularly when it gets to larger molecules. So not the typical chemical molecules, but larger molecules. Because if you look at, I mean, how a protein is coded, that those are letters, right? So there's a lot of interesting approaches that just look at, okay, can we use GenAI to predict how a protein folds and how that is going to interact with the target? So tons of really interesting approaches there. Then finally, I would also say GenAI is going to be really interesting then when we go from R to the development side of things. It can help you write clinical protocols, right, because that is the generative part of things. So that is also, you know, a lot of people are doing that. So the productivity gains in here are just going to be amazing. So these are the three examples of what you can use, plus, obviously, if you use it in research, as I mentioned, you can use these large language models to really understand unstructured data. And so that is if if you take that and use data sets that previously were inaccessible, you can just go so much further.

Okay. And in terms of like, identifying targets or identifying biomarkers, do you see any development happening in that area? biomarkers clearly, yes. By the way, biomarkers, I wouldn't say, you know, you only need GenAI there. The good old machine learning is still actually very useful. I mean, let me give you one example of a company I'm actually working with. They're called Genialis based in Boston.

So what they're doing is biomarkers, at least the molecular biomarkers used in the past are mostly one genetic mutation or something like that, right? So one single genetic change. So we're looking at DNA and one single factor, HER2 positive, HER2 negative. That's what biomarkers used to be. Jenny Altshuisdager has taken machine learning to look at RNA instead. So we look at, instead of the genes that are either active or not, we don't know, we're looking at the transcript or we're looking at the actual RNA in the cell. And so that is just the amount of data that in the past we just could manage. But with machine learning, they were able to build much better biomarkers just in that. GenAI is clearly going to help us much more because you can look at a lot of genetic data and identify more things. But it's not only GenAI that is going to be helpful in biomarker design. Yeah. And you bring up a pretty good point. Right now that GenAI is here, people tend to forget that the AI, the machine learning for the prediction, that's still here. And is just solving more problems, but the traditional or analytical AI is still solving a lot of great problems. And the predictions and the machine learning is all there because the more data we have, we're going to need a lot of machine learning to analyze that data. Yeah. And the two approaches are going to be totally complementary.

And you know, I'm a big believer in problem first approaches and not technology first approaches. Whenever you see technology going wrong, it's typically because somebody is trying to use a certain technology for everything. And that's not the way it works, right? You have a problem and maybe GenAI is the right solution. But GenAI is one more tool set that we're having, an extremely powerful one. But it's not just because we have a new tool set that needs to be applied for every single problem statement.

Because that doesn't make sense. So the two are very complementary. Absolutely. And what do you think about AI in general to be used, let's say, in the clinical setting? How do you see that happening with the patients? And so are there some specific examples you can talk about there? Yeah. So first of all, there's another company I'm working with called Quant Health out of Israel, a really exciting company. They've recently raised $70 million in their series A.

So what they've done, they've incorporated just tons of data. So they have 350 million patients in there. They have, you know, several thousand compounds. And what they're able to do with all this, you know, huge network of data is they're able to predict clinical trial responses, both at the level of an individual patient, but even forecast if a trial is going to be successful or not. So, I mean, you can imagine how powerful that is in clinical trial design. They're obviously now working with a lot of big pharma and getting tons of attraction for that because clinical trial is one of the biggest investments that we're having in the value chain. But, you know, when we're talking about AI in the other parts of things, so A, you can use AI to do much better monitoring of patients. You can use AI, as I already mentioned, particularly GenAI for clinical trial design. But in the patient interaction, you know, you can now include chatbots that are finally useful and meaningful to interact with it and not just the pain that we're all used to from, you know, trying to use Alexa for anything else in the kitchen timer, right? So I believe that also the interaction of patients with technology is going to be totally different. And what about in the area of, let's say, like medical, like EMRs or like, electronic medical records or health records. What kind of changes do you see like happening in those because with GenAI or AI in general?

So I continue quoting some of the companies that I'm working with because I'm just so excited by what they do. There's a company out of Paris called Prax.ai And so what they do with GenAI is they get rid of the stupid EMR filling exercise that physicians now do. If you look at what HCPs now do, right, they spend less and less time with actual patients and more and more time in filling in these huge EMR machines, which are basically just nothing more than billing engines these days but are not really made for optimal clinical care. So what you can do right now is have GenAI listen to the conversation with the patient, prepare an automatic transcript, but not just a transcript that is old fashioned. You can actually automatically summarize the conversation into something meaningful that basically can be used as a dear doctor letter to hand over the patient to another department, but then autofill all the relevant medical parameters that you've just discussed into an EMR. So you basically reduce the time the poor HCP spends typing into computers by 90 % which means that the physician can finally focus on the patient again. So I mean, this is just the beginning of something much bigger. Obviously, we will be using GenAI for other tools like clinical trial, sorry, clinical diagnosis and other points. But already that simple fact of freeing up physician time for patient care is such a powerful thing that I believe we'll see that rolled out everywhere. And if you're not using that, you're just going to be really frustrated by the really hardships of going through EMRs and how to fill them in. And then the other way around, as I mentioned earlier, you can use these systems to finally make sense of a lot of data that is in EMR that has not been captured in a meaningful way for machine learning. So there is a...

really significant potential in there as well. Because let's face it, I've just heard the other day, this number don't quote me in it, but that 80 or 90 % of healthcare data is never actually being analyzed. I don't know the numbers exactly, but there was something like, okay, about one third of the data generated in the world is healthcare data. But only about 10 to 20 % of that is actually being used in any analytical purpose.

So that means there's so much untapped potential. If you just bring that some percentage points higher up, you can imagine the potential we are able to unleash. Now, see, when we think about life sciences, there is the biopharmaceutical companies that develop drugs. There is the medical drugs and devices companies. And then there is the health care companies. How do you see all of those as you have this wide view of these companies?

How do you see each of these industries adopting AI, generative AI or so? What kind of patterns are you seeing right now? So first of all, with these new tools, I really believe we are finally getting back to something that we have really forgotten about, which is the patient experience. If you've ever been sick at a hospital or had a relative going through that unfortunate experience, you're lost, right? You have absolutely no idea who to talk to. You're lucky if you can grab one doctor between 3 and 4 PM while they're having their coffee break. Otherwise they're always too busy. So you have no idea what's going on. You're totally lost, which means you don't know what's going on. You can now build patient portals that really can, which any AI component can just tell you what's going on, right? And always keep you informed. You know what? There's this thing coming next. So this is like, having your own kind of assistant to navigate you through the complexities of a hospital system or even ideally then having your own assistant, your own health coach walk you through the steps even before and after the hospital, making sure that you're getting handed over to, let's say, the right physician in primary care or in some settings, let's not forget, many markets don't have good primary care. One of the really interesting things we built, unfortunately, without GenAI at the time, it just wasn't ready yet. But it was our partnership in China between Novartis and Tencent where we identified that heart failure patients are getting pretty good care in Chinese hospitals because they are actually up to Western standards. But unfortunately, the primary care, so the continuous care once you're being discharged from the hospital, just wasn't there, which means patients were totally dropping off and being lost in nature. So what we built with Tencent there is an

AI nurse to really look after the patients after they left the hospital and kind of replace a non -existent primary care system. This was all built on the digital technology we had at the time three, four years ago, but now imagine you actually have a chatbot that can take you through all this. It's not just the rep that we could do, but now you can just have this natural language interface that we just head up to a point, but it's...

It was never a good experience, right? With NLP, as you mentioned, NLP just doesn't have the power. But GenAI has the power of really taking patients by the hand and talking to them like, you know, a good nurse would do. As long as we can obviously make sure that we are always medical grade, which is, you know, with one of the, we are probably going to get to the limitations of GenAI right now.

And one of them is we are still having problems with hallucinations, we're still having problems with traceability and reference ability, which obviously you really want to avoid. So we are not totally out of the woods just yet, but the potential is clearly there in terms of building a much better patient experience that is finally as good as if you had your own nurse and you said that this was looking after you, which currently right now the healthcare system cannot afford. And particularly in emerging markets where healthcare system are even much more overloaded than in our Western markets. This is huge potential. And here, I believe the term digital health that we started at the beginning, talking about this, if digital health is truly looking after you, after all your data and making sense of it, we are not able to build these things because of GenAI. And I don't think we were able to build these properly beforehand. I cheer. So in terms of what you're saying, like for the patient experience and patient care, the digital nurse or so. So before it was based on this NLP, so just like the natural language processing where someone types something, then that gets like translated into specific activity. Now, because of this generative AI, it can actually be, the chat bot can actually have the ability to think a bit. I mean, quote unquote, think to actually provide the right answers and provide the right context, right? So that's something that you think is going to be a game changer in digital health then. Yeah, I really believe so because right now, I mean, what interfaces did you have, right? The interfaces were mostly your mobile app. So some form of a nice scale, plus/ minus, sliders, you name it, right? So you could report on a scale of one to 10, I was feeling semi dizzy, semi pain, semi whatever, right? But...

you could never just talk to the system as you would do. And let's face it, most of the healthcare issues that we're having you are getting when you're 60, 70, 80, right? It's not the 20 year old digital native on TikTok that is having health issues. It's the 65 year old grandma. And for them, they've always had an issue interacting with technology. Less and less now, obviously, as we emerge into a more digitally native society, but...

voice and just interacting in natural language is the most intuitive thing we can do, right? And that is just going to be opening up so much more potential in terms of building proper care that is also engaging because seriously, nobody likes to enter numbers in their digital interface. You do it because you're forced to, but it's so much more natural just talking about these things. Like, Emma, how are you doing today? How's your blood pressure? Yo, that doesn't look good. Let's you know, this is the human touch that physical nurse would give you. And this is why you get so much better compliance. It's actually human talking to you. I believe we can get more and more into that territory. We will never be able to completely replace the emotional connection that we'll have with a human body. Although obviously, if you look at, you know, some of those movies where you have the smart digital assistants, maybe you're going to get there at some point.

But at least it's going to make the interaction with the machine so much more compelling and so much more easygoing and just a much more agreeable experience. Now, you mentioned the term their hallucinations. So for people who are not familiar with the large language models and hallucination, can you describe that a bit? Oh, yeah. So that is...

what large language fundamentals do, again, you know, here, not a scientific explanation, but it's, if you want to really simplify, it's a smarter autocorrect. So, you know, when you're typing something at the autocorrect suggested word, right? Most often it's right. Sometimes it's not. This is basically because all these big models do is try to predict the most likely answer. So when you ask, okay, what is the first dog in space? You're going to get the name of the dog.

Then you ask somebody, what was the first dog on the moon? The model has no address to spill out something and gives you the same name of the Russian dog that actually went into space, although we know that there were two. I know there was never a dog on the moon. So this is the so called hallucinations because the model is just trying to predict something on likelihood. Now, what does it mean? That means you cannot trust because it is not hard coded, you cannot trust any AI response, any GenAI response yet 100%. It's getting better, obviously, and you're putting in guardrails, but it is still the possibility that you're going to get wrong answers, which obviously you wouldn't get if you just simply hard code every single answer, which is what we used to do in the NLP days. And obviously that is helpful because you always want exactly the same recommendation when you ask a should I be or should I not be eating whatever bread 24 hours before the operation, right? You don't want the model to make up the mind every time based on your data, you want that to be based on the best evidence possible. So this is going to be one of the hurdles that NLP systems, sorry, GenAI systems need to overcome. But there is now more and more smart ways and there's really smart people working on these problems. So I believe these will be overcome in the next 12 to 24 months if I'm being optimistic. Yeah, great. I mean, we've been working on these issues, right? And the problem, the one problem is, you ask Chat GPT, give me some suggestions for vacation. And even if one or two of those are wrong, not a big deal. But if you are an AI nurse helping a patient and patient is asking for advice and you give something wrong, that will be disastrous. Right. So we really need to avoid any wrong information. Right. I know this is still, as you say, all of this is in infancy. So, how do you see some of these challenges going forward, especially with patient data, the privacy of that, the accuracy of that? How do you see the world moving in this direction with this technology, which is really cool, but there could be some bad side effects of that as well. Well, on one hand that I see regulation emerging, the European Union just published the AI Act.

Biden had this executive order. FDA is trying to figure out what exactly they're going to do with AI and there's working groups on that. We have the medical device regulation in Europe. So there is a lot of emerging regulations like that. Put GDPR on top of that, right? And you actually have to follow a lot of the regulations, which usually are hundreds of pages to follow. So first of all, actually, you'll need at some point your own AI to just help you navigate AI regulation. But...

all these things at least are put in place with good intentions. And so indeed, it is going to be very critical that we are not going to go in there with the Mark Zuckerberg motto of go fast and break things because if you you break things in healthcare, you're going to kill people eventually. And this is something that we will always keep in mind, which means that the adoptions of all these tools will be slower than in most of the other industries. But it is the right thing to do to make sure that, you know, we are not giving work recommendations. That said, let's also not forget that physicians are not always right, right? There are medical errors occurring all over the place. Do we want to wait until the technology is 100 % perfect? This is going to be the same debate that we're having around self -driving cars, right? Nobody accepts that the self -driving car kills somebody. We accept that drivers kill people all the time, at least as a society, right? So the question is indeed, where is the right balance between zero risk, and also not just taking away all the potential benefits that we've discussed. So that is going to be a really interesting ethical debate for next five to 10 years and how to navigate, you know, the guidelines, the regulations, but at the same time, not just killing the innovation potential that this technology really has for making much better health. Okay. Now, in terms of the, let's say AI nurse or any of these, you know, the AI that'll be used, is being used, and will be used in the life science setting, how much domain specific knowledge do you think that the AI needs to be trained on? I mean, if talking about GenAI or machine learning? So machine learning, we all know, the quality of the data sets right now is the limiting factor.

Which is why we've seen less successful machine learning adoption in healthcare than we were hoping for just because healthcare data is, as we discussed, messy. And machine learning requires pretty clean data, right? And if somebody just talks about, okay, this is metastatic breast cancer, and another file talks about whatever, MBRC, you know, how does a machine know that the two are the same things, right? GenAI can help you fix that, but most of the data right now is not clean. So machine learning is really, really needing very clean healthcare data. A lot of effort has gone on both in pharma, but also in healthcare systems of trying to fix that. Now, GenAI obviously is a little bit more agnostic, but you know, I mentioned Prax.ia, the company I'm working with, a big part, they're actually using open source models as the basis of what they're doing. But a lot of what they're doing is trying to really make sure that you're using the right model for the right use case. So it's not always GenAI. For example, when we're talking about data extraction, it can be a different model. And it is really going to be much more about using the right tool for the right use case. And a lot of that will be how do we really make models that are useful and don't hallucinate? Understand complex things like disease names, drug names, all these things. So there's a lot of the effort that will go on to that part of the training. And then finally, any technology is only as good as the user flow. So I believe AI on its own is never going to be...

super relevant if you don't include it into a workforce. Right now, we're all copy pasting our questions into Chatt copy pasting the answers back into our workflow. The Microsoft co -pilots of this world are trying to make that a little bit more smooth, but ideally, this AI needs to be 100 % smoothly integrated into whatever the workflow the patient or physician has. And that's also required for a lot of domain -specific knowledge. So that is then not the training of the data set, but how do you integrate these into complex EMR systems, right? That are usually what one guards that don't want to really integrate with others. How do you integrate with Salesforce? How do you integrate with Cerner? How do you integrate with the big guys of the world that also try to oversee, sell their own AI solutions and all this. So I believe domain specific knowledge is going to be super important for training data sets, for making GenAI really meaningful and also for embedding it into proper clinical workflows. Yeah.

And how do you see, as you said, right, this technology is still just booming over the last year, year and a half. And if you had the crystal ball, right? Like how do you see the adoption in the different healthcares? So like we talked about the healthcare, right? But you know, so pharma, in med tech, right? Like what do you see? Like we'll have great adoption next couple of years, five years, 10 years. Like what do you see as kind of like the horizon in which we will adopt this new technology.

So I obviously don't have a crystal ball. There's two things I can say here. One is adoption in healthcare is always slower than anywhere else. That is why still a lot of doctors are using fax machines, right? Just because the workflows in healthcare are totally outdated and there's a multitude of reasons for that. So that means new technology in healthcare is being adopted slower than anywhere else. Let's keep that in mind.

Now, we are obviously now, we can take two kind of frameworks in mind when we think about adoption here. Number one is your listeners may or may not have heard of the Gartner hype cycle, which means you know, there was initially a totally hype phase of adoption. Then we're going to go to the trough of disillusionment when people figure out, oh, you know what, this thing is actually harder than we thought. And then eventually we're going to get to adoption. I really feel still very much on the first wave. Or we are very much at the beginning of the hype cycle. I wouldn't be surprised if the next two or three years, particularly in healthcare is going to be, you know, the trauma of disillusionment. But we're going to get there. And the other famous saying is, I forgot who said it, but you know, technology adoption, wasn't Bill Gates could be. Technology adoption is he always overestimates the impact that the new technology is going to have in the first two, one or two years. And you totally underestimate the impact that the new technology will have in the next 10 years. I believe if you're doing this demonstration again in 2034, AI is a technology that will be in everything. Like right now, mobile is in everything. Like right now, cloud is in everything.

All these technologies are now a mainstay and we don't even think about them anymore. Nobody talks about mobile as a platform anymore because it's just the normal way of doing things. Nobody talks about cloud anymore because cloud is just the way you do computing. AI is going to be one of those tools that will be embedded into everything. The more it becomes portable with large language models eventually fitting on your iPhone. We all know Apple and Google are working on that, of just putting an LLM right on your phone.

So that you always have it with you. These things will be embedded into everything and you wouldn't even know that they are there anymore in 10 years, but they're just going to be natural. But it's going to take five to 10 years for proper adoption because of healthcare being such a regulated, slow industry. And also because these things are always harder than people think initially. Absolutely. And that's a great point about the Gartner hype cycle, right? We haven't talked about that in the podcast, but you go through the hype, then...

there's a trough of, as I said, like disillusionment, right? And then you have like the sustained growth where the value actually comes and gets realized. And this is a great point, Christian, about talking about that. But also really appreciate you giving an overview of generative AI and digital health and the AI itself and how that is being adopted and how it will get adopted in health care. So Christian Hein, executive advisor and board member in HealthTech and AI. Christian, thank you for your time today. Thank you, Amar, for having me. It's been a pleasure.

Amar, that was really interesting. What did you think? There was a lot of great insights from Christian about digital health, about AI, and about both the two aspects of that, the generative AI, as well as the machine learning, how that's being used. I mean, it's getting adopted in a pretty fast way in health care, but there's a lot to be done, right? But then also, there's a lot of hype right now that we're going through this hype cycle. So we're going to go through some realizations about the value of this. And it's going to be pretty interesting. So there are a lot of really great insights that you provided today. You talked a lot about generative AI. I think people may be familiar with generative AI for creating images or video or in life science terms, drugs. One of the interesting things you talked about though is using it to take unstructured data and structuring it. How might this change what's possible? That I believe is going to be a game changer. Because until now, when we think about analytics, or even machine learning, which is a statistical analysis of data, we always think about like a spreadsheet, right? Like there's a tabular data. There's data where you get samples, you get different variables, and then you do some statistical analysis on that to get some results about some prediction.

It's been very difficult to do this kind of analysis on text data or image data, which is not structured. I mean, that's what we call unstructured data because it's not structured nicely in a table or in a database. So far, there have been a lot of attempts, especially with NLP, as we talked about the natural language processing, to do that, but they have been limited. I mean, like the kind of technology that's needed to really to convert unstructured data into structured data into the database or tables so that analysis can be done on them or even, and not just, it doesn't have to be like fancy statistical analysis, even just visualization of that to see how, you know, what things are where. That hasn't been done that much so far. So, generative AI is going to be able to do that. So, I believe that now we don't have to think about unstructured data, or structured data much differently. On unstructured data, we apply the right generative AI on that to convert it into structured data. And then we're able to do the kind of analysis that we're able to do on structured data on unstructured data as well. So I think that's going to be really helpful, especially when we're talking about the documents, right? Documents, websites, right? Where there's the free text or images. So how to be able to really structure all of that.

There are so many use cases we can think about around that, about regarding the patient records, regarding the clinical trial documents that get written, regarding the promotional material that gets written, even like scientific publications that are in. So all of that now can be converted into structured data. We spend a lot of time thinking about the impact of generative AI and drug development and biopharma more broadly. I think Christian offered some thoughts on...

how it might change the way physicians work and the patient experience. Where do you see the biggest impact of generative AI? I would say it is across the board in the life sciences. Definitely in the patient experience, in the health care settings, where especially when the patients are not getting enough care, the generative AI can definitely help in that area.

But I would also say that when you look at the value chain of healthcare and pharma, start with the research. The generative AI is definitely helping in creating new molecules. I mean, Christian referred to creating new biological molecules. There's also generative chemistry where creating new small molecules. So that is going to help us create drugs faster in terms of development, getting a lot of new aspects of our biomarkers, even making a lot of the processes and report generation faster. That's going to accelerate the drug development and also on the commercial and medical side, just getting the right medical education out there and then really personalizing the experience for the healthcare professionals and patients. So I just see the impact across the board and I think it's going to be a significant impact across the board for this. One of the things you asked him about was the extent that domain expertise is needed. And he distinguished between machine learning and generative AI. What's the potential to use generative AI to address this challenge with machine learning? See, the machine learning, as he rightly said, right? You just need a lot of clean data for machine learning to do its job. And there also, that is its requirement, but also, there is the domain knowledge that's required to know exactly what kind of analysis needs to be done, right? Because there's a scientific question or a business question that you're trying to answer with that data using machine learning. The way you do the analysis to answer that, that's pretty much dependent on how you need the domain knowledge. Now, in terms of generative AI, the domain knowledge for that is because...

Let's see, if you are talking about, let's say, given the example of this, the AI nurse, right, we can take an example of anything, like generative chemistry or so. Inherently, you need to know about what are the different terms regarding the patients? What are the kind of diseases? What are the kind of symptoms that could be? What are the kind of answers you can give around that? What could be the potential treatment for different areas, right?

There's a lot of that domain knowledge that needs to be given there. So if you, I mean, right now, if you ask like, you know, ChatGPTs of the world, they have some generic knowledge about this. They're not experts. Like you can't get expert advice on them, especially in the healthcare setting. So that's where the domain knowledge is needed. So domain knowledge is going to be needed quite a bit. And that's what we're seeing as we're developing a lot of these generative AI solutions, that domain knowledge is needed to make these really powerful solutions. And so that's why I was asking him about that, right? So of course, domain knowledge is needed in both of these areas, maybe in a bit different ways, but you do need it in both of those. Well, before we end today, I did want to ask you about some recent work you've done. I know you just authored a white paper for Agilisium about a strategic framework for generative AI and use cases in life sciences. What's the purpose of the white paper? And who's it for? So the  reason I wrote that was that as we are working a lot in generative AI, everyone seems to have a hundred use cases and no one seems to have a roadmap about how to go about that. And also there's not a real understanding about what are the use cases that can be solved quickly? What are the ones that cannot be solved quickly? So what is really the feasibility around that? What is the kind of outputs we're going to get from those? And also,

What is the impact? Like some of these, the impact may be a little, for some of the others, the value that they're generating is much larger. So as working with this, I actually came up with a framework to think about, like to make sense of all of this, right? About all the different types of use cases. And what I noticed was that even though there are some frameworks that are out there generically for generative AI, and there are some papers out about life sciences as well, there was no kind of a paper or so that talked about structured framework specifically for generative AI, specifically in life sciences. And so that's why I decided to write that too, so that people get the understanding of how to do that. And so that's why I discussed the different functional areas in the life sciences, but also the outcome based like categorization. So I actually categorize the use cases according to different outcomes. So,

insight generation versus content review versus content generation. So there are different types of outputs that the use cases have, right? So people need to think about that. And then people need to think about what is the value that each of those different types of use cases bring. What is the feasibility around that? Some are easy to do, some are hard to do. Some have a decent amount of value, but some have huge amount of value, right?

So when people are, as companies are thinking about creating their own roadmap for generative AI, I think it will be very helpful for them to really go through that so that they get a really good sense about, well, how do we prioritize the use case? How do we sequence the use case? So that's the idea. So I would say in terms of who that's for, it is for any industry executives in life sciences who are really interested in developing generative AI framework.

Any analysts or even business leaders or even folks who are in compliance or regulation, right? Like I think it is relevant for anyone related to life sciences or even investors who want to get into life science. So I know there are a lot of genetic companies that are coming up, right? So it gives them a sense of, okay, well, what kind of use cases fit where? So for any of these, as I said, like executives, business leaders, analysts, analytics leaders, investors, it is useful for any of those people who...

who are associated with life sciences and who are interested in generative AI. I've also written it in such a way so that even people outside of the industry, I think they can get a lot of value out of that as well. Because even though this is specific to life science industry, there are many aspects of that that can be applied outside of the industry as well. The paper is available for free. If you're interested in downloading and copying, you can go to Agilisium .com and find the white paper by going to the Insights tab.

and selecting white papers. Amar, thanks so much. It was a great discussion today. Looking forward to our next show. Thank you, Danny.

Thanks again to our sponsor, Agilisium Labs. Life Sciences DNA is a bi -monthly podcast produced by the Levine Media Group with production support from Fullview Media. Be sure to follow us on your preferred podcast platform. Music for this podcast is provided courtesy of the Jonah Levine Collective. We'd love to hear from you. Pop us a note at dannyatlevinemediagroup .com. For Life Sciences DNA and Dr. Amar Drawid

I'm Daniel Levine. Thanks for joining us.

‍