Pharma Sessions
Pharma Sessions
Where AI Delivers Real Value in Pharma Today with Emily Lewis
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Pharma Sessions is hosted by Jonathan Kaskey
Follow along on LinkedIn: https://www.linkedin.com/in/jonkaskey/
Or join the Pharma Sessions Substack:
www.pharmasessions.substack.com
The biggest impact the soonest is going to be, in my personal opinion, in the regulatory space, because document generation takes a lot of effort and is probably the lowest hanging fruit. So we have existing documents for many, many years already in digital form. And we have ways to scan those documents, even if they're like handwritten, for example, that can digitize those. But we can then, and we have templates for how submissions are supposed to be made. And so it's just such a no-brainer for me that, you know, we can use the existing documents on a specific disease state, on a specific clinical trials that have happened to create submissions, whether it's INDs, whether it's clinical study reports, to official documents for regulators and for payers as well.
SPEAKER_00Hello, hello, and welcome to Pharma Sessions, a place for pharmaceutical leaders to come and learn from each other. I'm your host, Jonathan Kaske. Technology and market trends are bringing change at an ever accelerating rate, and no person, team, or company can afford to be left behind. Here, we dive into the strategies and tactics that our guests use to tackle these challenges and create new opportunities, and how you can do the same in your own organization. On today's episode of Pharma Sessions, I'm thrilled to welcome Emily Lewis, an AI and innovation leader at UCB and one of the most forward-thinking voices in the evolution of healthcare technology. With nearly two decades of experience across ClinOps, digital health, patient engagement, and AI strategy, Emily has built a career helping organizations rethink how data, machine learning, and human-centered design can transform patient care. Emily is also a global educator, advisor, and advocate, teaching AI and pharma, advising on ethical AI, mentoring future innovators, and contributing to thought leadership across organizations like Harvard Business Review, HitLab, and Women Tech Makers. Did I get all that right, Emily? Welcome to the show.
SPEAKER_01You did. Thanks so much for having me. It's such a pleasure being here.
SPEAKER_00Absolutely. Thank you. So, all right, before we get far into it, I always like to start with a little icebreaker. And you're out in the Denver area, correct? I am. Okay, perfect. So it is maybe 11:30 your time. What have you had to eat so far today?
SPEAKER_01Well, I have two small children, so I barely get anything to eat. It's usually whatever they don't eat in the morning. So if I'm perfectly honest. So it's been coffee. I tried several things this morning with my kid. I tried um, well, my small one, my 11-month-old, we tried infant cereal, which he's over because he wants solids now. And then we tried egg bites, which if in the air fryer are very good when they crisp up. So I had leftover egg bite from my son, if I'm being completely honest.
SPEAKER_00All right. Whenever I ask that question, the breakfast foods are all over the map, but everybody has coffee. So that's the apparently the universal in pharma. So let's just dive right in. And as you gathered from the intro, there's probably a lot around AI here. But you've talked before about separating hype versus potential around large language models, particularly in healthcare. So where do you see LLMs falling short? And what other forms of AI, symbolic reasoning, et cetera, do you think are essential for moving towards a more reliable AI-empowered system in medicine?
SPEAKER_01Yeah. So we've seen for a few years now that LLMs are great for predicting the next word, right? They're totally probabilistic. They've been trained on a massive amount of text data. So they're great at what words might be close to this word in meaning. They don't have a true understanding of the world, right? They don't have, they just know patterns, patterns in words. And so that fundamentally we've seen, especially we've seen this now when Google rolled out their AI overviews. It leaves a lot out, right? Because you're just summarizing what is on the internet. You're not necessarily, you don't have an educated perspective on what is really true in the world. You're just aggregating and synthesizing and summarizing. So LLMs really fall short in knowing like what is true, true. I mean, they do a pretty good job of making a pretty educated guess, but they don't have a real world view. And so in medicine, that's especially important because they can do really well on tests, these medical exams we've seen that published in theory. They know a lot about medicine, but they can't actually help us practice medicine because they don't understand physiology and biology. And so by nature of that, they're just not, they're a great tool for conversational uh purposes and for summarizing and for helping humans in the loop really reason through things, but they're not the end all be-all, right? And so we need a stack of different types of AI to really move forward. And we're seeing really interesting developments from Jan Lacoon out of he wasn't Meta. He's now creating his own company with world models with Alex from Nobla. So, you know, they're starting, they're really most forward thinkers in this industry of AI are really understanding like Silicon Valley, it's what they're doing is great, but they've kind of got it all wrong. They need to really move past large language models and traditional machine learning and into what's called world models, which are really how do you think about how the world actually works? So, sort of similar to reinforcement learning, where there's a reward for doing things in the real world. World models really explore actual biology and the world itself by living in it and getting actual reinforcement from the context with that they're in, like the ecosystem they're they're within. So large language models are great for the interface with users, but they need to be stacked with symbolic reasoning models which give you actual context on like hard and fast rules in medicine. So in our world, for example, in medicine, don't give X drug to Y patient if creatinine is at a certain level. Or if patient is pregnant, for example, you know, this drug is contraindicated. There are certain hard and fast rules that we need to program in with RAG, for example, so retrieval augmented generation. But there also needs to be more discovery work. The unknown unknown is really what world models will help with, because LLMs really regurgitate what we already know in the world, but don't really cover that unknown portion. So I think we need a kind of a layer cake stack of all three types of reasoning and understanding of the real world for products of the future.
SPEAKER_00And if you're working in a pharma organization, let's say you're part of almost any business unit, whether it's commercial or medical or clinical, how advanced should the person on the business side be trying to get with their understanding of this situation versus how much do you think should be more reliant on like a center of excellence, right? Like everybody can't be a complete expert on all the various AI model types. But what's been really interesting. So I spent a good deal of time working with on like a rag solution in the past. And it was shocking to me how many people on the business side that are in the documentation standpoint actually knew, understood RAG and could explain it very clearly. So, like, where do you recommend people draw that line?
SPEAKER_01I recommend there's no line drawn at all. I think everybody needs to upskill in AI. I think no matter where you are, you need to cross-train in the modern world. You have your specificity, your vertical, whether you're regulatory, legal, compliance, but you also need to understand how the modern world works. And AI is a part of that now. And so you don't need to know the details of exactly how things work under the hood, but you need to know some general things about how AI, what rag is. I think the people who don't upskill themselves will slowly find themselves behind and not in the important conversations that make you useful within the modern day company. So I really advocate across the board for everyone to upskill themselves. And that's why digital transformation within a business is so important. And I think the people who are really cross-trained and have a systems thinking approach to really learning about these technologies will be at the forefront of really making positive changes and having better control over what the future of the business is and really better foresight into being competitive in the business marketplace.
SPEAKER_00I'm assuming that you're talking with your colleagues at across the industry, right? So are you seeing, so you're at a large pharma, right? Are you seeing large pharma proactively taking this stance and providing kind of general upskill training to people?
SPEAKER_01I think people are providing, you know, LinkedIn learning courses. There's some general upskilling uh tools, but I think what we're not doing enough is actual hands-on use of the tools. So they'll roll roll out things like Loop, which are a new, you know, the Microsoft's version of Notion, and they do a couple trainings, but then if people don't use it, they don't use it, right? It's their prerogative not to engage in AI-enabled software. But there really needs to be a pressure within the company to use these new tools that are rolling out there that are AI enabled. We need to be using AI-enabled CRMs, fair and sales. Um, that is how people are working in the modern day. And so we just need to make sure that we are doing more than just rolling out a static training, that we are actually doing hands-on training and oversight and have it be a part of people's performance development plan that might might be controversial, but I think the the folks that are really at selling within their own positions and and leading are really um fluent in these kinds of applications of of AI.
SPEAKER_00Yeah, I think so I can share from my own experience at a much smaller company, right? So I lead a sales and marketing team here. And where we've seen people almost organically start to upscale is when they can see it making their own jobs easier, like for lack of a better term. So one of the things that happens all the time at tech companies is product goes over and talks to business development and says, What are you hearing in the market? What do people want that we don't have? What's resonating, et cetera? And traditionally, you're like getting together, it's a phone call, it's an email, it's a it's a Zoom, whatever, and you're coming up with like a PowerPoint list, and that becomes your product, product roadmap, right? What we've started doing, and this is something that one of the people on the team just did because he he just did it, right? And it was great, is he went, I think the tool is called lovable. It's almost like a Figma. You can, as a layperson, create wireframes that are way more than wireframes. It's almost like an app that doesn't really work, but it it gets the point across and it has helped us get the feet. It what he saw out of that was one, it was easier. There wasn't a lot of back and forth. And two, because product understood it better, his requests all of a sudden were like at the top of the priority list. And so that is a prime example of like organic growth and somebody seeing seeing the benefit in a way that will, you know, he he's continuing to do that right now. Others on the team have have picked it up. I just thought it was so fascinating because it's like it's kind of what you're talking about, right? Where it's one thing to provide training on it, but it's another to say, okay, now go do something.
SPEAKER_01Because you really have to have that internal motivation, right? And you have to see those results. And so if you're not actually playing with these tools, the number one recommendation I can make is like carve out some time in your day to just play around with these tools, right? Make mistakes in them, understand how they work. Where are you running into? And then if you run into a problem, just YouTube it. I tell my mom you run into a problem on how to fix something in your coffee machine, just YouTube it.
SPEAKER_00Well, it also helps when you do run into problems, if you can maybe try to figure out why, not figure out from a technical standpoint, but just at least a basic understanding of like where it fell down. What you're talking about with the LLM, sometimes it's they're not pointed at your specific data sets that are very bounded and controlled. It's pointed at the whole internet. So if you're saying, hey, what should my sales rep, pharma rep, be doing today? They're not actually looking at the call data and the prescribing data and blah, blah, blah, blah, blah. They're just kind of almost like uh looking at a Google search and pulling a bunch of random stuff from wherever.
SPEAKER_01Yeah, which is why they have enterprise licenses for these tools now. So there's enterprise ChatGPT and Microsoft Copilot, so that it is very specifically trained on your business's data. So you're not training those external algorithms as well.
SPEAKER_00If people experimented with it when all this stuff sort of first hit the news two or three years ago, it's good to go back every now and then.
SPEAKER_01See how far we've come.
SPEAKER_00Yeah, because I remember when there was so much hype, I was planning a trip and I was like, make me an itinerary, and I was trying to go to Iceland. Like, I want a restaurant every night and I want to see these, these, and these places, and I don't want to drive more than a certain amount of time. And it it totally failed. It was like had me going all over the country and zigzags and whatnot. It's different now. Like it just it just is, and you just kind of have to keep at it a little bit and figure out how to use the thing.
SPEAKER_01I mean, I really recommend using it in your personal life, right? Right, because that's where you can really make mistakes and really you can test it in different ways than just your business purposes.
SPEAKER_00Like trying things out in your personal life, maybe if you think a bit more systemically, right, where there are all these other industries out there that have less regulations and, you know, honestly, less consequences about getting things wrong. What lesson should pharma be looking at from other industries? And how can you sort of take that in and use that in your own work?
SPEAKER_01I think the biggest parallel we can draw is to the world of aviation. We've seen autopilot be rolled out within the field for quite a while now, and pilots trust it. And in so do you um consumers, right? We walk on an airplane and we don't think twice. And so that level of trust has been built over time, right? Through governance and through regulation and through testing, rigorous testing and validation. And so I think we need to bring those same principles to the world of medicine and drug development. It's been done before. And so it's a little alarming to me to see the current administration pull back some of that regulation most recently, then AI and wearable technology and say that it's not needed. To me, it is needed. To me, people need clear guardrails on where to operate and how to operate. And if they're not, then they're worried about being liable, right?
SPEAKER_00And causing real What was the regulation you're talking to about? Um I'm not familiar with that.
SPEAKER_01The FDA recently came out with some guidance saying that they're loosening the regulatory reins on AI products in general if they don't meet a certain bar. Like they're just not going to go out and audit these products if they're not very strictly software as a medical device. If they don't provide clinical decision support, they're not going to look at them, which is very interesting because they recently gave a warning letter to Whoop, which, you know, was in that gray area. So it seems like even within the administration, there's they're going back and forth through about things. And so it's hard to take this administration at their word about what they really intend here. But it is scary when they are talking about kind of having some lax oversight for AI tools when technology is so new.
SPEAKER_00So one of the other risks that you had talked about or kind of around this idea of consumerization is I guess the risk of desensitation of in some ways it's great to learn from aviation. Well, actually, maybe aviation is a life and death thing, but some other industries aren't, right? And so we are seeing patients as consumers, of course, the the ultimate user, literal consumers of the drugs, but their experience, their feelings, the overall impact to their life needs to be taken into account. How does this all factor together? When we talk about consumerization, what does it practically mean for clinical operations, for patient engagement, or even the ways we design interfaces between patients and technology?
SPEAKER_01When we typically hear the word consumerization, we think Amazon making the user face pretty clear and easy. And I would say, yes, that's exactly what we're intending to do in healthcare. We want to reduce friction specifically for patients using products. We want to not necessarily have an easy button, but really give them a sense of agency in the process. So they're well educated about what's happening with them or what might be happening with them. They feel confident about talking to their clinician about it. We're kind of mitigating this white coat syndrome with them because they feel well equipped to have the conversation as the expert in their own disease or condition or whatever's going on with them. So it's about empowerment, it's about agency and reducing kind of cognitive burden on them. So giving them the tools to really synthesize what's going on with them in an easier way to have better conversations with the experts and have those documented and able to refer back to them and have the data really work for them as well as the clinician in these interchanges.
SPEAKER_00If you've been enjoying the conversations here on Pharma Sessions, you should know they're made possible by the team at Xunt. XSunt helps life sciences companies turn complex data into clear, actionable insight. For years, XSunt has made complicated data sets simple to help commercial, medical, and operations teams map what's happening, predict what's next, and make stronger decisions faster. And now there's an added AI layer that makes everything work so much better. I was actually pretty jaded about some of the AI approaches I'd seen, but when XSunt showed me theirs, I actually left my job to come work for them. It's really awesome. So if you want to understand your market, your customers, or your performance with more clarity than ever, check out xunt.com. That's xsunt.com. All right, let's jump back into the episode. Is there an example you could share either your own or somebody else where you've seen this done really well?
SPEAKER_01Yes. So there are companies out there. One of the products that's being worked on across the board are kind of AI advocates for patients. So we've seen LLMs be applied within the electronic health record to help clinicians come up with clinical questions, to summarize the medical history, to communicate with between a specialist and a primary care provider during like an e-consult. But now we're starting to see AI applied to patient interactions. And so, how can we use LLMs to really make the verbosity less of the, you know, the exchange and also within their own level of understanding? So maybe a fifth grade reading level. So translating not necessarily the language of, you know, from English to Spanish, but translating the clinical language to a more understandable lay language.
SPEAKER_00Right. Take a protocol, make it an ICF, and like something that can the average person can understand and comprehend. You know what's funny, Emily? So I was working on this type of project a while ago, maybe a year or two back when I was at a different company, and there was a lot of hesitancy from the particular pharma that we were dealing with based around compliance. And what they found was that nurses at the trial sites were already doing it. They were just doing it in a fairly unprotected way by taking the protocol document and popping it into ChatGPT.
SPEAKER_01And that's exactly why OpenAI just came out with ChatGPT health, is because they know this is 80% of the use cases they've determined. 70 to 80% of the use cases with within general use chat GPT are healthcare related.
SPEAKER_00Really? 70 to 80 percent. That's crazy. That's so big. I say that, although like an hour ago, I was using it to try to diagnose my cat with whatever is is wrong with him. And uh totally understandable. Because I think it also, this is just my own personal theory, but it kind of points to some of the issues that we have, like with our healthcare system in general, where it's like every time you talk to a doctor, one, it's hard to get an appointment, two, the catch register rings. And a lot of times people feel like you have such limited time with providers that they don't actually, they're not actually able to really spend time digging down and getting into it. And so even like recognizing all the deficiencies of the LLM that you talked about at the beginning, people, me, like I'm still going to it uh because it seems like the best available option.
SPEAKER_01And it's gonna help rebalance this power dynamic that's always been there. That it's doctors have always been paternalistic. But now, with you know, by equipping patients with better access to their own information, to what clinical guidelines are, they're really gonna be empowered to have more educated discussions with their clinicians and advocate for themselves to say, you know, my treatment's not working for me. I hear there are other treatments on the market and they work in X ways, you know, and they can really understand, have a better understanding of how the other treatments work and say, you know, I really want to switch. What do you think, Doc? So I think we're gonna really see, I think clinicians are probably already seeing this since it's been a few years that ChatGPT has been out, but that dynamic is gonna continue to be more equal between the patient and the clinician going forward.
SPEAKER_00That's very interesting. Right. So if you were to look at all the different areas that pharma touches, which is a lot of, you know, the patient, the regulatory, the payer, every commercial. I feel like this is a layup for you, but like which ones do you think are are very likely to be impacted by AI? Or is it everything? Is there is there some type of ranking that you can put as to where you think that you're gonna see the biggest impact the soonest?
SPEAKER_01The biggest impact the soonest is going to be, in my personal opinion, in the regulatory space, because document generation takes a lot of effort and is probably the lowest hanging fruit. So we have existing documents for many, many years already in digital form. And we have ways to scan those documents, even if they're like handwritten, for example. Um, that can digitize those. But we can then and we have templates for you know how submissions are supposed to be made. And so it's just such a no-brainer for me that you know we can use the existing documents on a specific disease state, on a specific the clinical trials that have happened to create submissions, whether it's INDs, whether it's clinical study reports, to official documents for for regulators and for payers as well, to put the data that we've generated in the way they want to see it. So we have many different stakeholders, right? And so not only is it document generation, but really tailoring the content to the specific user of that content.
SPEAKER_00Yeah, it's interesting. I saw somebody at conference a while back talking about imagining a world where not only were submissions electronic and kind of very AI enabled, but also review, because that's an area where both sides are so structured. And this idea of working within structured data, it seems like such a natural starting place because there's so much less unknown.
SPEAKER_01And it's so easy for things to become inconsistent, right? Like one document says this, but another says that. And so if there's a through layer, a thread of consistency where AI is looking at all the documents together, they can point out whether there is inconsistency and make sure that it's rectified and brought to the human in the loop who fixes it. So ultimately, you know, humans will always be responsible for these AI submissions. It's not the vendor's responsibility, it is your responsibility as a sponsor company. Um, and so I think that's one point I always try to drive home is you know, we can't offload the responsibility for this to anyone else but ourselves. So even though AI is doing the work, we are still very much accountable and responsible for making sure that we have oversight and we are still very much in looking at what it's created to correct things or and provide that human oversight.
SPEAKER_00All right. This is a quick tangent, but I'm like this close to coming and proposing something similar to my little local government here in Rehoboth Beach because my my wife is in the process of opening up a yarn store. And like, I swear to God, the submissions that we are requiring in the city of Rehoboth are like FDA level for a retail shop. Like, we this there's a form, it's structure. We could we could do this for review. Like, this doesn't need to take six weeks. This could take six minutes.
SPEAKER_01Yeah, you can systematize it right across industries, any kind of regulatory paperwork.
SPEAKER_00We'll start that business after the call.
SPEAKER_01Sounds good.
SPEAKER_00All right, so let's bring it back to just kind of a couple of final closing comments. But looking at your work in education, mentorship, AI advocacy. Again, the audience of this podcast is always pharma people, right? What capabilities do you think, if somebody's trying to advance their career, right? What do you think they should be focused on to develop right now so that they can be a part of this coming workflow?
SPEAKER_01So I hinted at it earlier. I think we need people who can look at workflows and processes and have that systems level thinking perspective because we can't simply digitally optimize what we've already been doing. AI is presenting this new opportunity to do things completely differently, right? And it can offload administrative work and free up humans to do more complex thinking and orchestrating. And so I think what we need to do is have folks who are ready to learn and who recognize they don't know it all, but are very much every day trying to keep abreast of the new things that are rolling out because the productization of this is happening at a dizzying speed. So you really need to have the right sources of information to stay up to date. But you also need folks who are just curious, right? Who are willing to really dig in to use the products, to understand what's out there, talk to the right people, have the right connections. It's no longer okay to just stay in your own lane, in your own vertical, and just be the regulatory person. You need to be somewhat savvy of how technology is really changing every aspect of our lives, including the work we do. So I think we need people who can be cross-functional in nature, who understand product, who understand software development, who understand cybersecurity, who understand patient engagement and clinicians and medicine itself and the whole gamut, right? We need to understand the whole industry we're in and how it's being impacted by technology every single day. It is a moving target. It is never static. And so you really need to be on your toes rather than on your heels by staying informed, by doing your own research, by being curious. So listening to podcasts, what I do is just drop a bunch of documents into notebook LLM and listen to my own personally generated podcast on my way to pick up my daughter from daycare.
SPEAKER_00You know, if I'm prepping for something, I if you're trying to keep up on some news or something, you've got several sources that you're popping into notebook and saying, give me a 15-minute podcast of this that I can listen to on the drive.
SPEAKER_01So I don't just have my NEJM AI grand rounds podcast. I have my own personally generated podcasts that are curated based upon the data and information that I want to learn about. If there's publications that have come out in nature, for example, I'm popping those in. I just generally want to learn more every day because I am painfully aware that everybody else is also upskilling. And so to be competitive in this new present day, you really need to be cross-functionally trained, not only in your, you know, traditional area. For me, it's clinical development. I've done work in clinical trials for almost 20 years, but that's not enough anymore. You really need to understand how the work is changing and evolving every day.
SPEAKER_00I love that example too, because it's almost like they talk about in during the training surgeons, the like see one, do one, teach one. Like you are actually seeing how other people are do are using this. Okay, that's some level of understanding. And then you're solving one of your own problems with this approach. And I'm sure there's many examples of that that you're having, and you're teaching other people how to do it. So it's really, really smart.
SPEAKER_01And that's exactly how I got my blog going, right? I was just genuinely curious about all the things that are happening in our industry. And so I just started writing about them, researching them and writing about them, which helped me in turn learn more about these topics. And so I'm taking in this information in many different ways, right? The podcasts, the LinkedIn blogs through conferences, the old traditional ways, but I'm synthesizing my own information with AI and with exposure in many different ways to really fully understand what's happening and grasp it and start thinking about the future, right? Like I not only follow the pharma industry, but I follow the social media industry. Like I said, Jan Lacoon and Meta and open sourcing and everything that's happening with open sourcing things on GitHub and making the developments available to researchers everywhere. I think that's so important and something I feel really strongly about is not only keeping this stuff for ourselves, but the power of AI is to amalgamate data from so many sources. And so I really applaud Lily's effort lately with um Toon Lab to create an ML, to create a platform in which people can contribute their own data and have one master model. And we're we're not all in our silos building our own models, but contributing to one where we can all benefit.
SPEAKER_00Very cool. The general takeaway is stay curious and be afraid to try, right? And in fact, be afraid if you're not trying because somebody else probably is.
SPEAKER_01Unfortunately, that's the case. You know, I have my influencers that I follow on Instagram who keep me up to date with what how they're using Claude skills. And it's very obvious I'm not a coder by training, but I can, you know, create a GPT and ChatGPT and have reusable skills there for the repeatable tasks that I use every day.
SPEAKER_00100%. It's not that hard to get started. Like let's just let's just say that.
SPEAKER_01Like it's or for somebody who's a millennial, right? It's a little harder for my mom's generation.
SPEAKER_00I'm about the oldest millennial, so maybe.
SPEAKER_01I mean, I still remember phone books and catalogs to order your clothes, but I am basically a digital native, right? And so I have both business experience and technical understood the advent of the internet and what it was like before that. And so I think you and I, our generation is probably at the greatest advantage because the fact that we had that business experience and the um were digital natives.
SPEAKER_00Well, there's also something about realizing that it's not set in stone, right? Like if the worst thing that happens when you create a GPT is it doesn't work.
SPEAKER_01You you tweak it. You say, All right, stop using m-dashes. Yeah.
SPEAKER_00You're right. Oh, that was that was in day one. No more bold than M dashes, please, for the love of God. Be human. Yeah. Excellent. Emily, thank you so much. I I really enjoyed this.
SPEAKER_01Yeah, this was a lovely conversation. Thanks for inviting me.
SPEAKER_00Take care. And that's a wrap on today's episode of Pharma Sessions with me, Jonathan Kaske. If you enjoyed today's conversation, don't forget to hit follow or subscribe and share it with someone else in the pharma world who might need to hear it. For more on pharma trends, career growth, and business strategies, connect with me, Jonathan Kaske, on LinkedIn. Until next time, thanks for listening.