Can AI Repair the Pharmaceutical Industry? with Wendy Rockwell

0:00 Can AI fix what's broken inside of the pharmaceutical industry? Let's find out with today's amazing special guest, Wendy Rockwell. Now, Wendy, I'm so excited to have you here because one of the big things that really interests me and one of the big challenges we face a lot in the AI industry is telling like the real from the fake. The Things that have like real AI components and the things that are kind of mimicking it or that are just chat GBT rappers.

1:22 And I can't think of any industry where this is more important than When you're kind of looking at these tools or looking at these vendors, what are some of the thoughts that go through your mind or some of your process to go? There's something that doesn't seem quite right here. Like how do you detect those things? Because that's very challenging for a lot of people. ~ you know, and I'd probably bring it back to the context of where I am in the pharmaceutical industry.

1:43 I work with health plans. So I spend predominant amount of my time looking at payer policies, how that lines up with patients that are trying to get specialty drugs, oncology drugs where they don't meet the qualifications, they get a rejection, there's an appeal. You know, some of those sort of different things that go on in the industry and I use it to understand where the payer's coming from first and to really look at the policies and then to look at the patient's pathway of getting access.

2:13 So those are the some of the realities that I have and in going forward. And I'm just want to make sure that I'm aligned to the context that you're talking to as well. I I happen to be in a startup where we are doing everything from the ground up. And before our systems get really old and how we do things, we're evaluating where we can go forward with the data that we wanna have. How do we wanna generate it? How do we wanna be aligned internally?

2:38 And what is that gonna look like in the new space and time of having the fourth industrial revolution and with AI? I think that's kind of what I'm trying to get to is that challenge for people, which is there are so many options now, like AI doctors and AI telehealth and all these different tools. And there's a spectrum of usefulness. So from someone who's kind of on the inside, what are some of the things that kind of say this is interesting or this

3:03 is not interesting or this is like because I'm already hearing from some people like, the my insurance company used AI to deny my claim. And Yes. It's like then you say, I can get an this AI tool says it will help me get my claim approved or help me improve it. And I deal with this all the time. I test tools a lot of the time that are much lower stakes and then have a wide spectrum of it's fully AI or

3:23 they're completely fake in their AI. I've seen both ends of the spectrum with low stakes tools. So you're in a much more high stakes industry. So I guess that's kind of for people that are thinking about Yeah. these tools. What are some of the signs or ways you notice this seems good or this seems wonky? I haven't found it especially in terms of patients getting access to medicine or it in the same case it would be procedures as well. I haven't found ~ anything, any resource

3:47 out there that really is a standard. This is the way you go. It's a very fragmented ~ medical system. And I'll give you an i an idea how I recently used Claude in terms of this. I have a ~ my husband's a rare disease patient who had a hard time. Getting access to his medicine, even though he qualified, because the AI on the other side of the pharmacy benefit company was ~ was denying him, even after a lot of we I know the system well,

4:17 my husband knows the system well. Over the weekend, I used Claude, for instance, and AI to say, my husband's up, he has he needs to get a procedure where he needs to get a special dose. And how are we going to do this with this company, knowing that we have obstacles of AI pushing back? And what COD was able to do was pull up that policy from our PVM specifically for that disease state. And this is an example where we will be

4:38 able to go in the industry to say, don't send another prescription in, do a quantity override. And then here's a PA form that you can fill out. Here are all the resources to get it through, and that will be your plan A. And then the plan B was a different drug. So that's where we will will be able to go as an industry of understanding these nuances that are often hidden within the payers in terms of the policies and what their methodology of getting drugs approved.

5:07 And it it and the savvier we get, of course, on the other end, is AI that is making the the pol the process more stringent. So it's a, I don't know, there's an opportunity for us to really get a handle on this fragmented system and pull the data together to make it easier for all parties. ~ payers spend a lot of time denying, but they're also wired in really old processes that have administrative burden don't want. So there the the potential is kind of linking all

5:35 of this information together. And that's where we look. we look at the patient journey all the time of what What takes it from the doctor writing a prescription to the patient actually that medicine in his or her her hands or being administered the ~ medicine to really help them maybe fight cancer or treat the rare disease or what have you. And it's that process that AI has the potential to really smooth out. I have not seen a solution for it that you know though,

6:01 yet. It's coming, some hopefully. I think that's the challenge is that there's kind of two ends of the spectrum. One is that AI is going to find new cures and new medicines and allow kind of more access to people that services, but also most people's first experience is dealing with their insurance company and kind of it seems like they're seeing an opportunity because it feels like the only thing insurance companies want to do is deny claims. I know that I've never had a good

6:24 experience with an insurance company and very rarely do meet someone who has. And It just feels like, there's just another way because if you get a person, there's a chance you can break through their emotional barriers and get some mercy from them and they'll help you. But with an AI, that chance goes away, right? So you don't even have that one in a million chance that the person will decide to save your life. And that's kind of The real challenge. And I think that many

6:46 people find insurance as a whole very baffling. There's so many companies and so many policies. Like, you're on policy A, but if you're on policy B, you'd be in a different situation. It's like, didn't you recommend policy A? And like, yeah, because we didn't want to help you. So I know for my family, we went on an insurance plan specifically to cover because it covered births. And then when we had a kid, we you have to wait a year before you're eligible.

7:07 And then they go, actually. And the The amount we were asking for was less than we'd even paid for the insurance over the previous few years. And it was like, well, then what's the point? Mm-hmm. So, like mo with this challenge, I think that's really kind of the thing people are trying to navigate now. And I know that the most common way people use AI now is to interpret their test results because for so long, Right, that makes sense. Yeah.

7:29 right, the results are murky and there's a huge gap between what the test results say and then what your doctor says. And having that knowledge can be like forearmed forewarned is forearm. Like you have a better chance because sometimes they won't let you look at the test results until the doctor interprets. And I can get that because I've definitely misread a scan of my own and been like, I'm dying. And they were like, No, that's not what you think it is.

7:50 That's not a shadow. Stop it. So I've done that too. But there is this dissemination of information, which is good, but then it also leads to paranoia or people Everyone's self-diagnosing now with these like pretty dramatic mental illnesses or physical illnesses or ailments. And now you have to ask this question when people say they have something, I go, did a doctor diagnose you or did you self-diagnose? Because Yeah, that can be really dangerous. Yeah. I I agree.

8:13 And I'm and I know that people do look at that. I mean, it's always you you always want to look at your test results with a trained professional in it. And but you know, on the other hand, I think what's also really interesting is ~ how physicians are really utilizing AI in their space and and improving their own ability to look at patterns and diagnoses. I think that's really been a strong point. And it it and utilizing it to self-diagnose it,

8:36 I I would never say that's a good idea. I I'm I'm with a system, health system here in Colorado where our our results are immediate and that was chosen by the chief information officer to make these results available to the patient, because here's the con conflict that your patients always have, at least in the US, is who owns the data. So the data's not owned really by the patient. And so that's a really big factor of how do you collect your own data,

9:04 how do you have that in front of you? In some other countries, you own your own data, like I probably like Switzerland, where you own that your since birth, like your vaccinations, diagnoses, everything like that. But in the US, we have a very fragmented system, and that's not usually the case. So if you had if you were running your whole history in combination with your physician and you're looking at different things, that's that's one way to look at it, or if you're pulling your genetic background,

9:30 if you have different gene mutations that could factor into your health or something like that, that that shapes things differently. ~ but I th I'm always worried when somebody goes just like people did when the internet was new, they would go and self-diagnose and ~ go down that path. And I'm sure in AI the people are going down that path quite a bit of my doctor missed this. But I think there's a collaborative aspect about it, especially if you do have your own data

9:55 and you can make that own your own decision with your physician. That's that's kind of where I stand on that. There's there's enough obstacles in our fragmented system. Yeah. It's like there's a certain amount of usefulness of knowing what these things Like I've noticed that when I get blood tests at different places, they use different measurements. So it's like this one's using moles and this one's using milliliters. Yep, sure. And that alone can be very confusing for people to get the metric.

10:19 Like, ~ where I live, we use completely different systems. So I can't tell my scores to an doctor because they go, Yeah. What are these measurements? You're a thousand times too high. I'm like, No, they're using a different like they're using liters instead They're different yeah. of milliliters. And that's all it takes. So I think that part is really useful where you can interpret data, but I have also seen the further end where the problem with AI is that it seeks affirmation.

10:41 So if you tell it what answer you want, it will give you that answer. It's like, I think I have this. Do I have it? It's always gonna say yes. And that's can lead to like these ~ Moonchausen cycles, right? Where you start off thinking, I have a cold and end up going, It's cancer. And then like you start to get depressed and like all of these things before No. you get to the doctor. And the doctor's like, what are you talking about? You don't even

10:59 have a like, but you so you kind of have this agreement cycle, which can be very dangerous. So that's kind of one area I worry about. The other thing that I kind of want to dive into is like the big we keep hearing these things like, ~ we found 500 million new cures with AI in the last week. And they're like, Well, then where where are they? Right. We keep hearing these big promises. And my real question is like Sure.

11:21 How much does AI actually accelerate the discovery and testing and medical release cycle? Like, is it actually going to be faster? Because the trials can't be sped up. You still have to do human safety and all of those things. So like I don't think we're lacking for ideas. I think there's other areas in the system where the problem exists more than the idea phase. Think about a lot of the AI is focused on, and that's not my background, is not the science part

11:48 of the pharmaceutical industry. How are you know, some of the things that are being looked at are there generics that are already on the market that can be used for different diseases that we hadn't thought of and that are widely available already? Kind of going back to your point. And then AI is also looking at different ways that molecules can be understood in terms of the disease state of the for the patient. it yes, trials still take. a long time. They they have to be designed.

12:13 They cost millions of dollars. You want to be able to check out the efficacy and the safety before you even go to down the route of does this treat my, you know, my disease state better than another medication on the market? There is the potential of a complete explosion of you know different ways of treating diseases more efficaciously, safer, quicker, whatever that is, with this new ~ age of looking at, you know, with AI looking back in the company. And I see these large corporations that

12:41 have already started to utilize in the in that ~ that can bring that in-house, AI in-house, and start looking at the different th molecules that they might ha have on hand already. Because there's plenty of discoveries within a pharmaceutical company that don't even make it to clinical trial because they don't know what to do with it. They don't know the direction. And if you look at the at the history of so many different drugs, It's it's amazing that they even can't like like GLP once,

13:09 like we hear about them all the time, the weight loss drugs. It you know, there was a scientist I knew about the Hula mist monster saliva, and that was just his knowledge. But can you imagine if it was accelerated a whole different direction where we could go with clinical trials and maybe not even have to do kind of kind of hit and miss sort of scenarios and it can be much more precision medicine? There's a potential of really having medicine balanced against.

13:34 If you have a gene mutation, if you have a group of genes that need to be that are affecting this disease state that can be modified, not the gene modified, but can the disease state be modified with new discoveries? I think we're on the verge of really finding more. There's there's a potential for so much out there. And ~ I don't know, it's exciting times. Yeah, I always feel like there's a lot of excitement discoveries. I remember when they were like,

13:59 We're gonna finally figure out the human genome like thirty years ago and then it doesn't feel like that much has changed since that happened, Yeah. I got it wrong. Right. right? They were like, Once we get this, it's all gonna change and then they did the twenty three Me thing. Once we have everyone's DNA we'll figure out how to cure and then like they went out of business because it didn't do anything and I think that sometimes our eyes are bigger than our stomachs and

14:18 I know a lot of people kind of imagine that now, like instead of doing human trials, you can have an AI do a human trials inside of its system, which we're so far away from, like so far away from that being possible, because the biggest problem with AI is inconsistency. Like you can ask the same question twice and get two different answers, and that's fine if you're writing a letter and very bad if you're doing science, right? Like it's like remote

14:41 There's a human there's a human total that. Yep. Yeah, like the most important thing about the scientific method is is it repeatable? Like that's the critical thing, Yes. and that's where AI kind of falls down. And I think that another area that is very interesting is we're changing our definition of privacy again. So I saw something recently which is kind of interesting, which is that like, I mean, when I grew up there was a level of privacy that doesn't exist anywhere. Now everyone's like post

15:06 wants to be famous so bad they're posting videos to TikTok and sharing different stuff. And like, you know, when I was a kid, if you were sick, it would there was a sh shame associated with it. Now it's kind of flipped in the opposite direction where it's almost like a badge of honor to have a different thing and tell everyone what you have. So it's completely reversed. Not that shame is a good thing, but it's gone in the other direction of like from shame to pride.

15:24 Yeah. And so now we kind of and now this the amount of data people will share with an and I've seen this having worked at a healthcare company myself, is that we don't There's this assumption that nothing will happen to me. Like I won't be the one lawyer who gets in trouble for submitting fake cases that an AI wrote, or I won't be the doctor whose patient data leaks. And it seems like we're so excited about these technologies that a

15:48 lot of companies use hope as a strategy. And rather than and everyone in the security industry is like paranoia is but certainly me as someone who leans heavily towards the security end of the spectrum, it's like, We don't know if these tools are trustworthy. Like the real measurement of if you can trust someone is time. Like a friend you've known for 30 years is more trustworthy than a friend you've known for two weeks because you have a larger data set.

16:10 These AI companies have been around for three years, five years, and it's like, well, would you let the person babysit your kids? Well, why would you give them your most personal of medical data? They promise they won't look at it. Like, how many friends come over to your house? There's no yeah. And look in your medicine cabinet. Like everyone. Like everyone does that. Yeah. So the promise doesn't mean anything, but we're so excited and I've seen like buying cycles for medical companies used

16:34 to be like three or five years for a hospital. And now they're like, we have to be faster because of AI. So it has to be three months. And the due diligence phase is kind of what's getting compressed in the security phase. And it's like more and more of Can we do it rather than should we do it? Or is this a good idea or is this a long term benefit? And it's like ~ the mistakes are happening once you've already adopted

16:53 a tool and you're kind of locked in. So how do you kind of see that new world where and how can users I guess develop Mm-hmm. a sense of when and where to be cautious with their data? Yeah and it's something that you know as a startup company that we're really profoundly looking at. I think we had talked about it. We ~ we currently aren't even using AI in our space because we really want to be careful about where our data goes

17:16 and how secure it is. we're probably not gonna build our own app, our own AI. And so now we're looking at enterprise or you know, individual and and how far do we are we willing to go just because of that risk. And I I think the start of it because we're a startup company, we don't have legacy systems that we're trying to modify or adjust to this new this new thinking, this new enterprise level thinking incorporating AI. What we did put together was ~ a council,

17:47 like kind of a general council, and I'm on that from different people within the company looking at what are our processes, what where are our gaps, like where could we accelerate, where can we go more. And how much of w what are we actually what's our risk tolerance? Kind of going back to the data. like I might have a risk-averse company, but maybe we could utilize to a certain extent and have certain privileges and make sure that we have guidelines and

18:10 make sure we have the right training so that there aren't mistakes and the wrong data gets in the wrong space. I'm currently getting my doctorate basically in the fourth industrial where I have The dean of my school, for instance, is working on A AI for the last 30 years. And we have this question in our classes, you know, how much do you really share? How much confidence do you have in a in a a public company where the CEO can change?

18:36 And we're not very clear about how where the data is gonna go or what could be shared or how secure it really is if you don't own it, if it's not your own company holding all that data. And so those are the just there's ethics questions, you know, what's the governance council really going to stipulate is enough is enough in terms of sh putting data out there? What do we really need to do to synthesize? We're really, really, really guarded,

19:01 even when we work with third party vendors and ask them how do they how are they using AI and are they going to use our data as a vendor in that space? And so we have to be really careful because it's it we have innovation and innovation costs millions of dollars and we want to protect that data. On the other hand, there's some clear avenues where we can go that aren't necessarily proprietary information that has a risk of being on

19:25 the internet that we can really we can we can look at patient data, tokenize it, make sure that there's no HIPAA violation possible and go, Where can we do better as an organization on that patient journey, getting access to a medicine that's prescribed for a certain disease state? So those kind of potentials are there. And I and I say that every company is going to be a mix of, you know, their risk tolerance. There's some, what I see over the weekend,

19:53 I think it was Bristol Meyer Squibb has just did a ~ an alignment with Claude, having all employees g getting very ~ involved in using AI. proactively and actively within the company, maybe BMS's risk tolerance is higher. They have higher security. So it's that that do due diligence on the other side, being upfront, obviously having the governance and teaching exactly how it's appropriate in the company and what can be used in that space or not. And then it's ~ you know, it's the cyber

20:22 risk part of it. You know, how do you protect yourself? Do you have the right sort of situation? And if your data does get out, what's the risk that goes back to the company? ~ in the in the long term? What are you willing to work with? And it's just a new thinking space altogether because you're not just having take data from a company and sending it, you can protect yourself or people in the company sending it out to their private email, for instance,

20:46 and you can do the cybersecurity around that. But do we have enough, do we have enough knowledge of how to protect ourselves in this age where AI is used so prolifically? challenge with these AI technologies and with how they work now is that if you accidentally push one button, all the data is out there. There's no undo button, there's no reverse, there's no chance to stop the mistake. And we've really shifted our security posture in the last 40 years. When I first started working in technology,

21:12 most companies gave you no internet access. So even at large companies. Right. You would just have an intranet, which people don't even know that word anymore, Okay. which is like you would just have an internal network and separate computers access the internet. So even I knew a lot of companies where you couldn't even see your own company's website. What they would do is download a copy of the website the night before and you could see like a photocopy of it,

21:32 but not a live version. So only some buttons were so you could see like what current promises and prices and terms of conditions and stuff were for support calls. And now we've switched to well, we just banned some and the thing is like. Bad website links change constantly. So this shift towards giving your employees more and more access to stuff. And it like you mentioned with vendors, it doesn't even have to be you. You know, a vendor can have access to some of your data.

21:54 They only need one dumb employee to make one small mistake or to log into the wrong website. And especially because a lot of these tools now want to ingest everything on your hard drive. Like if you have Windows on your machine, it's like taking screenshots, which Yeah. I've had a like it used to be how you would track employees if you were remote and you're like, we'll take a screenshot of 15 minutes, and some employees, like, I'm not doing that. That's an invasion of privacy.

22:14 And like now everyone will do it for free, which is crazy. Like, ~ you know, I don't want anyone watching me. What you know what I mean? Like I work from home and all those things, and it's like now we're just putting those into our systems and it's supposed to be more helpful, and we're kind of trading more and more freedom and privacy for these kind of panacea promises, but All you need is one person has that feature turned on, and then your data's out there,

22:34 and you just have to hope. And that's seem to be a lot of strategies. Yeah. And having worked in security and worked with some of these larger companies, like they're not as security conscious or risk averse as you would hope they are. What they actually are is ignorant or excited. I've worked with companies in healthcare who sign up with a company. I go, they have no security policy. And they go, but they're really nice. And I was like, Well, that's That's not good enough.

23:33 A great defense on your that'll be great at your trial. I wilt flip. Yeah. Like when I've worked on those projects, I'm like, listen, if I tell you not to do something and you do it, you get to make that decision. But I will if they ask me questions, I will definitely not fall on my sword for you. If it wasn't my decision, I will just say I told not to do it. I will testify. Let's do the immunity thing.

23:53 Let's I'll talk. Like, but if it's something that's my fault, then it's my fault. But I just find that it's so exciting and we're maybe we're at the end, everyone keeps thinking we're at the end of the AI bubble, which is the excitement bubble, which is like anything with AI gets more funding, gets more investment, gets more attention. And that's kind of the problem, which is like AI is not the answer to every problem. It's can do a lot of cool stuff,

24:17 but like, you know, ~ if I break if I lock myself out of the bathroom, which I did the other day. AI can't fix the door. It's happened twice in the past two weeks. Okay. ~ Like you just forget that you locked it from the buttons, like simple things that happened. I think that's important to remember that there's a lot of reasons why not using AI is actually better than using AI. I deal with it constantly where AI will give me the wrong answer nine times

24:40 in a row and then I'll say this is the answer. And then it will forget. It's very worrying. And the think of like using and I use very high end tools. So like It's not the quality of the tool I'm using as someone who's at the cutting edge of it. Like I trust it less than most people, having seen what's possible and how easy it is to trick it. And it just opens up these new vectors where now you have a software that

25:02 can be social engineered or tricked or manipulated or all it takes is Course. one mistake in the coding. And I think about it this all the time. And so there is this really kind of brave new world we're trying to figure out. And I think what you're doing is very interesting. I think that you're someone who's at the intersection of two things, who's been on the bad end of having a bad insurance experience. And that kind of colors how you treat your patients,

25:23 which I think is a good thing. So I'm very excited for what you're working on. For people who want to know more about your projects and kind of are interested in the direction you're going in, all those things, where's the best place for you to find you online and some of the amazing stuff that I've seen you writing about recently? I think the only thing I have right now is just my my LinkedIn. it's where where I utilize most of my business or academic resources or

25:44 or to leave comments and things like that at this time. Well, I'll link to your LinkedIn bef below the episode and in the show notes. Thank you so much for being here, Wendy. Wendy for an amazing episode of the Artificial Intelligence Podcast. Yep. Thank you.