From Clinic to Cutting Edge: How Physicians Can Lead in AI and Innovation with Dr. Mark Allen

In this episode on Startup Physicians, I talk with Dr. Mark Allen about his journey from medicine into the tech industry and the transformative role AI is playing in healthcare. Mark shares the challenges he’s seen in adopting new technologies, the evolution of AI, and its potential to reshape medical practice. He explains why generative AI matters, how it can drive efficiency, and why it’s critical for physicians to engage with these tools now. Our discussion focuses on how embracing AI can enhance patient care and help us adapt to the rapidly changing landscape of medicine. 

Episode Highlights

[00:00] - Introduction to AI in Healthcare
[00:59] - Mark Allen's Career Journey
[07:53] - Challenges in Healthcare Technology Adoption
[10:05] - Transitioning to AI and Drug Development
[14:13] - Generative AI: A Game Changer
[21:32] - The Role of Physicians in AI Development
[27:15] - Preparing for the Future of AI in Medicine

Transcript

Mark Allen: 0:24

I can teach you something in an hour that is going to save you hours every week, yep. And that's kind of my promise. So you can be, effectively, a magnified version of yourself. So that's really the possibility and the power of this. And the quicker you because this is, this is a rocket ship that's taking off. The quicker that you jump on it and start to use it as a daily part of your life, the more that you're going to be caught up and then profiting from it, profiting it from in every way. It's your business life and your personal life as well.

Alison Curfman: 1:00

Welcome to Startup Physicians. Please like and follow our show to join our community of physicians who are reimagining healthcare delivery. Hi everyone. Welcome back to Startup Physicians. This is your host, Dr. Alison Curfman. I'm excited to be joined today by my colleague, Dr. Mark Allen, who is someone who's always been really inspiring to me since I met him about really incredible knowledge in the AI field. He has had a lot of different successes, starting different companies and running different companies, and so I thought he would be a great guest to come on and share with you guys a little bit about his career pathway and how he used his medical background to enter into new areas of industry and make contributions there. And then we'll spend a little bit of time talking about some of the ways that AI is changing healthcare and ways that physicians can get involved. So thank you, Mark for joining me. How

Mark Allen: 1:58

wonderful to be here. Thank you for having me. Alison, awesome.

Alison Curfman: 2:02

So can you walk us through kind of your career journey, and how you, you know, started your first company, and when that was and other things that you've done?

Mark Allen: 2:12

Yeah, absolutely. So let me start here. So I went to medical school because I wanted to help people. That was kind of the way that I framed it is, you know, something that I do, the output of my work should help people. And as I got into medical school, I recognized that there was a challenge which we can only help people one at a time while we're working. There's very little scale in our ability to help people. And, you know, I now frame it as, you know, we're trading our time for that value. And so I, I was fortunate to get involved in an AI project during my medical training, which was at UCLA. So it actually started my second year in medical school. I hooked up with some senior faculty in the emergency department that were building medical expert systems, so systems that doctors used to effectively. And they started out being just helping to diagnose, but they were trying to incorporate it into the process of care itself, to provide actionable recommendations. So this as a timeline. This was in 1995 when I started in that

Alison Curfman: 3:31

research. I didn't even know AI existed back

Mark Allen: 3:33

then. Yeah, no, AI has been around as a term since the 50s, and it has been slowly improving exponentially, pretty much every year. But you know, when it's really, really dumb as an ant, you don't really notice it. But what's happened very recently, in the last couple of years, is that there's been a crossover point, okay, where now all of a sudden it's smarter than humans. So first it was smarter than high school graduates, and then that was really interesting. Chatgpt, 3.5 was there. That was when it was kind of a breakthrough. Oh, this is really interesting. Oh, but it's not that good, right? It's not that useful. But it kept getting smarter, and then it was smarter than college graduates. Now it's smarter than PhDs in their field, but it's improving exponentially. So, you know, an exponential curve looks like this. Okay, so it's, you say it's really deceptive when it's growing super slowly. But we're now at like this part of the curve. Okay, so the next step is it's twice as smart as people. Okay, right now they're already creating exams that are too hard for people, people can't even get a single question right on these exams. Humanity's last exam. You might have heard of that, and the AI is now starting to get like, in months, it went from single digit percentage. Now it's the top models are like 40% but it keeps getting smarter exponentially. So way back you could build really quality applications. So these doctors, going back to my story, these doctors had built a system that worked inside of the UCLA network that guided doctors to it was they had to build it out for every single encounter. So they started out with low back pain and getting stuck by a needle or splash. By blood. That was the one, first one that I started working on. And it would take them many months to build a system, and then they would test the system in clinical trials, and they showed that when doctors use the system, that they were twice as fast in the encounter. So this is way back, and I'm talking Well, we did. We're doing the testing in like 9798 they were twice as fast an encounter. They ordered more of the right tests and treatments, but they ordered far fewer of the wrong tests and treatments.

Alison Curfman: 5:55

Okay, so it's like a clinical decision support tool. It's

Mark Allen: 5:58

now we call it clinical decision support, yeah, okay, so back then we called it expert systems, and we so I had had some experience from the early days just building web apps as a side business while I'm in medical school, was the early days of the web. And so I helped them to write a grant. The CDC ended up funding to build what was, to my knowledge, the first ever web based medical expert system. So we got this working in a number of different encounters. Most complicated was fevers in children, and this is in the emergency room and urgent care setting. So it would take a lot of time to effectively define the clinical practice guidelines, the algorithms and my team would effectively put it into software. And so we showed these these improvements time after time. And so I during this time, I took a year off. I did medical school in five years, instead of four. As I was working on this research, I then took a position as a half time emergency medicine resident to stay at UCLA. And so I had my my double time internship here, and then I'm a half time resident. So I was equivalent of 212 hour shifts a week, and then working on the software the rest of the time. And as I would go into a shift and I'm working on the software, I'm like, God, if I just had my software to help me with this problem. I don't exactly know what to do. I don't have time to look it up. And so I sat down one day after a shift with my mentor, who a guy by the name of Larry baroff, who's the head of the emergency department at UCLA. And I said to Larry, I said, you know, I think that I could make more impact bringing this software to market than practicing 1000 lifetimes.

Alison Curfman: 7:40

Yeah, it's that scale impact, right?

Mark Allen: 7:42

Yeah, the scale. And he said, I agree. How much do you need to start the company? This is the end of 99 and pulls out his checkbook. It's like, well, hang on, let me think about that for a little bit. I thought about put together a business plan. I didn't even have a business plan at that point. Put together a business plan. Over the next couple of months, I went and talked to a couple of other investors. The next two people that I talked to were like, you, we want to invest in this just based upon our clinical trial results. And so I left my residency, so it's not even halfway through it, and just took my residency. Director told me this is a one way door. Don't expect to get back into this program. So the I left it, and I hired a team, took a million, little over a million dollars from friends and family, and we were as we're building out the commercial version. So we had these prototypes, we had apps that we had put into production. I went to sell it to hospitals, and I'd have meeting after meeting, and the hospitals would not buy. Finally, six months into this, somebody was finally kind enough to explain to me. They said, We love the physician productivity, that they're twice as fast when they use this software. We love that. It improves the quality. But what you call cost, all those tests and treatments that your software is going to tell us to avoid. Well, we call that revenue. We get reimbursed on that and if we were to at that time, we showed statistics that we reduced costs by 27% not including labor. He said we would literally be out of business if we did that,

Alison Curfman: 9:19

if they were ordering appropriate tests Exactly. Well, that's disappointing. So what did you do after that?

Mark Allen: 9:27

So I went back to my team. I was like, Okay, so, you know, we kind of, like, I really didn't think about this. And this is a lesson in financial architecture to really think through your business model. And we're like, okay, there's fundamentally two customers. There's the Kaiser and there's a VA it's not enough to support the company. Like, what other industries can we go after? Because

Alison Curfman: 9:50

back then, it's not like there were, like, a lot of risk bearing entities. It

Mark Allen: 9:54

didn't exist. So so we pivoted and started selling into the insurance industry and financial services industry more broadly. So doing things like trade compliance for stock and bond trades, took a big investor in that mutual fund market. We got into the area of insurance claims adjudication, and took big investor, couple big investors in that market and in banking for lending to. Decisions. So we became, over time, a 12 year overnight success used now that product's called Corticon, c o r t, i, c o n, and it automates billions of enterprise decisions every day. So Cigna is an example in the health care field. Most of their claims go through the quarter count system and every Don't blame me for claim getting denied. I don't control the rules. The it was interesting, and it wasn't like it was, you know, the original vision, why even started the company, was to help people's health at scale and to keep the company alive like I defined. Where could we sell this software? So we were in the right place at the right time when Obamacare happened. Okay, so we we had had a number of government customers that were using it for eligibility for various benefits programs, and then with Obamacare, every single state had to redo their eligibility for health benefit system. And now we're on the right side of financial architecture, because the government said we're gonna you have you must do this, and we're gonna pay 90% of the costs. And we already had a few successes, and the company just took off. Boom. So sold it at that time to a publicly traded company. And then I worked for this, this company called progress, for four years. We have when you sell sometimes, like, you get these golden handcuffs. So I made money on the sale, then made and then it's like, okay, you're going to get more vesting stock. We want you to do this. And this, it was actually a lot of fun to be part of a public company and have different roles in it. And then I left, and I really went back to my drawing board after four years there, and I said, Okay, so how do I help as many people on this planet as deeply as possible? So I kind of added that it's like if you help a billion people in such a way that they pay $1 each. You've got a billion dollar business, right? So if you help people in a deeper way, then you know, it could be an even bigger business. And so, so I really that was kind of the question that I started with. And so I got into, then this area of drug development and specifically targeting aging. So drugs that target the aging process have the potential to treat and prevent many different aging related diseases as like that's one of the biggest problems in humanity. It occurs to everyone on this planet and man, that's a big problem that people would pay a lot of money to be able to address so I basically built, founded a biotech company. I on that thesis, I researched dozens of projects. I ended up partnering with three Harvard professors to found a company called the LeVian like E, L, E, V, I N, and one of the most exciting projects in in my mind, in this field of longevity, they had discovered a natural protein called gdF 11 that promotes rejuvenation and repair in aged animals. Okay, it's an essential protein. It's identical across mammals, even insulin, another essential protein has differences across mammals, this, this has no differences. Any mutations are incompatible with with life, if they're on actually both chromosomes. There are some mutations on heterozygous mutations that have severe malformations. So really exciting kind of target. So over seven years, $40 million we took that from a breakthrough academic discovery to just starting clinical trials in the field of stroke recovery with some incredible animal data. And the market changed as the interest rates went up. We couldn't fund the clinical trials. I could not I'll take responsibility. I could not get those funded. So you know, of my company, 1100s of other assets, got stranded, and we had to sell it for pennies on the dollar. Super sad story, you know, disappointing, yeah, the new buyers can take it along. Seven years. You know of, you know, my time, my team's time, $40 million you know, gone into that project, and so, you know, I learned a tremendous amount, you know, in that process, about company, about drug development. One of the things that that I learned, though, is that in biotech, you have massive risk, and you have massive amount of money that's required, and incredibly long timelines. So you know your average to get a drug to market now this is not what a biotech company usually doesn't take it all the way to market, right? Average to get a drug to market is over $2 billion and over 15 years,

Alison Curfman: 15:22

right? That's such a long timeline. Right? My

Mark Allen: 15:24

God, and, you know, so I took a big swing at it, and, you know, it was a great thing. And what happened, right? As we had to put the assets, we had to basically put them into a they call it an ABC process, to liquidate the assets. It's like, okay, so now I'm going to ask that same question, what do you do next? Okay, so I want to help as many people as I can, I want to help people as deeply as I can, but I'm going to add another dimension. I want to help them now, right? Yeah, I don't want to wait 15 years with a bunch of risks help. Yeah, so right, at this time generative AI is taking off, okay? And I knew, from my perspective, we're at that crossover point. It's smarter than humans, and it's still improving exponentially. And so I started to, I would say, play with the technology initially, and just apply it to the kinds of problems that we would apply to Corticon to. And so Corticon, let's say some of these enterprise applications that we would do take 10 people six months to a year to put something into production. Okay? And cost millions of dollars. I could build a similar application with generative AI in a matter of a few weeks, just myself. And I was like, Oh, my God. So the difference is that this generative AI, it has all of the data in it, so you don't have to feed it the rules. All you have to do is focus it like glass is a magnifying glass, right? You focus it to do the task that you want it to do, and maybe there's a little bit of tuning associated with that, but it can then do it, and it performs at insanely high level. So today, it diagnoses better than doctors, right? It can write contracts better than lawyers. It's better than the best of the best. Okay, now it still hallucinates a lot. Still has some issues, but to build these applications, and really to determine whether it's outputs high quality, takes experts to help you to build it right. But it is capable right in every field. And so I got really excited. I saw that right away, so I kind of learned it. And then I built a course to train other people. And I began to focus on, like, weird, who do I want to train? Okay, I want to train business leaders. I want to train physicians. And it's kind of like that combination became my core market, so people that are kind of physicians looking to do entrepreneurial things. I got a lot of those kinds of clients. And then I have entrepreneurs, small business owners, business leaders, CTOs, of different organizations. So I so after that initial course that I built, I then was like, this is changing so fast I can't even keep my course up to date. So I created a mastermind that allows me and my team to focus on, okay, what's new? What are the greatest technologies that are out there right now, and how do we train people on them so that they can use them? Yeah,

Alison Curfman: 18:20

make a plug that I am in Mark's mastermind, and have already been extremely impressed with the amount of up to date knowledge, because I'm someone who's definitely been keeping up with AI in the form of, like, all of the health tech advances that are happening. And I think you and I can talk a little bit about what doctors can really expect coming down the line, and what what what sort of hesitations we have. But I think that you're right. It changes every week. So I really value the opportunity to be part of your community and have like, a up to date, like, up, this new feature came out up, let me demonstrate this entirely new thing that could actually save you, potentially, like, 12 hours this week. So that's my little plug for your mastermind. It's called Hero Forge, right?

Mark Allen: 19:08

That's right, yeah, yep, Hero Forge. And it's like that the AI makes us all superheroes, gives us superpowers, if we just know how to use it. And I thought it might be interesting to kind of sharing a little bit about what are the most powerful and transformative AI technologies today. And, you know, kind of, how do you put it into context? So, you know, the core of this whole AI revolution is the model itself. So we call it the foundation model. We call them llms, large language models, kind of the first ones, but there are versions that are specific to images, versions specific to videos. There's versions that can do multiple things. We call them multimodal. So those models keep getting smarter and smarter and smarter themselves. Okay, they're exponentially improving, and a lot of that comes down to the amount of compute you put into building it. Okay? It takes many months of actually, years of compute to do it. They put huge server farms, like some of the latest ones, like grok fork, 200,000 of some of the most expensive CP GPUs. They call them NVIDIA GPUs, to work for months to create this model, okay, that's that is all of the as much data as they can get in the world. They feed into it and crunch it into this neural network, okay, which is similar based on the architecture of the brain, and it effectively creates this brain that has all that data in it. Okay, so that's the core of what a model is. Now, a model itself has no memory. It doesn't it only knows what it's been trained on. It doesn't know the most recent data, and it doesn't remember anything about your conversation. So data in, data out, so you have to plug it in to an architecture that we broadly call an agent to make it work. So a chat bot, which is things that everybody's familiar with, chat, GPT and those bots, okay, so those are the core model. And in fact, you can select different models. You select the newer models, the more reasoning based models, okay, but what it does is it's part of an agent that also has memory, so it has memory of your conversation, okay? It also has the ability to use tools. Example of a tool is web search, okay, so the more that you build, and there's becoming standards to build these agents, so we're starting to see way more powerful agents, okay, that have more tools associated with them and the ability to run long running tasks. So the kind of the first example that consumers really used a lot of as a super powerful agent was deep research. So Google came out with the first deep research agent. It's actually just February of this year. Think about how recent all this stuff is. And so that was a model that could reason over a long period of time and basically hit the web at speed, Google speed, review lots of different documents, analyze that with its reasoning, and put it into the form of report. Okay, that was an example of an agentic task, or an AI agent, okay, so now we have AI agents all over the place. We have AI agent tools that we can use personally, okay, that allow us to select different tools that we can connect to. We talk about those and train on those, like Gen sparks and other cool ones, but all the chat bots are becoming these powerful agents that allow you now to, example, connect with your email. That's a tool to connect with your email and do a task like a simple task of review all my emails in the last week and let me know if there's anything that I still need to get back to and, oh, by the way, draft helpful responses based upon the way that I respond to emails. That's just a simple so

Alison Curfman: 23:14

what you're describing is a prompt. So I am assuming that some of the people listening to this probably don't do very much with AI yet, or they just do some really basic things, and I think you're hitting on something that one of the most important, like the model, can be really incredibly brilliant and smart, but if we're not prompting it right, you're not going to get the right answers. And so I think that, you know, I'd love to hear your thoughts on, like, how all of this is folding into, like, the daily practice of medicine and how we as doctors can stay on top of these things. And also, I mean, I do a lot of work with health tech companies, and I know that I have this very, very strong desire to get more doctors to understand these technologies so they can actually be used as experts to help develop them instead of just receiving a finished product.

Mark Allen: 24:10

Yep, oh, that's key. There's so many opportunities right now to get involved in AI development. And what if

Alison Curfman: 24:19

somebody says, I'm not an AI expert right now? Like, is anyone? Like, I mean, I feel like know more, you know way more about AI than I do, but even just a couple of months of immersive learning, I mean, I feel like we're so early in this that, like anyone can choose to become a really put your head down and stay connected to this development and choose to become an expert in this, because I think It's going to be super important for medicine, for doctors, to understand this

Mark Allen: 24:49

absolutely super important in every job you know, it's, and it's, it's going to change every job you know, as I say, right now, at this moment in time, okay, well, we again. The point is that it's exponentially improving, but at this moment in time, it's already smarter than all of us. Okay? So the In other words, if somebody really focused, they could have an AI do most jobs that we do, most tasks that we do better than we can do them. Okay? So the. Wave is coming, okay? And you can surf this wave. You can enjoy it and have fun with it by learning how to dance with AI, okay? Or at some point, it's going to crush us if we don't change and learn to use the AI. So that's really important. It's a scary thing to think about, but it's going to happen. And, you know, it's we've had waves of technology, we just haven't had them move quite

Alison Curfman: 25:51

this fast. It's going to be uncomfortable.

Mark Allen: 25:54

Yeah, you know, at the turn of the century, of last century, okay, 90% of jobs in the US were in agriculture. Wow. Industrial Revolution happened, and now it's a small, single digit percentage. They're still in agriculture. So it jobs change, and we adjust, but usually we have a generation to adjust, so it's going to be a lot, lot faster this way. So you know, what can you do? So first of all, you by using AI, you can do just about any job better and faster. And there are so many ways to use AI that are that can be incorporated into the workflow of your work, okay? And, you know, just the tools that you use. So, you know, one of the things that that I do is I will have conversations with as you can put it into a voice mode, and you can have conversations with all the leading chat bots right now. And it's like talking to a to a person that knows everything. Okay, you can give it assignments to do, and it'll search the web. So sometimes I'll, I'll give it I run. So this concept of deep research. So deep research is like, I want to research a topic, like these topics that I would research, let's say a disease. Just use that as an example. I want to research this disease. I could go and spend hours researching that disease, okay, for a patient, or I could kick off a deep research report. And it takes me a few minutes now to kick off a deep research report, and then that deep research report comes back with an analysis, reading hundreds of papers and coming back. Now, you can feed it the you can feed it a lot of the papers, if they're behind paywalls, okay, you can feed it right into it, you know, but it has access to it, everything that's not behind a paywall, and then it will come back with an analysis, and you can tell it the analysis that you want. Okay, so, you know, here's who I am, the more that you kind of understand how to build context. And this, there's, there's a couple skills here to really understand. So it's, first of all, it's prompt engineer. So how do you talk to the AI, okay. How do you write a very clear prompt so that you know it's a difference between using it like a search engine, something very simple, and using it like a pro, okay? Part of kind of a breakout skill of prompt engineer is also called context engineer. So it's what context you provide it with. So I write these prompts that I have context of depending upon what I'm doing, that could be hundreds of pages of documents long. I could have context like my email, okay, for certain things that that, or my Google Drive, things like that, or a bunch of papers, right? So that's called context engineering. That's part of prompting now is to get it to work that way, and you can do that just for a one off task. You can also set it up for a repeat task, and if you repeat the task often enough, you can set it up into an automation or even an app. Yeah, you saw like our class last week. We're talking about, how do you build the apps? Right? It's this,

Alison Curfman: 29:05

maybe, yeah, maybe kind of a little advance for some of the people that might be listening and won't have heard all these things before. But I think one of the pieces that I'm hearing that's interesting. First off, you're a very forward thinking visionary founder. I think that your journey through multiple startups has been really interesting to see how, especially your thoughts on how physicians can contribute to even like biotech and drug discovery and and I'm sure AI is going to make a massive change in that timeline for getting drugs to market. And then now you're focused on AI, which I would, I would highly encourage any physician listening to this to follow some of Mark's content and consider some of the things that he offers, because I think that we can continue to stay as educated as we can about things coming down the line. Because, I mean, I personally do have fear about some of the things that may happen in healthcare, and I'm sure we all do, because we all want to stand for, like, incredibly high quality for our patients. But what would you say that people is, like, the one thing people can take away for, like, how they can prepare for this sort of change that's coming.

Mark Allen: 30:22

So, you know, one of the things is. Like I say, where do you learn? How do you learn? You know it's right now, if you just like, do a Google search for how do I learn? It's overwhelming. There's so much information. Where do I even start? Okay, so you know, finding, finding a community, finding a teacher, a mentor that can that can guide you, that you resonate with Okay, is is super key. So, you know, I recognize, I appreciate you so much being a part of the mastermind. And you know, our folks that are part of the mastermind, we try to build a community of people that are really passionate about learning this and using it to elevate humanity, to create abundance, create prosperity, right? That's really what we believe, and it's very much possible. So, you know, I know some of the things I say may be overwhelming. This is wow. Is a time of change, also a time of change is also the time of greatest opportunity, right? So you see the wealth that's being created out of this movement. Now it's it happens to be in the hands of the people that are using the technology and that are building with the technology, but it's like the greatest gold rush that we've ever seen. And there are companies that have, you know, there's a company that had one person got sold for $80 million okay? There's companies. There's multiple companies. I'm talking not talking about the AI labs. The AI labs like building the foundation models that we talked about that is a sport of kinks. It requires billions of dollars, okay? But to build on those models,

Alison Curfman: 32:05

anything you can create all sorts of things. Yeah, no, I've, I think it's really incredible. The one thing I would kind of encourage people is that, you know, when we think back on other technologies that have come through healthcare and made us uncomfortable, I remember, I very vividly, remember giving a talk on telehealth back in like 2014 and having, like, a line of people in the aisle ready to yell at me that, like, this is never going to be quality care, you can, you know, it's going to be cutting corners, and this is not how we deliver medicine. And whereas now it's, you know, we know there's some cases that, like, telehealth is absolutely not appropriate, but you know, we've adapted to rapid change, which was required by the pandemic, and then a lot of policy changes, and as we adopted this technology, most of us couldn't do our job without the idea of telehealth. Like I don't think I would start at a new practice if they didn't have telehealth as an option for certain types of visits. And so I would encourage people, when it comes to AI, to really latch on to other, you know, trusted resources, especially in the physician community, so we can all continue to learn together. And when it comes to health tech, and, you know, the companies I support, and some of the companies that I, you know, connect other people with, I think having a really good knowledge of this stuff is is super important.

Mark Allen: 33:21

You you kind of brought up this point, and I want to really get this across, is that, like, I can teach you something in an hour that is going to save you hours every week, yep, and that's kind of my promise to you, so you can be effectively, a magnified version of yourself. So that's really the possibility and the power of this. And the quicker you because this is, this is a rocket ship that's taking off, the quicker that you jump on it and start to use it as a daily part of your life, the more that you're going to be caught up and then profiting from it and profiting it from in every way. It's your business life and your personal life as well. So

Alison Curfman: 34:09

I appreciate all of your insights and encouragement. And so if you guys want to check out Mark's website, it's hero for hero forge.ai. Correct, that's correct, and we'll put it in the show notes as well. So thank you so much Mark for joining me.

Mark Allen: 34:25

Thank you so much. Allison,

Alison Curfman: 34:27

thank you for listening to Startup Physicians. Don't forget to like, follow and share.