DGTL Voices with Ed Marx - Even If They Feel Ready, They're Not (ft. Hooman Rashidi)

Episode Date: August 20, 2026

Dr. Hooman Rashidi is Associate Dean of AI in Medicine at the University of Pittsburgh and Executive Director of CPACE. Before Pitt, he founded Cleveland Clinic's Center for AI and Data Science and directed AI at UC Davis Medical Center. Ed asks him whether medical staff are prepared for AI. His answer is that even the ones who feel ready are not, because people bundle every kind of AI together and the newer generative tools don't behave like the predictive models the field has been using for decades. In this episode of DGTL Voices, he explains what's missing from tumor boards, why he leads with education before deployment, how Pitt runs a hybrid strategy rather than committing to vendors alone, and why his center puts junior contributors on patent filings when most institutions don't. https://bio.marxadvisory.com/

We talk about medical staffs. Are they generally prepared for AI? Even if they feel like they're ready, they're not ready. Nearly all of these conferences are missing somebody who's well-versed in AI. As more and more of these AI tools are making predictions and are influencing particular clinical decision outcomes, it will be very important to also have an AI-enabled practitioner at these multidisciplinary conferences, who can also put in the two cents about the limitations of some of these models. AI education is not just something nice to have, but essential to have.

Welcome to Digital Voices, where healthcare and life science leaders explore the real work behind transformation. This podcast is about people, leadership, and the conversations that move healthcare forward. Now your host, Ed Marks. everyone welcome to another edition of digital voices thank you for listening we know you have a lot of different choices you made us top 10 in the world and it's partially or mostly because of awesome guests like dr humann rachidi humin welcome to digital voices yeah thank you so much ed it's it's a pleasure being here thank you for the invite yeah this is going to be so much fun we have a lot of commonalities in our background in addition just to being handsome men you know it's

It's going to be fun to unpack all this. And what you're all doing, and I don't want to jump ahead too far, but what you're all doing at UPMC and the University of Pittsburgh, it's just amazing, just mind-blowing stuff. So we first met, we crossed paths over a clinic, but we got a chance to meet in person, which we'll probably talk about a little bit in Pittsburgh. It was, again, amazing experience.

But the most important thing, Human, that we ask in digital voices, are what songs are on your playlist? What kind of music do you like to listen to? Yeah, that's, that's a, so one of my favorite bands is the Steve Miller band, flying like an eagle and apricadapra. And I'm also a Depeche mode fan. So that's in excess in Depeche mode. There's a bunch of stuff from them as well, yeah. That's good music.

And I'm hopefully going to remember to put in the show notes a rap song you did. That's on, we're putting. Oh, right. You get a rap song. And we're putting that on our playlist. We do them a Spotify playlist. I do like hip hop as well too. Yeah, that's true.

That was actually a good tune. That's where it shows AI can help any soul, including me, to make a nice hip-hop song. That was great. I sent it like I told you, I would, to my producer's son. What about life message or mantra? Are there words that sort of guide how you live? I like to, you know, be able to reflect back. and think that that I've had more positive impacts than negative impacts.

I'm just like any human. I think we're all flawed and we're never going to have, you know, 100% perfect positive impact. But if I can have more positives than negatives, I think of, you know, I've lived a good life. Yeah. That's what I live by. You know, and I want to, you know, I want to make sure that I'm helpful to my friends. neighbors, you know, both local, national and global communities.

I mean, so I, you know. Being a good human, which you are. Yeah, just be a good, good human. I think it's, I think there is karma. I believe in that. And I think if you do good to others, good things happen to you. Well, I want to talk a little bit about like you growing up. First, you got one of the coolest names.

So, you know, also about your name, where you grew up, you know, what's your story? Yeah, so I, I, the name is, is a Persian name. So I don't, don't quote me, but I think it means like it's, it's, it's, it's an old Zoroastrian name from, from Persia, you know, so, so, so my, my parents named all the kids with names that start with an H. So literally they named everybody with an H. So I'm, my name was Human. And then they're all mostly were Zoroastrian names.

And then my brother's name was Herbod. And then my sister is, kind of like more could be, you know, pronounced, you know, more traditional. It's Halle. But some people say Holly. And then my, my youngest brother, the reason I'm telling you about all my siblings is because my youngest brother, his name has become very popular because it's Hormos is like the, you know, the, Hormos is right now in every single channel talks about Hormos. That's pretty cool.

And so where did you grow up? Like, are you? So, so I grew up. My early part of my life, I was in Iran, where, um, But then after the revolution, we left and I went to school in Italy for a little while until we then came out to U.S. and I, you know, did my high school. And, you know, initially at Santa Monica High School, then parents had moved out to Arcadia. Then I was in Arcadia, where I finished my high school there.

and then went to undergrad at University of California, San Diego, and then grad school there. And then from there, med school, residency, fellowships, all of that. Yeah. Very cool. Do you speak multiple languages? I do. Yeah. Yeah, I do.

I still speak. Like Farsi? I'm pretty, you know, I would say, depends to which one of my friends you ask. but they would say I'm still pretty fluent also in Italian, you know. But I'm probably a lot better with English, you know. No, that's a great multicultural background, that's for sure. Was there a pivotal moment in your life that fundamentally changed your trajectory? I think I was not planning on going to med school initially, but I think, you know, I saw some, you know, friends and,

folks that I looked up to who went into medicine. And so that, that made me think that, you know, because I was doing bioinformatics, you know, what we call machine learning type work before met school. And I started thinking, you know, could we apply some of these to medical stuff? I was very naive back then, you know, I did not know about HIPAA or anything else. So I think that's, you know, meeting some of my friends who, you know, were going down the medical trajectory. I think their influence, you know, got me to move into medicine along, I mean, to be fair, also with my parents, of course, you know, typical Persian parents who want their kids to be a physician, you know. So I think that helped as well, too.

Yeah, that's pretty cool. Before we get into that, your healthcare career now and make sure I got all of these titles correctly, Associate Dean of AI and Medicine, Professor and Dow Chair, Lombardi. Yeshunezaku, yeah. Experimental Pathology Research, Executive Vice Chair, Computational Path and Informatics Division, Executive Director C-Pace, which is where you and I hung out, AI Center, and just general, a cool person. Did I get all the titles, right?

Thank you. That's very kind to you. Thanks so much. So, yeah, that explains a little bit why you chose health care, right? The influence of others and certainly your parents. Did you ever see yourself, you know, the AI was around then, but not as we know it today. Did you ever see yourself like as a young person, maybe in college even? Like, man, someday I'd be like this world clinician leader. No, not at all. I mean, I think I credit probably mostly people around me who basically I got lucky, you know, know, who influenced me, mentored me. That's one of the reasons I like to mentor people. It's just, I think the best things that have happened in my life have been, you know, because of people who mentored me and put me in the right trajectories, like, you know, that going back to my graduate school time, Doug Smith, who was like one of the best graduate mentors on the planet, who basically took me down the right path. similar people since, you know, like, you know, I've had fantastic mentors and medicine,

you know, like Ralph Green from, from University of California, our old chair at UC Davis, Lydia Howell. So I think, I think it's, it takes a village to, to mentor, you know, the, you know, tomorrow's leaders. And so I honestly always want to kind of, prepare for that as well. Yeah, do what you can do that's best for now, but also plan for the future and how you're planning on passing on the baton. So I think that's probably why I still enjoy teaching med students, being around met students, residents, fellows, junior faculty, because I think they can probably do a better job than I did, you know, once they're prepared and, you know, for their journey.

Yeah. No, that's a great philosophy. And the world needs more people like that who are willing to give and build, help build others. So in that earlier education, you learn a lot about bioinformatics. And obviously, that's helped shape you today. Like, how has that, like, been key, right, to your more recent success with AI? I have to say, you know, obviously the journey has changed quite a bit, right? I mean, so many of us kind of have mixed feelings.

about how AI has changed our world. I mean, as somebody who, I mean, I do also still wear the, you know, R&D director had for our AI center. You know, I am a software developer. You know, I've developed various stuff. So I do enjoy coding. I like building things. And I have to say, the whole journey has changed quite a bit from, you know, let's say go back

three years, four years ago till now. I think the whole pre-chat GPT era, I don't think any of us thought that, you know, this whole vibe coding environment would be a major part of our, you know, workflow. And now I don't think anybody thinks twice about it. It's part of our system, part of what we do. And it's got a lot of advantages, but it also has created some challenges for us to overcome. I definitely see, you know, certain parts of AI that are really helping. And then there's other parts of it that I think we need to educate the masses so that they're aware of its limitations because it could also take you down the wrong path. Yeah. No, that's very provocative.

We should definitely come back to that before we're done. So tell us a little bit about UPMC and Pitt and your varied roles and responsibilities. I mentioned all of them purposely, but give our audience like a flavor for like all the different things that you end up doing. Yeah, yeah, I wear a lot of hats here. I mean, on the UPMC side, I have more of an advisory role and I obviously oversee our, you know, division, the computational pathology and informatics division in our department. and that includes multiple centers and various informatics frameworks. On the pit side, that part I do have both a strategic operational, educational role,

you know, in terms of how we are, you know, integrating AI into our pit landscape, especially in the health sciences. And obviously I'm starting in the School of Medicine. That's where I live. Yeah. And even though we are tackling various operational needs as well, we are starting off first and foremost with education piece. Because I'm a firm believer that if you're end users or even your developers, if they're not properly educated within the space, then the adoption rate is not going to be maximized. And so how do you adopt these?

How do you integrate these? Deploy them in the most responsible way. And that's, I'm a firm believer of the education piece. The other reason I feel the education is very important is because I'm also realistic about AI. I know AI, it's similar to what happened to us when Internet basically took over. It's going to have some challenges around upskilling. reskilling needs within, you know, the population as a whole, and even within our healthcare system, people becoming AI-enabled practitioners and staff.

I don't believe in AI replacing people. I'm a firm believer in AI being a tool that should be helping us and having human in the loop to, you know, it helping make us more efficient. but I'm also not naive. I know there are certain things that AI is starting to do that, you know, may seem as a replacement and some of which may become a replacement. But so because of that, I think it's our job to prepare our masses, regardless of which industry we're in, to make sure that that's upskilling and rescilling that's going to

be needed soon, if not now, gets done in the fastest way possible. because I think as opposed to the internet wave, the AI wave is coming at a much faster pace. I think there's a lot of positives, and I'm more on the positive side, you know, but I'm also realistic, and I think we should be sensitive to what people feel about how these things are affecting people's lives.

And I am optimistic that most of it, long term, especially, is going to be for the better. But I'm also realistic, just like anything new, I think it's, going to have a disruptive nature to it as well, too. Well, no, I appreciate your pragmatic approach. And it's actually refreshing to hear from a, you know, world-class AI leader in the clinical space that you have that balanced sort of view, right?

Because sometimes we get one extreme or the other. And I don't think either of those extremes help us. But I think your approach is super good. So we talk a little bit about UPMC and Pitt more specifically. But now let's go into one of your roles a little bit deeper. and that's, we'll call it C-PACE for now on, but that's the computational pathology and AI Center of Excellence. Tell us a little bit about C-PACE. Yeah, so C-PACE is, as you know, I do like, you know, starting new, you know, centers, and as we've done at Cleveland Clinic and,

you know, before that. So C-Pace is kind of like the latest, greatest on what we're trying to do to set ourselves up to be, you know, a world-class AI center within pathology and lab medicine, although we don't just tackle pathology lab medicine, you know, ventures now. We do things in surgery, internal medicine, operational needs. And given that, you know, we're all on the same page, you know, meaning our leadership in terms of what our, you know, what our needs are within, you know, pit as a whole, especially. Our goal is to have C-PACE becoming a major driver for various needs that we have within the PIT environment.

That could be, you know, operational tools that, you know, we need to integrate. It could be educational tools, research tools. So, and before I tell you more about C-PACE, I think it's important. important to realize that our strategy within Pitt and, and I believe the UPMC leadership are saying similar things, is a hybrid one, meaning that given that we do have the resources and the talent of being able to deliver a hybrid approach, we are going down the hybrid camp of things on a strategy thing for AI rather than a non-hybrid vendor-based only. So let me explain what I mean by hybrid.

I think most people are used to the traditional vendor-based AI frameworks where you get access to a known vendor. It could be OpenAI, it's chat GPT, or Google's Gemini or Anthropics Claude or any of the Microsoft or others. But regardless of which vendor you choose, reality is, you know, there's those challenges with vendor-based frameworks, you know. And we have great relationships with our vendors. We use lots of them.

We don't have just one. We have many of them that we actually have partnerships and work with them closely with. And that's great. I think whenever there is an ROI or return on an, investment that makes sense with your vendor partners, you pair up and you do the right things for your institution, for your organization moving forward. I think knowing that, you know, the vendor-based frameworks for certain tasks may actually cause cost and data security issues sometimes for us.

It's also important to have backup maybe home-brewd frameworks as when you do have that capability. And we do have that capability, especially through, you know, centers like C-PACE, where we have the talent set to be able to create these things. So that's what I meant by hybrid, is when there is an ROI with our vendor partners, we go with them, and that's great. But when we find out that we could become cost prohibitive in three years, five years, 10 years, whatever the trajectory is. Then we start planning and saying, okay, what kind of homebrewed tools that we could have internally that could drive that strategy forward as well. So that combination then gives us more flexibility and makes us also not only more financially nimble, but also data security-wise,

I would argue that we become more secure and have more flexibility there in terms of things that we would want to kind of deploy. Yeah. And I think, you know, in that argument, one thing we sometimes forget is how it helps you with recruitment and retention. 100%. Yeah, because you have all these smart people around you and they're going to find, they'll go to the vendor if they have to.

100%. They don't want to. They want to be with you in that local community in the region and doing great things. And you're giving them that opportunity. So it helps, it helps everyone. No, I'm glad you said that because you're 100% right. like our AI center now has, I mean, we have quite a few people now in our center. It's like over 30 people.

Yeah. And it's a combination of, you know, physician AI scientists, you know, computer PhD scientists, AI software engineers, administrators, students, PhD students, you know, fellows. postdocs, all sorts of people who are basically part of this journey. One of the top things that I think stands out when people come up to us, and I don't want to jinx us, but is that, you know, we have maintained most of our people, even though we're not a, we're only a couple of years old as a center, but most of our people are still here now, nearly all.

And I like to think that we also have a really inclusive. culture here that, you know, we want to make sure people are having fun. We want to make sure that they feel they're valued. And we don't want to do it based on just talk. We want to also do it through our actions. And so one of the things that we've done here is, and this is something, you know, my partner who drives this with me at C-Pace, Leron-Pantanowicz, who's our chair of our department. So it's really, with the help of Liron, Matthew Hanna, you know, Namtran, all of us feel like by giving ownership to everyone on things that are being developed, you then, you know, people have more loyalty around the environment. And so, for example,

we, I believe we've now have 10 filed patents or more now with. in the past year. And I have to say, we don't have the traditional patent filings that others have, where they exclude some of the early developers because they feel like the hierarchy might not be there and they should not be part of the thing and only faculty should be included. We don't have that. If you were an actual contributor and you contributed to the project and it was valuable contributions that led to the final product, you.

You should be on there. You should be listed on there. And I think that brings on value to a lot of the people who are part of this process. And so I think that's part of the culture that we've built around here. Yeah, no, no, it's fascinating. You know, we touched on this a couple of times, surface level. But we talk about medical staffs. Are they generally prepared for AI and how much education change?

You brought that up a couple times now, kind of the education. Yeah. So I think most medical staff, just students, you know, staff, faculty, I think even if they feel like they're ready, they're not ready. I think there's a lot of subtleties there in AI. And obviously different AIs are different. You know, people automatically bundle all of them under one thing. We've been using AI frameworks for decades. And so I think your traditional.

let's say cancer predictor tools or sepsis predictor tools or readmission predictors, those we've been using for a long time. And I think most end users know the performance measures and how to evaluate them, how to assess a vendor, how to basically build something as a homebrewd frameworks. Versus if you built something on the Gen A.I. A.I. A multi-agentic frameworks where we don't have the same performance measures, I think most end users don't understand the limitations of those frameworks since they're probabilistic approaches that spit out, you know, let's say next words, for example.

And so because of that, I think it's very important for people to understand limitations through education and also for us to set the right pathway forward as people tackle this AI space. So I quote one of my friends and colleagues, Dr. Rosenstock, Jason Rosentock, who's our Dean of Education. Him and I, when we were doing our early AI series for the students, he came up with this mnemonic for responsible use of AI so that it aligns well with our policies within the university. And I love this mnemonic because it just makes it easy for everyone to remember, regardless of which AI tool you use. Obviously, he was mainly talking mostly around generative AI, but this mnemonic was just a simple DVP mnemonic. So D for disclose,

V for verify, and P for protect. And if you just follow these three words, DVP, or disclose, verify, and protect, our point is, if you are using AI, you need to be honest. And disclose it. You know, this whole idea that people have that sometimes they don't disclose and act like it's their stuff that they made. I think that takes away from, you know, the legitimacy of what they're providing at their final product. So I think people need to be honest. They need to disclose. This is just another tool that they've used it for what they need to do. And when it serves them the way it needs to serve them. And then V, you have to vary. You have to very the output of these things, especially the text-based frameworks. If this thing is spitting out

some answer, you can't automatically accept its answer. We know that a subset of these things are incorrect, these responses. And so given that we know there is no perfect AI framework out there in the world, it just doesn't exist. And given that they all have some error rate, especially within medicine, it's so important for us to be able to verify this, Because imagine what kind of a, you know, harmful effect you may have if you take it at face value. Because as opposed to like a Netflix recommendation or something from Amazon or a basic movie recommendation or something like that, if they make the wrong recommendation to you or what purchasing power recommendation that you may have, the wrong recommendation has usually very little harm, you know, versus in medicine, the wrong recommendation could,

hurt or sometimes even kill people. So it's very important to verify the information as you tackle this and make sure that the information is based on legitimate sources, that you verified it because you're putting your name on there at the end of the day. And then finally, P for Protect, because we're dealing with sensitive data and some of it also may violate institutional policies. You want to make sure that you're following, you know, the institutional policies to that you're not violating HIPAA

and you're basically being protective of the data that's being uploaded, especially for vendor-based frameworks as they're being used. Because if you don't, you then could get people in trouble, you know, and yourself in trouble for that matter. So DVP, I love it.

I always tell, follow the DVP. I think it saves you 95% of the time. Super, super practical. Speaking of super practical, I imagine 90% of listeners are like, oh my gosh, I really am behind. What can I do? Give us just one simple, practical thing we could do to maybe improve our AI literacy. Yeah, that's a good point.

So I would say, obviously, I'm biased because it's our own, and you know about this, it's our seven-part review article series. So this was, so modern pathology, which is the, official journal of the United States and Canadian Academy of Pathology, as part of our education and AI series globally, we put out this seven-part review article series, which is free to the world. It's open access, and please share the link with the audience. And in this series, we did on purpose, we recruited about 40 AI national AI experts from across various institutions. Of course, it was driven by Pitt, but it included our old institution, Cleveland Clinic, Mayo,

a lot of other ones like it. Long story short, the goal and the focus of that series was educating the masses. And educating the masses without any heavy code, no code at all, as a matter of fact. No heavy math. It was just conceptual of for you to understand the capabilities of these tools, but also their limitations. And just to let the audience know what the seven parts series basically entails, it starts off with a very basic, you know, glossary of terms of several hundred,

you know, of the most commonly used AI terminologies with examples in our case in healthcare. world. And then big picture, what is generative AI? What is non-generative predictive analytics AI? And then the second part is a deeper dive on generative AI, where it shows you what is chat GPT? How does it work? How do you build a rag model or retrieval augmented generation type model? So how do you customize AI tools and also on text, images, so on multimodal, so on multimodal, so on so forth. And then the third one goes into non-generative AI for like things like classification, like supervised learning frameworks, unsupervised learning frameworks. And so deeper dive into the traditional machine learning studies. And then the fourth one is around the regulatory

aspects of AI and how does the FDA sees them? How does the European body seize them and other country sees them. So dives in deep with that. And then the fifth one is around study design best practices and statistical aspects, performance measures of both generative AI and non-generative AI and how do they similar? How are they different? All of that. And then the sixth one is around bias consideration and, you know, ethical and sustainability aspects of AI. You know, how do you actually make sure that the AI minimizes these inadvertent effects in there. And then finally, the seventh one is around, you know, multi-agentic frameworks, machine learning operations.

How do you deploy them the best way? How do you keep track of them as you basically deploy them? So hopefully that gives the audience a good framework of, you know, what is it that these things can do? But just as importantly, what is it that they can't? not do and they're limited by. Yeah, no, it's super practical. We'll definitely throw down the link. Human, this has been

super, super fun and interesting. I wish we had more time. We start off with songs on your playlist and brought back some good oldies like Fly Like an Eagle by Steve Miller Band. You know, kind of your philosophy, being a good human and just, you know, building others up and making

positive impacts wherever you are, locally, globally. I talk about your growing up and speaking farcey and some good of talent, good Italian. And then we talk about pivotal moments and how you got into healthcare and people who helped you along the way. And we talk a lot about PIT and C-PAS, C-PACE, specifically, all the great things that

you're doing. And I think we got pretty practical on things that people who are listening can do to improve their own sort of AI understanding and make a difference in the world. And it was super helpful to have the DVP acronym that you mentioned about disclose, verify, protect when it comes to the AI. utilization and a lot about upskilling. What did we miss? Or is there anything you want to double down on? I'll give you the last word. No, I think the only thing that we haven't talked about is, specifically also within the healthcare, why do we think that AI education is not just something

nice to have, but essential to have for the society as a whole? And if you're wondering why I'm so passionate about this is I'll give you the classic story I tell people around typical multidisciplinary conferences. The classic one is a tumor board that we do in medicine, right? So I think maybe some of the audience know and some of them don't know that if somebody has, let's say, a particular tumor, you know, the way that we currently manage tumors in medicine is we have these multidisciplinary conferences called tumor boards and where people discuss their individual subspecialties, you know, capabilities, but also limitations. And we do

that on purpose so that we can figure out collectively as a consensus between the pathologist, radiologist, oncologist, the surgeon, the social worker to try to make sure that the patient is, you know, compliant with the chemo and whatnot. The pharmacist, the nutritionist, everybody's at the table, usually, at these tumor boards, and they're putting in their two cents about how to best fine-tune the therapy, prognosis prediction, and final tracking of the patient,

of make sure that we're giving the best care possible to the patients that we serve, which is fantastic. I think this is the modern era of how we manage patients, and this is how we should do them. But there is one piece missing right now in these multidisciplinary conferences right now. And this is true globally, by the way. So this is not something that's missing just in U.S. And that is nearly all of these conferences are missing somebody who's well-versed in AI.

And the reason I think it's important is because as more and more of these AI tools are making predictions and are influencing, you know, particular clinical decision outcomes, it will be very important to also have an AI enabled practitioner at these multidisciplinary conferences who can also put in the two cents about the limitations of some of these models. And that may change the way. So you need that additional patient advocate from a patient safety aspect to be able to drive better care. And I think that's why education is so key. I think the more we educate, the more people basically become well-versed, the more of these AI champions that we have at the table. These AI champions can be serving in these multidisciplinary conferences whenever they're present there. They can help with vendor AI frameworks that the institution is purchasing. Are they purchasing the right stuff?

Do they understand the limitations of them? Do they understand of the homebrewd frameworks that are basically coming to table? So yeah, lots of that. So that's my pitch is why it's so essential is because we have to continuously find ways to improve patient care day and day out. And as long as we're on that continual patient improvement, quality path, we're then doing the right things. Dr. Human Rashidi, perfect way to end this episode of Digital Voices. Thank you so much for being my guest. Thank you so much, Ed.

Thank you. Thank you for listening to Digital Voices. We hope today's conversation sparked ideas, reflection, and connection. Subscribe on YouTube, Apple, and Spotify podcast so you don't miss an episode. Thank you.