Hi, everyone. This is Kedar Amate, co-founder and chief medical officer of Qualified Health. Thank you for tuning in to the Becker's Healthcare Podcast series. I'm really excited today to be talking to Dr. Nabil Safdar, the chief AI officer of Emory Healthcare and Emory University. Today, we're going to be talking about change management and how to help with AI adoption in our healthcare institutions. Nabil, thank you for joining me here today. Yeah, thank you, Kedar. I'm looking forward to it. Well, Nabil, you and I have had a chance to work together a little bit between Qualified and Emory in the partnership that we have. But let's start by naming an AI deployment that you've been particularly excited about, particularly proud of at Emory. What's worked about that? What's made it successful? And then we can get into what made it challenging. Yeah, Kedar, great question. Plenty that I'm proud of, plenty that I'm ashamed of as well. But I'm going to go back to one of my first AI deployments, which was about 11 years ago at Emory. This is now considered pretty routine, but CT perfusion for stroke imaging is a pretty common thing now at most stroke centers. The brain attack teams, the stroke teams, neurologists, neurocritical care docs, neuroradiologists like to use this, especially since we know brain is time. But 11 years ago, this was kind of like two guys in a research project just started a company and we were actually implementing CT perfusion in research mode only. And I remember there was a neurologist who emailed me and I had a conversation with him and said, can we please use this clinically rather than just in research mode? And I was new to the organization. I thought to myself, geez, why don't we use it clinically? And what I discovered was just there was no contract to pay for the darn solution. You know, it was available. Just no, there was a divide culturally between the physicians who wanted to use the tools and the administrators who knew how to get that contract through legal, through security review, all those types of things. And I said, well, I could figure this out. Like I started calling around administrators. They're like, yeah, that's not even that expensive. Let's just do it. So we implemented it officially and all that changed was there was no longer the little for research use only, you know, label that goes on to the AI. But I learned a lot of things from that. One is we hamstring ourselves unnecessarily by barriers to adoption that don't really exist. It's just that we're not willing to, you know, take a little bit of risk. I was new to the organization, so I could ask stupid questions like, who pays for this thing? Who's our lawyer? And that helped. And then out of that, what happened was, it was useful for the stroke team. It was useful for patients. I think the evidence bears that out. But we also found that there were many ways that it was not useful. And some of our team actually published in a paper led by Chung on all the ways the CT perfusion can be misused and it can go wrong, the artifacts, all the way it goes wrong. And then, you know, 10 years later, in that same group, there's a whole paper on AI governance in that department, you know. So these things tend to have a life of their own. There tends to be a flywheel effect, like little breaking down of barriers compound over time so that now that seems like a pretty routine thing. So I was, look, was it a major technical breakthrough? No, absolutely not. But for me, it was like the first time I was like, I'm going to ask some dumb questions and let's figure out if we can do it. And I was, I'm very proud of that. Maybe good guidance for how to, how to approach AI implementations at scale in almost any organization. But tell me, tell me a little bit about what has been a, you know, a big surprise for you as you've been now tracking many projects as they've gone through your system. What surprised you about receptiveness of staff team members to artificial intelligence technologies? We hear a lot about health systems being, or healthcare workers being particularly nervous about it, worried about job security, worried about trusting the accuracy of these technologies. How have staff responded to the technologies that you and your team are trying to deploy, Debra? Yeah, great question. I'll tell you what surprised me. This shouldn't have surprised me, but I think it speaks to the circles that we all travel in. And it has to do with culture. I just kind of assumed that, hey, everyone's talking about AI. This is the latest greatest thing ever Once I started using it I think I lost like 48 hours of my life I was like this is amazing The generative AI stuff you know it reminded me of when I had my first Apple IIe and I had to, you know, program things in basic or load up those five inch floppy disks. I felt like so smart because I was like, oh, it's not like the movies. You have to give it very specific instructions in order for it to do. It only knows what you tell it. And, I don't know, 50 years later, 40 years later, you know, now I was like, oh, my God, what I imagined as a child computers would do in science fiction, they're able to do. So I had a very, I think, intuitive understanding, maybe built on some experience of what Gen.AI tools can do and how to interact with them. I mean, you got to take that with a grain of salt because everybody's new to this game. But, you know, we were early in the games and the circles that I traveled in were similar, right? The people that I spoke to were similar. Even the folks who were not, you know, ostensibly tech, you know, nerds were using the tools and they were curious. But when we released, you know, one kind of rag architecture chatbot that looked at policies to nurses at an entire hospital, You know, one of the first things that our team recognized was when we were when the users were asking the AI chatbot questions, they were interacting with it like it was Google. They were entering keywords rather than prompts. And that just never occurred to me that you would not know that somebody might not know that this isn't a search engine. This is a prompt engine. It's a chatbot. Did you have to educate them differently? Did you have to change your approach? Absolutely. We had to go through and start to give, you know, by analogy, hey, this is like you're talking to a person, give it a specific question, you know, talking a little bit about prompt strategies, some upskilling, some education. I'm not blaming the nurses, by the way. That was total, like, you know, blind spot on my team's part and my part. You know what I mean? We pivoted. We started giving them much more basic information. Why would a busy nurse? I mean, unless they're in that 20% of early adopters, bleeding edge folks who are trying this at home, they're busy, right? They've got busy lives. They've got busy jobs. And this is one more thing we're asking them to do. Of course, they're going to use the common framework, a common paradigm. It reminds me, Zainab Tufekci gave a talk recently at Emory, and she talked about when horse carriages were first being replaced. one of the early models of cars actually had a faux horse head on top of it so that it would be a familiar, you know, paradigm for the drivers and also for other horses, which was the majority transportation mode at that time. And I felt like that's what these users were doing. It was my fault for not recognizing that not every, this is a journey, right? And everybody's a different place. You have to meet people where they are. Yeah, that's an interesting point. Giving this idea of making the AI familiar, the horse, placing the horse kind of idea. Do you think we have to actually meter the dose of AI into our health systems? Let me explain what I mean by that. Not do the most radical, Rosie from the Jetsons kind of solution right away, but instead build our confidence slowly. in the technology through what I'm calling meter dosing, like small, almost micro dosing kind of the artificial intelligence into our systems, especially with clinicians kind of building their trust and their confidence as they see ambient documentation successes stacking against, helping with prior authorization, stacking against, et cetera, et cetera, and then moving towards more complex and more wholesale use cases. Do you think this is part of the strategy for winning clinicians over and building trust in the technology? Yeah, man, so much to talk about here. I don't know where to start. So, you know, we have done a couple of projects, what we call at Emory, and I'll credit my colleague and friend Rachel Silverman with this, like with the YOLO approach. Just big bang, turn it on for everybody, very little training. We'll see what sticks. And in other cases, we have done like, you know, death by a thousand pilots where it's like, okay, you only get five users and now there's another 10, 15 users. And I'll tell you what I've seen. You know, Kate, we've done stuff like this before. You know, I like car and travel analogies, right? So, you know, there's a certain capacity that a highway has, right? And if you overwhelm the capacity of that highway with new traffic, it's just going to reject the system and people are going to go around it. Right. Yeah. So that's the YOLO approach. And that's what we found. You know what I mean We said turn this tool on for everybody Everybody use it We even did some kind of rah leadership moments And we plateaued around like 30 utilization You know what I mean? Now we're going to go back and try to do more. It was a cheap 30% adoption, meaning like it didn't cost us a lot to just turn it on and say, go for it. And you get a certain amount of adoption because it's a helpful technology. I do think metering it in is valuable. And it depends, right? So, you know, for the most resistant users, those who are feeling maybe the most threatened because a high percentage of their job can be automated as tasks that are automated. Not the entirety of their job, but a certain portion of it. Those people that have had bad experiences with previous iterations of AI, let's call it, maybe it was, you know, NLP two generations ago or something like that, you know. So for them, I think metering it in is the only way that you avoid a total immune rejection of this new technology, right? And you got to do it. It's almost like allergy shots, you know? But if something is so compelling and so useful that it speaks for itself, you know what I mean? Like people will see it and they'll say, why don't I have it? Then I think you start YOLO. You know, you only live once. Let's give it all to them and we'll see who adopts it. then we'll come and get the laggards, you know. So, you know, I like to commit to one or the other. So it's a bit of toggling between different strategies, depending on the nature of the environment. It makes total sense. And I love our blending of car metaphors and biology, what we do together on some level. But maybe as I come close to the end of our time here, you know, I think you and I agree that there's going to be lots of different solutions necessary for a health system. It's not going to be one thing. You know, AI, there's so many different potential applications, maybe at least dozens, if not hundreds of potential application areas and things we're going to need to implement. So how do you create a, you know, change ready culture in an organization? This is not going to be one time with one big thing or even a couple of small things. This is going to be a constant for us for the next, you know, at least five to 10 years of many, many things coming. How do you prepare an organization like Emory for this constant pace of change with these technologies coming online? Yeah, you know, I do think that we're seeing an acceleration of availability of new technology driven by AI that really is unlike what I've seen in previous parts of my career. and others share that, like the availability of new technology rapidly, new solutions, you know, the velocity of it, you know, the volume of it, kind of like the big data aspect of, you know, how much there is, is something really new. I think, you know, we're going to have to figure out how we manage this change. I like the persuasion, you know, kind of literature, especially Caldini. And Caldini talks about a couple of levers of influence. One of them is authority and social proof, or there's authority from top-down leadership. There's social proof from people who have been through it before, who have validity with their peers, but particularly commitment and consistency. You know, I'm thinking about downtimes. If you asked me, hey, does Emory have a great approach to how it manages expected and unexpected downtimes five or six years ago, I would be like, I don't even know that we have an approach. But now it's a totally different ballgame. And I'm thinking about how we have had expected downtimes from major upgrades, and we've had unexpected downtimes like the crowd strike or some of these other things that have come. And I'll tell you, it's a different world now. We have a team of people who may be analysts who are consistent. We have administrators who are consistent. We have stand-up, incident command center policies and procedures, which are very consistent. And so we have established culture when it comes to downtime. People know what their roles are. They know what the expectations are. And mostly that's driven from consistency. In healthcare, we all get a little bit cynical and we start to say, is this the latest fad, the latest trend? Is it going to blow over in another two months? Let me wait and see. But if it's there consistently, and then you also leverage all your levers, reciprocity, scarcity, liking, social proof, authority. But that consistency and that commitment is what changes culture over time. Because look, we're dealing with literally tens of thousands of people from all different walks of life, different levels of education, different backgrounds, cultural backgrounds even languages So how do you create a consistent common culture It what you tolerate and what you expect in behavior and that people only learn that by seeing that over and over again And so that what we got to do with AI. We've got to have consistent rollouts, consistent communication, consistent support from top-down leadership, consistent engagement of the front lines. And then it will slowly build that culture over time. I like that advice, consistency and commitment. That makes a ton of sense. Maybe I can ask you a final question just in rapid fire here as we get to the end, Nabil. If you could give one piece of advice to any other health system that's trying to embark on this journey on adopting generative AI in their environments, what's the one quick fire piece of advice you'd give? Yeah, my quick fire advice is do not relegate your AI strategy to be a service that only responds to tickets. And, you know, if it's really becomes, hey, I've got something broken, I need you to fix it. Or it becomes a thousand requests from various departments and business units. You know, look, we need to be responsive to the front lines. Absolutely, we should do that. But there needs to be capacity reserved for top-down transformational AI strategy. And that only comes if you position that conversation at the right level in the organization. So it can't just be operational AI governance. It can't just be security, risk, compliance, legal counsel, governance. It's got to be true corporate governance when it comes to AI, which isn't about what we do or how we do it. It's about who we are as an organization. And that means having AI embedded in those corporate discussions, in those small group discussions. This isn't just self-serving. I don't have to be personally in that conversation, but that conversation has to be happening at those top levels. Fully agree. This is exactly how we've been pursuing AI transformations and health system partners around the country. Absolutely fully embedded at the top level of leadership and with organizations that see AI as part of who they are going forward. It makes a ton of sense. Last word, Nabil, Chief AI Officer's new role in health systems. You are one of them at a major institution. What do you think is going to make your fellow CAIOs successful in this new role? Yeah, I'm going to use another car analogy here, Kedar. So with cars, you have, and let's talk about a racing team or something like that. You know, with a racing team F1 or rally, whatever it may be, you have your mechanics and those would be akin to the like really techies, you know, chief AI officers or chief data analytics officers playing that role. Great. They're a part of the team. You have the drivers who are actually sitting there in the cockpit racing their cars or the rally cars or motorcycles, whatever it may be. Great. They get to say that we, you know, we actually made the decisions at the last minute at the front line. Then you have kind of the safety officials sitting on the side, like with the green flags, the checkered flags, the flags that slow down. And then you have the folks who are actually the marketing team, like all the people selling the advertisements on the car, right? At the end of the day, if you're a chief A officer and you fill one of these particular profiles, you're the mechanic, meaning you're very technical. You're the salesperson, meaning you've got all the advertisements and all the connections and you're a networker, right? You're the driver, meaning you're the operator. You've been in there, you've deployed, or you're the safety official. You're a governance type. You know, you're a kind of ethics and compliance type. It doesn't matter. What matters is you get to the finish line fast enough and far enough. So we can stop counting, frankly, how many AI tools we've deployed. I think that's cute. I do it myself, so it's not a criticism to other systems that tout like, hey, we've deployed 200 AIs. Everybody's deployed 200 AIs. It's impossible to count them. They're embedded. They're like the oxygen we breathe. They're like the electricity we use. And what matters is, did you get outcomes? And so five years from now, we're going to be judged on what outcomes, measurable or less tangible, soft or hard, doesn't matter. whether it's green dollars, whether it is time saved, whether it's satisfaction, did you contribute to? Because nobody really cares if you're the best mechanic in the world if in the race your car didn't win. Totally agree. You're helping us understand the outcomes that are going to matter most, and we're going to see in five years' time. We'll have another interview, Nabil, five years from now. We'll see how you're doing against that performance report card as you're thinking about it. But Dr. Nabil Safdar, thanks for joining me here on the Becker's Podcast. really a pleasure to have you and a pleasure to work with you every day. Likewise, thank you, Kedar.
