Who Teaches the Machine? AI and the Future of Precision Education

Medicine has embraced personalization. Every day, clinicians tailor treatments, medications, and interventions to the needs of individual patients. Yet medical education often remains rooted in standardized pathways, where learners progress through the same curriculum despite arriving with different experiences, strengths, and learning needs.

In this episode of The Precision Educator, Dr. Larry Chu and Dr. Viji Kurup are joined by Dr. Rishi Kadakia of UCSF to explore how artificial intelligence may help bridge this gap and advance the vision of precision education.

The conversation examines the limitations of one-size-fits-all training and the challenge of understanding what learners truly need in order to grow. Dr. Kadakia shares his work using AI-enabled educational tools to support learner preparation, identify knowledge gaps, and create more individualized learning experiences. Together, the guests explore how data generated through learner interactions may provide educators with new ways to understand learner development and tailor educational support.

A central theme is AI literacy. As large language models become increasingly accessible to both faculty and trainees, how can educators ensure these tools are used thoughtfully and responsibly? The discussion explores the opportunities AI creates for learning, as well as the risks of overreliance, false confidence, and the gradual erosion of critical thinking and clinical reasoning skills.

The episode also highlights the importance of maintaining a “human in the loop” approach. While AI can help organize information, identify patterns, and expand access to educational resources, it cannot replace the contextual judgment, mentorship, and human understanding that define effective teaching. From subtle shifts in patient behavior to the complexities of clinical decision-making, many of the most important educational and clinical signals remain deeply human.

Ultimately, this conversation reframes AI not as a destination, but as a tool. The future of precision education will depend not only on technological innovation, but on educators who are willing to understand these tools, shape their use, and ensure they serve the goals of learning, professional development, and patient care.

Key takeaways from this episode:

• Why personalized learning remains one of the greatest challenges in medical education
• How AI may help educators better understand learner needs and support individualized development
• The growing importance of AI literacy for faculty, trainees, and educational leaders
• Why maintaining a “human in the loop” is essential for safe and effective AI implementation
• How educators can help shape the future of AI rather than simply react to it

Especially useful for:

Medical educators, residency and fellowship leaders, clinician-educators, faculty developers, medical students, residents, and anyone interested in the intersection of artificial intelligence, learning, assessment, and precision education.

Related episodes:

For an introduction to the foundational principles behind this series, start with Episode 1: What Is Precision Education? Rethinking How Physicians Learn.

For a deeper exploration of coaching as a core mechanism of precision education, listen to Episode 2: Talk Less, Listen More: Coaching as Precision Education.

For a systems-level perspective on how data can inform learning trajectories and assessment, explore Episode 3: When Data Becomes a Coach: Rethinking Assessment, Coaching, and Learning Trajectories.

For insights into how educational leaders balance data, trust, and learner development, listen to Episode 4: From Systems to Signals: Precision Education for Program Directors.

Additional Reading

1. Budzyń K, Romańczyk M, Kitala D et al. Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study The Lancet Gastroenterology & Hepatology, 2025; 10, 896-903. Link.
2. Teach (Teaching Educators AI Competency Hands-On), Link.

Transcript

Larry Chu, MD 0:09

Welcome back to The Precision Educator. I'm Larry Chu, joined as always by my colleague and co-chair of the Society for Education in Anesthesiology's Precision Education Task Force, Dr. VG Kuro.

Viji Kurup, MD 0:24

Hi Larry, great to be back with yet another exciting episode for our podcast. So one of the biggest challenges I feel for medical education is this dichotomy that we have between how we perform in our clinical role and how we perform in the education side. So in clinical practice, we live in a world of moment-to-moment personalization, the dosing, physiology, hemodynamics, pain plans, all of these are kind of tailored to the patients, right? But in medical education, we still often teach groups of learners as though they are one homogeneous mass, that they all learn the same way at the same pace with the same needs. So, you know, I think that that's something that we could explore today with our guests.

Larry Chu, MD 1:15

So today we're exploring the question that many of our SEA members have been asking, have been exploring, and it's the theme of the work that we're doing in the precision education task force, which is what really does personalized mean in education and in precision education? And how could AI, for instance, help anesthesiology learners develop better expertise? How could it empower us as educators to better deliver precision education? Today we're joined by Dr. Rishi Kadakia from UCSF. Dr. Kadakia, welcome.

Rishi Kadakia, MD 1:58

Thank you, Larry. Thank you, Viji, for both having me. It's an honor to be here, and thank you for just your initiative in this area. It's uh a much needed one.

Larry Chu, MD 2:06

Welcome, welcome. So, you know, one of the things that we've been exploring and discovering in this podcast series is taking a look at where we are today and where we want to go, and looking at this arc of one size fits all of education, and thinking about how we can begin to tailor what we do to the individual. So, with that in mind, what do you see as the limitations that we're facing right now as educators in this model of one size fits all education?

Rishi Kadakia, MD 2:49

Yeah, thank you, Larry. Uh that's a loaded question. And I think uh when we think about where we sit in the midst of sort of precision education and how do we take a learner's individualized needs and factor that into what we deliver on the next iteration of their education, for me, it comes down to data. And that certainly I know that is also equally one of the most challenging aspects of all of this is how do we collect the data, how do we decide which data is worth utilizing in the data set, how do we label it, and then how do we use it on the other side? So there's a lot of questions. And I think where AI and large language models can really play a role in this is both the data collection, how we label it, and then potentially also the delivery system. And currently my interest in my area is really focused on the data collection and then very early, early sort of work in the implementation side as well.

Viji Kurup, MD 3:42

Yeah, that that makes sense because uh with the current system that we have, it's you know, you could have two residents who've finished the same rotation and come out at the other end with very different experiences. And we don't know if each of those residents have hit the milestones that we, you know, that we've set out for them. And that's variability, unless we are really tracking it, like you said, is not visible to the learner and to the teacher, right?

Rishi Kadakia, MD 4:16

Absolutely. And I think, you know, just to give a little bit of background, I'm a pediatric cardiac anesthesiologist. And when I think about what I do in the scope of both anesthesia and also in medicine, sometimes I feel like we're in a dimly lit corner here where we have some of the rarest forms of diseases, and they're ultra-concentrated to just a few operating rooms in a very specific location in the Bay Area because UCSF is sprawling. It's not just one location anymore. And so I think about those challenges as well. And how do we ensure that the learner at different sites, of different backgrounds, different prior experiences ultimately gets the experience that we want as educators?

Larry Chu, MD 4:55

I think one of the interesting frames is also to I think put in context and perspective what we mean by data too, because I think especially for a lot of our listeners, when we say data, sometimes it can seem abstract and maybe our eyes glaze over and we don't really understand what that means. But I mean, we're collecting data all the time as educators. It's it's what we see, it's the observations that we make every day when we work with learners. It comes down to interactions that we have and how they're making decisions in the operating room that we observe, it's how they interact with patients, it's how they explain their clinical reasoning to us in the moment after the case is over and we're debriefing them. I think what's interesting and maybe what you're describing and you'll talk about today is how in 2026 and maybe going forward we're able to capture those moments that maybe we as educators are observing, but we can we can now capture those better and maybe across educators, across timelines, across sites in ways that we haven't before. And what if we are able to do that? What can we do with that information? And how can we use that information in ways to personalize the educational experience better?

Rishi Kadakia, MD 6:26

Absolutely, I fully agree, Larry. And um, when I think about your last podcast and what Dr. Mahoney said about choosing to take either the whole bread or taking half of it, um, when we think about, you know, how do we tackle this, right? Data is, as you said, very nebulous on its own as a concept, and that it's just one part of it, it's even bigger picture. And so certainly we do take little bite-sized nibblets of this and decide, you know, where do we want to start? And then hopefully the points that we do start can have sort of learning points in themselves that we can extrapolate that to sort of the larger areas. And just for example, currently, I'm really focused on sort of the scaffolding phase of the educator's experience, the learner's experience. I apologize. Um, and so when we think about how does a learner prepare for a case before they come into the operating room so that when they are in the operating room, rather than learning for the first time how to manage hypotension, they've at least conceptualized it prior to coming in to the OR the night before. And so then perhaps then when we think about, okay, there's these scaffolding events that occur the day prior to being in the operating room, potentially there are opportunities there to start collecting data and take that and then provide a better experience the next time. So that is obviously just one area. There's obviously many more, including the interoperative and potentially the post-operative experience. Put all this data together to create a better profile for our learners.

Viji Kurup, MD 7:48

Yeah. So that's one example. Uh, Rishi, when you're thinking about personalized learning and beyond for our learners, what's the picture that comes to your head of how you're going to create this? How does it look like?

Rishi Kadakia, MD 8:04

Yeah, that's a great point. And again, this are so many ways to attack this. And when I think about our specialty and what our learners in pediatric cardiac anesthesia tend to go through, uh, the very first thing in the way, sort of describe it visually, is sort of a bunch of hikers approaching Mount Everest. And there is just this insurmountable task coming up tomorrow where they're going to be for the very first time in the Pete's cardiac OR. And there's all this knowledge and uh underlying information that goes into being able to actually have an educational experience in the first place. And so I'm thinking about that moment, the learners encountering Peds anesthesia or Peds cardiac anesthesia for the first time. So, how do we make it approachable? How do we allow them to come in with some basic subset of information that they can then build upon it and actually have a beneficial experience in the operating room? So that's where we started. And when I think about that, it is contextualization to a large degree. So when we have a learner engage with our platform for the first time, we do collect some basic information about their background, their learner type. And from that moment on, we're building sort of a starting point for their knowledge. And then from there, as they interact with the platform and how they interact, the types of questions that they ask, we can categorize them. Is this a knowledge-based question? Is this critical thinking or clinical reasoning? Are they asking about decision making? Are they asking about drug dosing or what drug should I use for an induction? And so then we can sort of get an idea of where the learner sits. And there's still so much more beyond that in terms of what we can collect and what we can derive from their attempts at learning to then factor into the overall larger learning experience. And like I said, I think I'm just capturing a very tiny sliver of everything. But even in that in itself, is very a very large task. Um we think about it as, you know, when Larry talked a little bit about data and what does it mean to someone and how do you collect data, how you decide what data is important, I think it's a little bit of a like cyclical logic. You think that this is important, you collect the data, and then it turns out actually it's not very useful when it comes to outcomes or learning. Um you want to go back and you go back and you make make a mega modification in your data collection strategy. And I think that's no different in the world of AI, and we're gonna decide what we need to collect and avoid noise and sort of strengthen our data collection for things that are that are important. And I think we just need to start doing it, and then that's how we're gonna get the cycle started.

Larry Chu, MD 10:28

You know, let's talk a little bit about AI, because I also think that's another term like data that gets thrown around a lot, and and maybe educators don't really know what it means. And I know that there's a lot of excitement around it, and people talk about the potential of our artificial intelligence in so many different domains, not just education. I also think there's a perception that people think that you just throw it's just like you know, sauce, you throw it on and it's just gonna make everything better. I think something that would be really interesting to talk about really is how do you see artificial intelligence being thoughtfully used here, at least in your experience, how have you implemented it in ways that have helped you deliver a better educational experience? And maybe what have you seen?

Rishi Kadakia, MD 11:28

Yeah, absolutely. Uh what I'll also mention is that we're a slightly different, smaller hat, faculty development, where we're working on literacy program pilot for faculty educators who are just as you described, I mean AI is an emotional experience for most people. Some people think that uh some people believe that it's going to uh, you know, there's fear that they may replace physicians or at least components of the patient-physician relationship. Um, others, as you said, they they think it is this amazing tool and it can fix everything. It's the pasta sauce. So we started navigating sort of basic literacy back in September. And when we think about how do we use AI critically and safely, I do think literacy is a key component. We need to understand how these tools work. And it's no different than when pre-AI, when we think about evaluating research or evidence, right? We need to have at least a basic understanding of statistics so that we can evaluate the research and the data and make our own opinion about whether we trust what we read. And I don't think this is any different. And I think understanding, again, at a basic level how our different models operate, um, we can potentially, first of all, uh use them better. We can prompt more efficiently, then we can also be a bit more particular about the responses that we receive from AI. So we can decide when an AI is actually worth the praise and it's doing something really meaningful, or potentially it's introducing some error in our process. And so when we think about AI and how it may impact the educator experience, I think it does really start with our own literacy and an understanding of how AI operates. And then from there, we can slowly integrate it into our practice. And there's a very narrow window where AI can be useful, at least at this current moment in time, which is 2026. We use it too early in the sort of education process, and we run into a risk where we actually may misskill or never skill. Whereas if we use it too late, then it's a missed opportunity, or we may be using it redundantly. It's a tool that we don't need. Um, so I think first understanding how it works will allow us to sort of use it at that critical moment where it can actually serve as an advantage to our thinking and our reasoning.

Larry Chu, MD 13:50

So, as an example, is in 2026, is AI best used as generative AI to help us make our PowerPoint slides, to help us generate learning objectives for our talks? Is it best used to give feedback to our residents? Or should we be using it to make uh admissions decisions for our residency programs and be reading applications that we receive to our programs? Uh, you know, these are I think uh the questions a lot of educators have in in terms of, you know, where what is the state of the technology and and how could it best be used? And and uh I'm wondering in your experience, you know, what uses have you tried and where do you see the technology?

Rishi Kadakia, MD 14:42

Right. I think you've described uh a variety of scenarios that we've already seen sort of a leak or an implementation, early sort of strategy for AI implementation. Um but just to dial back, remember that these are large language models. So really their magic happens in sort of the communication, the language between you and this computer system. And all of those sort of applications are possible. But I think that we are still, it's very nascent, and I think we have a long way to go before we can really say that AI can sort of have a meaningful impact in any of those areas. And I think, like we mentioned, we need to start accepting that these tools will become uh more and more prevalent. And so, how do we navigate that scenario where we are using a tool that we don't completely understand? And we're using it in very high-stakes environments. You talk about medical admissions or patient care. And so it's it's less about can this tool do this, and more about, well, where are we in this process? You know, as experts, whether we're an education expert or a clinical expert, or maybe you're a school of medicine admissions expert. We need to ensure that as we use these tools, we do have a system in place to both monitor and also intervene. And that's essentially the concept of a human in the loop. And you can replace human with any sort of expert you want in that particular process. And a great example of this actually is uh not from education, but it's just from our, you know, one of the largest EMR systems in our country. Um we have had a pretty significant rollout of AI features. And um, one thing I would give it credit, um, even if the outputs are not what we want them to be as medical providers, is that it's very transparent and it's not uh a forced process. And both to the end user, so as a physician, say writing a patient message in the uh message and messages interface, you have the option to choose from a pre-generated AI response, or you can choose to write your own, or something a hybrid of both. So that's one level of sort of human in the loop. Two, on the administrator side, as this is a new feature rollout, there are people in the background who are also verifying and uh sort of checking that the humans that are deciding to use these tools are doing so in a in a relatively safe way. And so while we're working along the early implementation route, um having sort of strict monitoring and um the ability to intervene is crucial, um, as with any other technology that we implement for the first time.

Viji Kurup, MD 17:11

We talked a little bit about uh how educators are using these tools, but our learners are already using these, right? So they're using it and we don't we haven't set up any guardrails. They are beginning to use it in ways that they think is going to help them. So if you had to, you know, as a learner, use these tools, how do you envision uh it should be used?

Rishi Kadakia, MD 17:40

Right. You brought up a great sort of I almost want to call it a problem, but a challenge, I would say, that we have in this world. And unfortunately, like we think about what Google was 15 years ago, or Wikipedia was 10 years ago. And there are just some tools that are so incredibly accessible that it'll be always hard to completely eradicate sort of the unsafe use of those tools. I personally believe that one of the most helpful things we can do to sort of mitigate that particular problem is provide our own set of tools and make it as easy and usable as possible. So this is where uh sort of putting on my technology hat and computer science and coding background, it's important that we do make tools accessible, have have a user interface that is approachable, really do our best to limit the barriers of access to that tool. And so when I think about our platform for peace cardiac anesthesia, that was one of the sort of main things we were working on is creating a chat interface that was very easy to use and had some immediate value to the user, that they would choose to come back to that rather than using an open source tool. And really the secondary objective there is that those conversations can be monitored for safety. And so we have the ability then again to ensure that the information is correct and then we can intervene as educators and still be part of that discussion as we need to be, as it does involve at the end point our patients.

Larry Chu, MD 19:06

I think that the comment of human and loop, I think is really pertinent. I think as you were talking about the idea of that, you know, you have to consider if you're going to I think accept the advice of AI, I think that's a emerging skill that needs to be nurtured and developed. I'm also though thinking about all the times that that Claude or ChatGPT asks me if I should just say yes. And I'm just like, yes, yes, yes. And I I think of the studies that show that there is a de-skilling of physicians who use AI over time. There's a study that shows that gastroenterologists who use AI for detecting, um, I think it was a study that showed they use AI for detecting lesions and colonoscopies or some GI study, uh, they get de-skilled over time. And I know there was another study I read that looked at students who use AI for studying, for instance, they're falsely reassured of their competency and they actually score lower on exams just because of the reassuring tone of the chatbots. So I do think that what you said is very important. The idea that we have to cultivate and nurture uh I don't want to say a skepticism, but the idea of AI as a learning partner and a consciousness of using our own judgment, of keeping our our own. I I'm not sure the exact skill set here, but uh engagement with learning, engagement with the learning process as we use AI as a learning tool. And I think that skill is something that needs to be worked out more. I don't know if you've thought about that, Rishi, or if you've been developing uh work in that area, or what you think about it.

Rishi Kadakia, MD 21:07

Yeah, I think this again comes back to literacy. And first we need to start with the educators and the faculty and ensure that they know these concepts, that they understand that there are limitations to these tools. At the end of the day, as I mentioned, and I mentioned that for a reason, was that large language models rely on text and language to do their processing. But humans are actually language and text is a secondary objective. We actually use a lot of other inputs when we think about the way we reason. So we are actually fundamentally very different. And so we start there in that we are two very different systems. And we are still responsible for what the learners will take away from their clinical experience and eventually what happens to the patient. And so, first we inform our educators and our faculty, and that does uh need to sort of start high up at the university or the academic level, and that we ensure that the faculty that are using these tools have some sort of gate or some sort of educational priming process before they even use the tools themselves. And then we can sort of trickle down into sort of how do we ensure that the faculty members then invoke the same sort of guardrails and safety mechanisms on their trainees. And I think that's kind of the approach we've taken at our institution. And the project that I'm leading is really again focused on the faculty side, recognizing that the faculty are the ones that are going to be assessing the learners. And there was a paper that came out I think in October of last year in the NEGM that sort of outlined in a really nice way for even the uninitiated faculty who have not used AI extensively. And honestly there'll be a lot of faculty well they'll never catch up to our learners in terms of what they know. But still providing a framework to be able to have a conversation about AI and safe AI use, whether or not they actually fully understand it themselves. And that was the uh the deft model with the deft AI model, I apologize, that provides a sort of a rough framework and really it comes down to just awareness and having the discussion, right? Did you use AI for your patient presentation? Did you use it for your note? And then that opens the doors to discuss potential limitations of AI, biases. Perhaps they uh missed a potential differential in their sort of line of thinking because they use the tool. And it's not so much as to discourage the tool but a reminder that the tool can be helpful but it's not the only tool we have when we think about our responsibilities as physicians.

Viji Kurup, MD 23:39

Yeah and I'm as you're saying this I'm thinking about the paper that Larry authored and will probably be in publication very quickly where he talks about things that an AI cannot detect and cannot be used to sense the anxiety of a patient, to catch upon the hesitation that a patient has, to catch upon a subtle shift in the mood in the OR, right? Like so these are things that as humans it's easy to you know see and respond to it. But if we were relying on an AI to capture all of that in a record uh those that would not have been you know captured. So so there's a lot that is missed when we purely look at AI or any machine for some task.

Larry Chu, MD 24:34

I think uh the example so I was asked to write a chapter on AI and anesthesiology for an upcoming textbook for medical students on the use of AI. And I tried to position it in a way to say that human perspective and context is really irreplaceable. But I think one of the examples I think I used was and and you know I gave real examples of how there are AI algorithms currently that predict intraoperative hypotension 15 minutes before it occurs based on data that's being recorded and arterial lines. But the example I gave was that your intraoperative anesthesiologist listens to a change in the signs of the room and how the suction changed from intermittent to now it's just constant suctioning of blood. And then you know a couple minutes later the surgeon says we're having a little bleeding here. But you know the anesthesiologist has already by that time ordered blood opened up the fluids started administering intraoperative measures to counteract this ahead of time before the AI algorithm has even noted that you know something has happened with the vitals so the the point is as VG mentioned that as human beings we're capturing data right about the tone of the room the fact that the nurses are hovering a little closer to the surgical field that the surgeons aren't bantering as much are paying more close attention and maybe someday the AI will know I've given away too much now in my chapter and the AI is going to read it and it's gonna do these things now. I don't know but at least in 2026 human in the loop as you mentioned Rishi it's still really important.

Rishi Kadakia, MD 26:44

Yeah I couldn't agree more.

Viji Kurup, MD 26:46

Yeah I I had a question also Rishi you brought up AI literacy right so this it's not a one time and done deal right like this should be an ongoing thing that the tools are evolving so quickly. So how do you see that the literacy is gonna keep up with the evolution of these tools?

Rishi Kadakia, MD 27:09

Yeah Beijy that's a really tough question because I think in its current state I think we've hit sort of like a reliable rate of change at the moment like when we first talked about AI sort of exploding at the start of 2025 it was extremely disruptive and we were really starting from scratch. And I feel like now this in this current stretch we're very iterative so we have time to catch up and sort of understand how all these things work. But your question is very apt and then what about the next generation where there's maybe it's no longer about typing things into a chat box, but there's now a camera and maybe there's a sound sensor and so now the suctioning can be heard by the LM. And so there is really not in my opinion too much we can do except understand what we can understand. And in this moment I feel like we're on a stretch where we can sit back and sort of learn how these tools work. One, by reading about them, but two also by using them and experimenting. So there's no better way to determine how great an LLM can be or how terrible it can be until you sit there and try to write a practice question and think like well at the end of the day this needs to be a high caliber question. I need to adhere to say the MBME item writing guidelines or use a reliable CBL format, whatever you decide to do, we are still the evaluators but I think we also need to be able to start using the tools and learn about our limitations. And one really easy one a low-hanging fruit that I use to teach others who are using AI for the sort of the first few times is try writing a large, you know, a practice question with a fully developed STEM, several answer choices and an explanation with references. Try doing that with a single prompt in your favorite LM and most likely you will fall flat. You will realize that the LM cannot produce a high quality challenging question that tests either knowledge or application or some other parameter like we would have written. And the reality is that taking a question writes takes three to four hours to write a good question, at least maybe longer, and not including the research. But there's no other way to really teach that concept than to say hey like you should experiment and try doing it yourself and realize that there are pitfalls in this process. Is there any place that you would suggest yeah I don't want to self-plug but we have you know the developing project from UCSF uh which is a teach.medicineworks.org there's no requirements to use it you can just basically uh go to the website and start working at it there are two different modules currently one for writing practice practice questions um and one for developing a clinical vignette so if you're uh say you're in you're in medicine and you want to walk your residence through a case it helps you design um a nice CVL type of case so that's there and sort of the sort of the metacognitive approach to all of this is that you can obviously go through these modules, learn about how LLMs work, learn about the sort of the medical consensus of how do we uh create these items. But then you can actually go and actually talk to an LM that's been designed to sort of evaluate what you've learned. So after you finish doing a module and say running a practice question, you could then actually go sit and write a practice question within this specific sandbox and it'll actually then provide feedback on the way you wrote your prompt and give you suggestions based on what you had just learned. So there's sort of an opportunity there to both learn and also experiment as I mentioned. And I'm sure people could do really build on this concept but the whole idea was to experiment and do so safely so that you can develop literacy along the way.

Larry Chu, MD 31:00

And Rishi, which large language model are you using or are you recommending right now?

Rishi Kadakia, MD 31:07

I don't have a particular one and I suggest each person use it and determine what works best for them. There are some models that are tuned for better types of work than others. Conversationally I think open AI, Chat GPT have a great model. As most institutions are starting to roll out their own AI protected systems that support P3 and P4 data, you'll quickly realize that using your institutional system has its own set of limitations. So open AI on the consumer end and the public end may have great reasoning capabilities, but you may realize that on the enterprise side you may be more limited. So the model and how you use it and what format you use it will significantly differ. Claude has also become quite popular as well for its ability to create applications, do coding and so certainly when I think about coding or code assistant work, I do tend to lean more on Claude and the higher end Claude Opus models to do some of my work because it just seems to do the best job. But that is again my subjective sort of take and it probably will change again in three months knowing how things go.

Larry Chu, MD 32:12

The other I think unique affordance that I see with Claude though you obviously you can implement the town extent with open AI's large language models but Claude certainly makes it easier to implement MCP with its connectors. So it has a very easy implementation to connect PubMed to connect site AI to connect gateway which is connected to a large number of medical publishers. So this just allows Claude to access the scientific literature directly from within the uh Claude large language model and reduce um hallucinations. And uh so that's one unique affordance that I've just seen very directly that makes a lot easier for lay users who might not have a lot of time to fiddle around with connecting PubMed and other things directly to other LLM Cloud makes it really easy so you don't have to worry that any references that you're getting or you can ask it to you know find go and find me the latest evidence about you know society guidance for neuroxial anticoagulation and neuroxial blockade uh that kind of thing and it can go into the primary literature for you. So I think from that perspective Claude I think I I've seen there's a unique affordance there. It has a rich I think plug-in marketplace. But again I I have no financial and I think we should insert a disclaimer here that nobody is being paid by any of these I certainly am not uh no one's being paid but I uh but along the lines of what Rishi Rishi and VG both mentioned of that the landscape is changing so fast and that each of us have to go in and explore and try things that in that in that way I think um talking a little bit about what each of us have explored and learned might be useful for our listeners.

Viji Kurup, MD 34:22

And I think we also have one of our working groups at the Society for Education Anesthesia Task Force, the Precision Education Task Force I think one of the working groups is also working on this very issue of AI literacy and we may, you know, hopefully we'll have uh some material coming out from that working group I really agree with both of you about having a larger discussion.

Rishi Kadakia, MD 34:49

I think that is so important. I think as a few times actually I can think of where bottling up and going into a cave is going to help with any sort of knowledge acquisition or tackling large sort of issues such as AI and its effect on society. So yes, absolutely I think having these discussions and that group is and has been fantastic for that specific purpose. And yeah so I recommend anyone uh who's interested in using these tools stay connected to the others and it could be a conversation in the break room talking about tools or something more uh formatted as well. But yeah I I couldn't agree more Vidy and Larry.

Larry Chu, MD 35:26

Well I think you know we could keep talking forever on this topic. We've just like really skimmed the surface but I would want to close out this episode by asking I think a future looking question which is if we zoom out okay this is 10 years from now okay and SpaceX IPO has happened. Alon Musk is now a multiple trillionaire and um he's rebranded as Skynet okay and uh AI has now taken over uh every aspect of our lives Rishi I'm not just putting you on the line here but uh I'll put up the everyone here on the call on the podcast but how are we training our anesthesiologists uh 10 years from now uh and what role is AI playing?

Rishi Kadakia, MD 36:23

Yeah that is that is a great question and I I always think about where we stand in this whole process to start with when we even before AI, right? So we're sitting in the operating room, we're hearing the beeping we're hearing the suctioning and we're integrating all this information and then we're making a decision. Is this important? Is it not important? And then we're acting on it or we're choosing not to act on it and watch it further. And there's uh so much happening in a short amount of time that when I think about okay in the in a real in a realizable short-term future what would that look like? How could that change? And I think having that knowledge and early reasoning will be the next largest step for us. So when we are sitting in the room and something just doesn't seem right there may be a way to describe it more succinctly with these tools. I don't know if the decision making or even putting in the line or intubating or all of that can be done by anyone but our hands in the next 10 years. But I think that access to information will be significantly accelerated and our ability to make decisions with us still in the loop will happen so much faster. And my hope is that that'll lead to better outcomes for our patients. All right VG?

Viji Kurup, MD 37:42

When I'm thinking about that question I'm really thinking about worldwide there's like five billion people who don't have access to safe surgical and anesthesia care. Right? And this might be the way in which we can bridge that gap that we will be able to have hubs of areas of specialization and then in the rest of the places there could be they could have access to information and AI would help but where you need that human that they that we would need fewer humans there and still be able to you know give that access that is so necessary worldwide. So I see hope in this I'm very optimistic and I think that if we are able to scale this level of specialization uh that we'll be able to solve this problem that has been you know it has seemed so daunting until now.

Larry Chu, MD 38:54

I I want to give the most absurd extreme example since you you both have been so you started with Skynet reasonable. So I don't know what I could do beyond that except maybe okay um we will all have neurolink implants in our brains so therefore in 10 years there will be no alarms in the OR because there will be no need for any display surfaces because we will receive all information directly via our neural link implants into our brains and and the AI will handle all vigilance for us. So uh therefore yeah.

Viji Kurup, MD 39:41

So why does it even need a human then?

Rishi Kadakia, MD 39:44

Well yeah yeah like if that's why I said it's rebranded as Skynet I I think I think the end point here is to get more time for crosswords.

Viji Kurup, MD 39:56

There we go more sudoku.

Larry Chu, MD 39:58

Okay obviously my vision of the future is horrible and will not happen. So um yes we will not let I really hope so we will not let that happen. But this has been a fun episode and I think what it really means though is that we as educators we are the ones who will shape the future of AI and medicine and anesthesiology and education and and that means that we we owe it to the specialty and to each other to be involved to as Rishi said to learn about the technology and if we want to shape the future of how it is used in our specialty. And I hope that this conversation and this podcast has helped inspire you to learn more about it to be engaged and to be a part of helping to shape how it is used in medical education and in anesthesiology. So with that you have been listening to the Precision Educator where we explore how data coaching and innovation can transform the way we teach and learn in medicine. In today's episode we explored how AI could help move anesthesiology education beyond standardized paths towards more adaptive and personalized learning and how AI could support anesthesiology learners through better preparation and more equitable learning opportunities.

Viji Kurup, MD 41:28

Our goal is to help you move beyond one size fits all education towards approaches that are more proactive, personalized participatory and predictive I also wanted to say that if you enjoyed today's conversation please subscribe and share the podcast with your colleagues to connect with the Society for Education and Anesthesia's Precision Education Task Force follow the Society for Education and Anesthesia program and join this conversation.

Larry Chu, MD 41:57

Until next time I'm Larry Chu with my co-host and this has been the Precision Educator,