Yeah, thank you, Doug. And thank you, Leo, for allowing me to debate you on this. And in the year of presidential elections, first of all, in the same format, I would like to concede to Leo. Leo is the master of this. Some of the data that I'm going to show is limited, but I'm sure his slides pack all the data to support my cause, because all the research that has been done has been done by Leo in this field. So thank you, Leo, for supporting our cause over here. Just some disclosures. Yes, I am a founder of BrainX and BrainX Community, both AI and healthcare companies. Now here is the first question for you guys. Are you using any one of these or have you used any one of these? Obviously the answer is yes, right? And you're all critical care practitioners, right? So we already answered the question. AI and critical care is booming, right? Now there was an amazing moment this past year, where just like the Mac was released by Steve Jobs, the Chad GPT, for the first time, made AI say hello world to masses, to public. And that was a transformational moment. Think about that moment. And of all the applications, the fastest adoption of an application was Chad GPT. Now that's amazing. So a lot of users are already using that. I'm sure some of you over here are already using that. To demystify some of this AI, because to a lot of people, it seems like black magic. It seems like, oh, you know, it just something goes in there and supposed to output whatever I want. Well, it is a science. It is pretty significant. There is a significant amount of math. There is a significant amount of equations that you see in there that are done by people who work with people like Leo to make it scientific. And that's why it's important. Now I want to show you the landscape of AI publications. This is some of the research work that I do with a lot of my colleagues. And if you look across these 26 different specialties, you will find critical care is pretty close to most of the specialties. And what this slide represents is year-over-year volume of publications. We do this via PubMed search. This year, 21,000 publications that were published in 2023, and we are working on the review for that. 21,000-plus publications for AI and health care. And then Leo would say, well, I don't think a lot of these are good quality publications. Right, Leo? He'll say, like, there's a lot of stuff just getting published. So to go through these publications, we used an AI model, OK? Using a validated AI model to take these publications and tell us which one have significant data and are close to real-world application. That's what we did. So AI to read through AI in health care publications and tell us which ones are mature. Around 7%. Around 7%. And this is using a validated model that was published in Lancet. 98%, 99% accuracy. So a lot of good publication, a lot of good research is going into AI. And we think about AI in critical care. It's there. It's there with a lot of different specialties. And I also preface this by saying the best AI is the one that you don't even realize is running but is helping you. And when I say these specialties, a lot of publications, a lot of research work that is going on in the specialties represents the tools that we are already using. So think about the neurointensivists who are here, who are using the CT scans or looking at the CT scans and the interpretation to diagnose stroke. Look at imaging over there. That's where that research is happening. So yes, the direct publications in AI in critical care might be limited. But if you see the tools that we are using, and I'll show you some other examples, a lot of them come from a lot of these different subspecialties. And you would argue with me that, well, I don't know what's the state of implementation. You told me about all that research. Look at the number of FDA-approved AI technologies. Now there's 600-plus FDA-approved AI devices, 600-plus. There is an amazing amount of growth and implementation going on. Did you catch this at the conference? And thank you, Leo, for championing the cause for low-income areas. You have been championing that. He's great. And think about these tools being put in the hands of people where they don't have the matching skill sets that are available over here. How many workshops are you going to hold for this? Can we enable people? Can we democratize key skill sets through this? And here is a perfect example. Echo, focus, that's been a great focus for us, right? We have been trying to have critical care practitioners be enabled to do this. But we still struggle, right? We are only so many. So how do we democratize? Just like the president said at the beginning of the Congress, that's the big focus for this society. But how do we democratize these key skill sets? Now, think about that nurse or think about that resident or think about that physician or think about that other critical care practitioner who's there who doesn't have the expertise like you to get data. Look at this. If it's an AI-enabled ultrasound device between somebody who is not trained versus somebody who is skilled in doing this every single day, hardly any difference in the image acquisition. That's what you want. That's what our patients want. That's how you will perform at a higher level rather than just trying to get some data. And again, something that is not visible to us in critical care at bedside, how do you think these drugs are coming on your therapy or to your therapeutic committee for approval? These drugs are being discovered these days using AI. Now, here's a trivia. Did you know that majority of the ML and AI students, they are going to some of these pharmaceutical companies? No, not Google, not Facebook. That's where they are going. And there is enough data now to show that AI can help with clinical decision-making, can improve your clinical decision-making. I'm not saying it will do it independently, but it can help you. We all are suffering with this enormous amount of data which is causing burnout using electronic health record. Don't you want that readily accessible to you? Don't you want to use these tools and technologies? You already use it for your phone, for your emails, or Google search things. So why wouldn't you use it for patient care, especially when it shows improved outcomes? And can it help predict who's likely to get sicker? It's not just doing using the values that were there from vitals, but now the world is moving towards multimodal. So it can read the x-ray or it can read EKG images. It can read the data that are there in vitals, the lab values. And then just like we do, just like clinicians do, give you an interpretation that, yes, this patient is likely to deteriorate. Don't you want that? Don't you want that help when you're taking care of all these patients? And aren't you challenged with ever-increasing number of patients? Don't you want that for triaging these patients? Who should I keep back in ICU? Who can I transfer out of ICU? Don't you want that? Or are we going to just depend upon, I believe I'm the best? And can you look at cohorts in a different way? We see so much heterogeneity. Earlier, there was a talk about ARDS. We talked about sepsis. We talked about heart failure. There's so much heterogeneity amongst our patient population, but we have still stayed with the basics. Oh, they are heart failure. Well, it's not just one type of heart failure. We have learned that over time period. ARDS is not a homogeneous state, right? So why not start looking at things? Why not start looking at patterns, which we have never been able to do in a different way? And that's where we can do more precision medicine and treat our patients better. Can we listen to our patients in a better way? Can AI help us with that? So I'm sure a lot of your hospitals look at press Ganey scores and the comments. What do you guys do with these comments? Well, this is what we did, built an AI algorithm where we can take in these comments, and then we can look at the negative comments, and then use another algorithm to tell us, okay, tell me what are they saying? Once you build these algorithms, it can do it in less than one minute. Trust me, when we did this as a quality improvement project with four residents, it took us four months to sift through these comments to do something similar. Is that the best use of our resources? Don't you want AI to help you with this data? And this is where critical care is going with this. And billing and coding, yes, I see some of my friends from billing and coding over here. Oh, you want to spend all your time doing billing and coding, right? Those diagnosis, and then all these people who send you these emails, I think you need to correct your diagnosis over there, doctor or nurse. You don't want to deal with all that. You want it smooth. You just want to document what you want to document. Let it get billed efficiently. You don't want to deal with that, trust me. And that's where things are going, and that's where we are seeing adoption of this in critical care. Oh, yes, the fear. I'm going to lose my job. If somebody else is going to do the ultrasound, then what am I going to do? And this is Jeff Hinton. Jeff Hinton is considered the godfather of deep learning. And he said people should stop training radiologists. That's what he said. In 2016, he said people should stop training radiologists. And you know what happened since? Well, look at this. The number of radiology spots increased. The number of the demand for radiologists increased. And he was proven wrong. They don't know. They don't know what we want and what we need. That's why we need us to be at forefront of this. I'm going to pause for a second and let you look at this data. Let's look at these numbers. This is what the clinicians are asking for based on this particular survey, a 1,500 member survey, physicians and patients. I'm going to, just like Elon Musk said, let this sink in. Impressive numbers, right? This is not just critical care practitioners. This is people, clinicians delivering care. This is what our patients are asking for. And again, I'm going to pause for a moment and let you see these numbers. That's what your patients are asking for. I showed you a lot of these tools and devices are there in the critical care area. A lot of that research work is already going there in critical care area. This is what your hospital executors are planning for. It's on the top of their mind. How am I going to implement AI in critical care? And then this is where the money is going. All these big VC firms, they're pouring billions of dollars into generating AI right now. And Leo, you don't think this is called boom? I think this is boom time, man. What about education? So Leo is the champion of doing data phones, educating everyone. I'm going to give you this example. I participated in this. We had a team called Sugar High. Leo, do you remember this? We did very well with that. And we even published it, Leo. That was awesome. Thank you for helping us with this. And I call that boom. And Leo is going to argue about biasness. But I'm going to tell you, biasness, it's a real issue. It's true. But biasness starts with us as humans. Biasness goes from there to devices that we have been using for a very long time, right? That pulse oximetry? I think there are very clear data on that now. So yes, there is biasness in AI algorithms. But thanks to Leo, who has been championing this cause, we have availability of mitigation measures for that. But yes, it's a real issue. But I think we can overcome it. But are we there yet in critical care? Are we there yet or not? But I'll show you what RSNA is doing. And this is a call to all of us as critical care practitioners. They have a journal. They have a separate society meeting. They have separate data accounts. They have a lot of resources. They have open source data sets available to everyone. I think it's for us, who live in this data-rich field, to beat them. It's a call to all of us to do that. And I would consider that a boom. And now, for the first time, we have these large language models, similar to chat GPT, which are there for health care. These are trained on electronic health record data. And one of the biggest and most commonly used data set is a critical care data set, rightly so, called Mimic, that you have championed. Thank you, Leo. I consider that a boom. This is over time. Prior years were imaging years, because there was one network, one convolutional neural network, and they had the imaging data, and they could build all those AI algorithms. Over time has come with a boom. And what's coming next? Again, in health care, multimodal models. So it's not just going to read text. It's not just going to read the lab values, like I showed you previously. It's going to take in a lot of different type of inputs. It's going to output what you want. A lot of different outputs. So you can take text, generate an image, take an image, generate text. You can do a lot of different things with that. And that's where it's a boom time for us. And a lot of people have said that data is the new oil. Some people have said AI is the new electricity. What we need to form are the energy companies that will solve this problem for us as critical care practitioners, and then for our patients. And that's why I would say please join the Data Science Guide. It's an open member forum for all SCCM members. You can join it for free. You don't need to do anything. You don't need to apply for it. You can just join it. And that's where we provide a lot of resources for AI in critical care. And the key question I always ask, can AI save lives? But I also think about what Gene Kranz, the mission controller for Apollo mission, said. He said failure is not an option. It's not AI that is going to fail. It's us who are going to fail our patients and our colleagues. So failure is not an option for us. So with that, Leo, I'm going to turn it to you. Thank you for championing the boom time for AI.

AI in healthcare

critical care

AI advancements

clinical decision-making

AI democratization