Computer generated transcript

Warning!
The following transcript was generated automatically from the content and has not been checked or corrected manually.

Good evening. Um Thank you ever so much for joining us this evening for the first of our, er Royal College of Services of England Innovation Hub webinar series. Um Next slide please, Sean. Um It's a, it's a really great er, privilege to be able to bring this, er, event this evening. And I'm hugely grateful for our speaker who will introduce to you shortly, but with all of the excitement going on around, er, generative A I, we thought it was really a great opportunity for the I hub to um start a conversation about the role of A I in surgery. Er, and so that's the kind of the premise for what we wanted to bring you this evening and next time. So from the innovation hub and hosting this evening is, er, myself and one of the plastic surgeons at Buckings Healthcare Trust that will also lead the, er, innovation hub. Er, and my colleague, Dina Hardy, er, who's a consultant in Manchester and one of the robotics leads, er, and professor, um which I promised her I wouldn't get in by Gooden. Um and I will hand over to, to Dina just talk very briefly about the, the Royal College I have and what we do and how you can get involved. Thanks so much Ryan and thank you everyone for joining us on this uh Tuesday evening. It's really exciting to be here and it's great to see so much enthusiasm for innovation. But for A I uh next slide, please, Sean. So I'm sure those of you who are innovation enthusiasts or interested in either innovation, whether that's through entrepreneurship or entrepreneurship are well aware that the Royal College of Surgeons launched their IHAB about 12 months ago now. And we've had a really exciting 2023 and this is the first of our 2024 events. We're a team of eight. We're a variety of different disciplines, a variety of different surgical specialties and a variety of different grades. But we are all um young, enthusiastic and keen to connect surgical innovators um beyond just the college and use the college as a platform. We need to drive innovation for surgeons by surgeons. Next slide, please. And so the aims of the IH are really, really straightforward. They're to support innovation through education, resource and collaboration. Next slide please. Shan and our aim really is to educate the surgical community, those who are at the cutting edge of innovation, but those who are also new to innovation and those who want to adopt innovation, to teach them how to support and develop innovation within the NHS. Next slide, please. We want to provide you with resource through innovation, events such as this through specialist innovation, support clinics and through a variety of other educational research platforms. Next slide please. And importantly, we want to create a really rich robust ground for innovators by connecting partners that are either academic, educational or industry because it's only through collaboration and networking that we can create a really rich environment for surgical innovation. Next slide, please. So we're welcoming everyone. It doesn't matter how old or young you are, it doesn't matter what stage of your career you are, it doesn't matter how much of an innovator that you may think you are, you may think you're not. So I can tell you when Ryan first approached me, I was a bit surprised, but actually, he made me understand that within me, there was a huge amount of innovative potential, but a huge amount of interest and innovation. And I think that if it's something you're thinking about, then join our QR Code, log on to our QR Code and join us, we'd be delighted to have you and thank you for joining us this evening. I'm gonna hand back to our next slide, please. Thank you, Dina. Um So it's a real honor to introduce our speaker for this evening. So Taa leads uh the uh health care uh focus for Microsoft UK in terms of the generative A I um with a kind of huge focus on actually implementing meaningful um er impactful er projects and products into the er healthcare workspace, er and also is taking the lead on as I know he would be talking about bit later on around fair A I which is all about the ethics of A I. Um and how we make sure that whatever is implemented er is done. So in a kind of solid governance within a solid governance framework. Next slide, please. So before I hand over just to give a bit of uh uh kind of indication of what they say, um Titania is gonna talk about what er ga I is aba bit about how you get the best out of it and we're gonna cover some use cases and that'll be the first half of the um of the session today and then the second half, we want to have an open discussion. So please do use the chat box over on the right hand side, some of our college, college colleagues uh are gonna be going through those suggestions and, and highlighting them and we really want this to be a kind of fluid um er discussion at the end. So without any further ado I will hand over to um to Taa to take us through genitive A I. Perfect. Thanks a lot Ryan. So just a small housekeeping stuff. So I'll try to share my screen. Please let me know if you can see it clearly. Yeah, yeah, I hope you can see my screen Right. Right. Yeah, perfect. Yeah. First of all, thanks to the whole II innovation hub and also especially to Ryan for giving me this opportunity to present the in healthcare. And today I think I'm going to spend next 20 to 25 minutes uh talking about health care. So I know that it's, there is a lot of hype in the market across the social media across I would say X and even in your daily life. So I'll try to share the experiences what we are seeing from Microsoft end. And uh I am trying to bring the hype to the reality through the real life use cases and also how A I is being evolved as we start like every day, there is new things that are coming up. And if you are not seeing the news, there is a soda from open a eye. So you will see uh I'm not sure what so today, but I'll talk about how the evaluation of A I is happening. So firstly, you can see here, this is one of the quote that is coming from Forbes chart is Michael Jordan. And but in medicine, the support cast is the key. So the analogy here is um if you've seen the last dance documentary from Netflix, you can understand like, you know Michael Jordan is the hero, even though Scott Pi and Dennis Rodman are the supporting cast. But here in A I in A I world in medicine the supporting guys, the people who are forefront it, they it can be the doctors, it can be nurses. They are really key in terms of how the application of A I is. So with that, with that, let me go to the my second slide. So here II just will like to talk about the common questions that you all might have. So few questions could be what's opening A first of all, what is charge G be and how they are related to Microsoft and how Microsoft is with open A I? And then is a always right. You might also have questions around is A I governed is A I safe and, and to other end of the spectrum, you might also have a question like will A I replace humans or what is a copilot? And what is a pilot? And also there's a lot of bugs in the market on hallucinations. So will A I ever hau hallucinate or will A I give all the right solutions every time? I don't think so. So like then there are a few questions around. What's the prompt engineering? Is it the new technical skill set? And what are the implications of A I? So uh I thought like this could be a few of the questions that could be on your head. I will try, what I tried to do was try to answer most of the questions that you are seeing on the screen. And uh this is, they will be covered in the following slides. If you feel like something is not covered, feel free to drop a, a message as Ryan said, and I'll try to answer most of them if I know if I don't know, I'm happy to come back to you. So OK, let's kick start with the agenda today. So I try to divide the agenda into four parts. So part number one, I will talk about the generative A I. So before uh jumping into A I and all the things I try to go from the bo eye view where the A I is starting, what's the history of A I at a high level? And then I'll talk, talk about what Microsoft is doing with Open A I. And because our collaboration with open air is really pivotal moment, I would say uh that changed the whole A I industry. And then I would like to talk about uh what's Microsoft vision where we are driving towards? And then I'll come to the second topic that is open air offering. So what Microsoft is doing with Open A I? So are we trying to question A I? What's charge GPT? How A I is related to that? There might be a few questions around the models you are hearing in the market. I won't go into a technical level, but I'll try to cover the surface side. And next, the third topic is responsible A I this is really key for Microsoft because A I, we want customers to use A I with a mindset that it is properly governed and it's not something that is uh loosely coupled where use A I and you see a lot of problems later. And next fourth topic is on the use cases. So what I felt is uh it's better to jump from respon back to the use cases. So these use cases are not us use cases. These are more specifically around the UK healthcare system. And what Microsoft is seeing within the UK health care, we are trying to learn again here. So because we understand you are the doctors, you are the forefront, sitting in front of the patients, you know, things better and we are also trying to learn as a part of our journey. So there will be a good amount of use cases that you might know already there might be something new, we're happy to actually learning by doing. So if you have something new, feel free to share to us and we'll define ourselves. Finally, the fifth topic is the links. I'm not gonna talk anything on that, but I thought like I got some learning guides and certification routes of I know your doctors, but in case you are interested, you can have a look into that fifth section. Let me jump into the first topic. So what's the history of A I? So like it is not something that came into the market because of chap. So the history started all the way going back to 1956. So the early start of A I is the exam question was can machines think like humans? And that's where the actual research started. And the research went on from 1956 until there is another pivotal moment that is 1997 where the machine learning actually came into the picture. Machine learning means you have an algorithm that you fed a lot of data to that algorithm. And algorithm actually understands the meaning of the data and giving you some predictions out of the data. So that is called machine learning. And this branch of study is really critical because this is the subset of the evolution from A I. And then there was another moment in 2017 where the deep learning has came into the picture. Deep learning means um A I, you don't understand linearly, right? You ask a question, you give you the answer but behind the screens, it works like a neural network similar like our brain, you have a lot of branches in your brain. So you need to look into the different aspects of the data and then you need to give the uh right making. That's what the deep learning is from the evolution of the deep learning that came the urea moment. That is generative A I. So generative A I meaning is you are generating a text it could be audio, it could be video based on a prompt. Prompt is the question that user is asking. So the A I has moved away from the back desk of developers somewhere sitting in a remote location. And uh we brought the A I to the forefront of the business people are common people who actually can interact with A I using a natural language. Uh When it's a natural language, using the colloquial language that we speak to a person. That's why the generative A I is really critical and that's why it became very prevalent in the market. How prevalent is it? It is? That's where you can see the trend here. So this is a stat where we see the evolution of uh time to take by 100 million users in terms of adoption. You can see we started the smartphones from 16 years then to internet for seven years, then Instagram, it is 2.5 years. Now the GP that is what we started with A I, it came down to two months. So which is a rapid speed. That means the rate of adoption is very high. This is what the meaning of hype to reality. So the reality has touched base within the two months purely because everyone was able to interact with A I and ask those meaningful questions and get the response. Now, what is the total market size? So we are looking around one trillion uh market share only in health care. And this is what mckenzie says in their uh recent survey when I say recent uh it was July last year. And uh and even uh from Microsoft end, we are also trying to learn from this survey how the meaningful of one trillion will be from UK point of view. So what does A I have done a have A I have done the platform shift. The shift is happening in terms of you as a user sitting in front of the system asking the questions and using a natural language. And then there was a reasoning in, in running behind the screen and giving you the response back. So that is called, this is called in a simple example, what generative A I is and this is what in case you have seen the chat GPT that is what happens behind like you ask a question and the model will give you the response. Now, uh what's the love affair of Microsoft and open A I? So this is a really interesting thing because as a Microsoft, our goal is to make sure empower every organization, every person to achieve more and open A I. Their goal when they started in 2016 was to build A I that benefits the humanity. So together with Microsoft and uh open A, so we have developed different models. So, so you can see the models are 3.5 to 4. These are more like a text models that is what you are seeing on chat GPT. Then you also have models related to Dali. It's more about creating an image based on your question that you asked. You can say like I want to sit a monkey sitting on a mountain. So the Dali will produce a picture out of it. Then you have whisper, it will try to translate an audio file into a text. So for example, if you have a call center or like somebody speaking to another person and that is in a audio bite, you want to translate that not like a subtitles, but more like a text that is where you use Whisper. And finally, Sora, so Sora just launched uh in few days back. So it can create a video in a real time based on a question you asked. So, so think about open A is more like an incubator where the models are built. And Microsoft is like an enterprise grade solution which wrap that A I models that open A builds and we make that as enterprise grade solution. So that is what our Microsoft and Open A collaboration is to give some more context to it. So open A I, what you are seeing on Chad GPD. So that is where you do more like an individual basis. For example, you as individual want to understand with A I, you can use charge GPT if you use charge GPT plus that is you pay $20 per month, you actually can use that um latest model and latest data. And then like, you know, from Microsoft side, we actually provide the enterprise grade solution. So if you want to use enter, if you want, if your firm wants to use A I open A I is not the right one because you need to have all the security, you need to have the regulations, you need to have the framework uh the enterprise package behind that. That's what Microsoft does and how we are doing it. So we did investment of 1 billion in 2019 and this investment opportunity has given a levy from Microsoft that all models that open air is going to build, everything is built on Azure and Microsoft has exclusive privilege to commercialize all those models. And we have invested another $10 billion in 2023 to accelerate all the research that open air is doing. So open air build the models. Microsoft take the models and make it enterprise ready. That's the summation I would say now. Um so as you know, from Microsoft side, we are moving away from cloud company and technology company and we are now poisoning ourselves as an A I company and how we are doing within health care. So within healthcare, we are trying to divide the impact of A I into three categories. The provider side. So that is where you provide uh carers like private hospitals where you provide the care to the patients and then through the payers, the insurance providers, how A I can impact that section. And finally, the life senses that is where the discovery, the research and the clinical trials have all those happening there. So we have uh currently like penetrating into all the three markets and within the UK as well. So I look after the private hospitals and health tech, my tech within the UK uh territory. So, so that's more about setting up the leveling field on a, what's the partnership? Now, I will go through the Azure Open Air offerings. So from the Microsoft side, you think about like anything from if you cop those are like applications that you see at the front end and then behind the screen, these are the different A I models that works. It's not always A I, right? Sometimes you need to use a combination of vis uh cognitive service, we call cognitive service means. Uh for example, you want to understand like you want to simply convert audio to text, you not use A I every time you can use speech. So we always encourage customers to think about A I is not a silver bullet for all your problems. A I will be useful only when it's uh right business case or you have other options to evaluate as well. And, and also from Microsoft side, we uh we also came up with another uh technology called copilot. So why we have used copilot? Because end of the day, the pilot is human and uh human cannot, like, you cannot automate the whole process with A I. We need to have a human at the end of the loop. So the answer is a I can never replace humans. A I can only augment what the humans, what currently, what they are doing and it will provide as a support mechanism. So to to make an right word for the support mechanism, Microsoft came up with a word called copilot and then uh what Microsoft done with the A I is we try to inject A I into across all the products that we have. For example, we have M 365, you always will be using the MS word powerpoint Excel and all. So now we have a Copilot where you have a I engine which can create a magic and I'm happy to share the videos later which will blow off your mind in case you are not seen, then we also have something called github Copilot. This is something for a developer community. So right now, developers typically spend uh like 60 to 70% of the time reinventing the same code. So what github Copilot does is it helps the developer to actually uh use the code that could be used across the enterprise or outside and then they can accelerate the code development where you can see the real world products being built, which are used to take your time. Now takes a few months or a few weeks. Then we also have the power bi copilot, which is more around. Uh how do you build the new reports? Then we also have security copilot which will help you to uh like prompt, what are the security firewall threats that as enterprise you might have? And then you, it will alert the uh like respect to see so that the customer side security and infrastructure manager to make sure that you know, he uh address those issues before they actually arise. So we are seeing the two parts of it. One is the copilot section where all the products are involved. So A I is not just for one purpose, it is for the multipurpose. And then on the right hand side, we have A I which can be used for the front end applications. So you can use A I across all the spectrum of the works you are doing in a day to day life. Now I thought like I would show you a small demo uh how A I can uh can be used in real time. So this is for gi copilot, you can see on the left hand side, there is a cortex model, this was generated open A. So this is the model that runs behind the screen. And the GITHA of the centerpiece is where the customers will use to type the code. So in case you have seen on the right hand side here, think like you are a developer, you're trying to write a code where you want to validate the list of uh email addresses. So whenever the developer types here like validate each address uh with an API, so this is what he's trying intended to do. Once you type what you want instantly, you can see the list of code that is generated. So this think about this is like a suggestion. Now um it will be up to the developer to actually accept this code or edit this code or use part of it. So in the current world, he or she, she used to go to the private website like stack overflow. That's that's where typically think about stack overflow is like a common website where you look for the code and where you bring that code into your system. Now, the developer will not need to go anywhere outside. He just need to give the prompt. So this is called a prompt, then you get the response and then you can improve those su suggestions. So this is one way which you can use uh like copilot in terms of accelerating your day to day job. Now, I will talk about responsible A I. So responsible A I is very critical in the heart of Microsoft because we want to make sure all the A I that is being built. When I say A I, the models, the P models and then the daily model that you saw or it could be. So in future, all those models have been prebuilt with this list of principles. So there are six principles that are Microsoft has ensured that you can use A I with safe so that you don't get into the hallucinations. So the first one is the fairness. Then second one is reliable and safety. Third one is privacy and security. Fourth one is inclusions and fifth one is transparency and last one is accountability. So you might wonder what are the six all about, how does it make an impact in healthcare? So, within healthcare setting, this is how we made a definition. For example, when I say fairness, we want to make sure the model that is being used on the health data. Uh and it takes all the demographic basis into consideration before it gives an outcome. And the second one is reliability and safety. We understand that patient data is really critical and we want to make sure the the summary of what you are trying to achieve here is not a is, is a fair representation of the whole context behind that. Third one is the privacy and security. We all understand that uh we living in a patient where, where the data is really critical and making sure that it is uh rightly used. So that is where we have the security bo boundaries defined. And the next one is inclusives. So we want to make sure that A I, the response that A I is giving is considering all the population segments. And next one is transparency. Transparency is more about garbage in garbage out. So you want to make sure that we are giving the right input sources to A I and it is giving the right responses. Finally, the accountability. So accountability is all around how we are making sure we are re re all the models being legally and regulatory compliant and how they are working in a fairness way. So that A I doesn't get into hallucinations when I say hallucination giving you the wrong predict your outcomes. Uh for example, in a simple example, right? So A I can predict there might be uh a rain tomorrow but in in idle world, the forecast tomorrow doesn't show any rain. If A I says it's a rain or snow tomorrow. That's a simple example of a hallucination in a healthcare setting. You can think like when a healthy patient who has, for example, um he's doing really good across all blood sugar and also he's doing very good at BP. But A I tells like, you know, something else like even though his blood sugar levels are very low A I tells like he is prone to diabetes, type one or type two that is called hallucination because it is giving a wrong outcome. So, so these are the six principles are into all the models that we have well from open air through Microsoft. And that's this is what I mean by enterprise ready. So Micro Microsoft ensures that customers are aware of all the principles that are being employed and they can use A I for safe. Now, you could have a question. Can customers tweak these models? The answer is yes. So sometimes you need to make sure that uh you need to make sure certain things should be allowed. For example, um in a normal context, I want to use a knife to cut the body in a general term that is more like you're killing someone. But from a health point of view, you are doing a surgery there. So these are different types of anecdotes which A I needs to understand that is what we call learning. So in that context, uh we have given the option for customer to train the data so that A I understands what the data is, it provides the responses in a meaningful way. Now, in case you are interested more about reading Responsible Eye. So we have a book from our legal officer called Brad Smith and this is called Tools and Weapons. I would really urge to read this book. It's really insightful and very powerful as well. Next one, We always say that what the models that you are you within Azure, it is like your data is your data, meaning that if you want to test the A I, your data will not be going outside of your subscription. It resides within your subscription and all the data that is being trained on the model will be your model. It will not be used for a general purpose where you see an open air in charge. And finally, as I said, all the models that are enterprise ready from Microsoft, they are HIPAA and ISO and so two and starlit. So more are being added to this list as we move on this A I evolution. And finally, the last segment is the use cases. So I was thinking before jumping into the use case, start to la landscape, what we are facing the challenge in the healthcare sector in UK. So most of you already know about this one. So please um apologize if few of the stats are not incorrect. So first thing is you can see a backlog, we have 7.2 million backlog and we just keep on growing and we have challenges with staffing and there's a huge uh staffing uh like shortage. And then we also have a issue with uh A and D in terms of um how, how many patients have been seen. And then there is a lot of funding uh cuts that are happening in the social care. And then there is uh a lot of pressure on the commercial and clinical trials, which is really important. You can see there's a 24% decrease uh from 2070 to 2020. And so there should be some answer in, in terms of because we need to accept the reality that funding or funding cuts is going to happen further and how we from the industry, how do we support the health community? That's where A I will really be useful. So uh to accelerate and also to help address few of the things if not all. So these are the few use cases we are seeing from clinical point of view. First thing is finding the patient information. So here uh for example, you when a patient with a GP, most of the times you have a history of information stored about the patient, about his previous diagnosis, some information could be in uh x-rays, some mentioned it could be radiology reports. So how do we comprehend information both in the system like E or system C or TP? And then how do we combine that information back with your clinical reports? How do we summarize that information? That's where I could help you to provide the insights. And then second one is prioritizing your workloads. We understand that uh in your secondary or primary secondary referral, you will have a lot of patients to visit. So you need to make sure that you provide your time to the patients who need it the most. So a can help you to provide the sorting for those workloads. Next one is empowering patient with chatbots. So this is where uh it's more like a triaging purpose. So how do you make the intelligent triaging? Not just by asking like a standard chat board kind of questions, but how do we be able to understand what patients are going through and follow them for a clinical pathway that NHS uh provides the fourth one is imaging. So this is where um one of the beautiful example is NHS Grampian, where uh NHS Scotland team were able to identify uh they are using A I for identifying the like early detection of cancer in the breast screening. So this is about uh finding like passing A I through a list of images and finding those patterns that you require. So at the, at the same time, it's not like it's going to replace A GP here, it's more like a providing support to the GP in terms of uh like a secondary or tertiary referral. And next one is how do you streamline um like, you know, patient consent. So like because you are doing a lot of activities that involves patient consent that should be flowing through end to end how to use A I to actually augment that consent. Next one, these are administrator use cases. So one of the strongest use cases, call center. So we understand there's a lot of on the call center team to actually address all the calls, provide the summary and then uh provide the answers as well. So uh we have uh successfully used A I in terms of augmenting this one. Next one is the prescription automation. So NHS, we typically receives thousands of prescriptions coming from different pharmacies and most of the pe most of the times it is sorted manually and how A I can help to sort all those tasks um in a real time so that you can save the time uh in, in terms of dispatching medicines to the patients. And we have a real life use case working at the UK firm right now and then data optimization. So here like there will be a lot of people come as op but A I is one of the use cases. How do A I understand how many people are coming for P and how many actually turn up being uh going for a the uh going for operation the and finally, we have research and regulatory use cases. So this is more about clinical trials, I like discovery. So clinical test could happen through patient co from patient angle or from uh or from a clinical test angle. So how do you identify the right cohorts that should go for a clinical trials rather than looking into laundry list of information. And how do you make sure that you know you are making the right right amount of sales in case you are looking out for um wider adoption of your solution. And uh then we also have nuance. So this is a Microsoft partnership with the Nuance. Uh you might heard about Dragon Medical One. So that's now called Nuance and they are now accelerating air adoption to Dex Copilot. So in theory, what it does is whenever you have an interaction between uh A GP and patient, most of the times you spend time once the patient leaves, uh in terms of writing the summary reports and how A I can be helpful, Like think about it like a technology that is sitting in a device or in a medical lab, which uh we understand the conversation that a GP have with the patient and then provides a summary which a doctor can have a view as a draft and then doctor can submit the draft or tweak the draft and submit in the system. So it is going to reduce the time almost by 60 to 70%. And uh finally the R and D paradigm. So as I talked about today, there is most of the time the R and D is happening through wet lab. And uh we can think about most of the time experimentation, but the future is more about how do we find the right cohorts, how do we do the predictive modeling first and then go for a research so that you can save a lot of time. So with that, um I know it's a wh stop, I just uh wizzle through all the slides, I will try to show you a couple of demos which will show number one. How do you build an A I simple A I for your own uh trust. And secondly, I also do a demo with Ryan on how you can use A I in a call center uh set up if I just share my other screen. Yeah. So here I hope you can see my screen, right? Yeah. Thank you. So this is a setup I have done. Um think about you have a lot of patient information stored in different P files and you have you want to create a chat GP like chat GP to kind of experience on top of all your patient data, this is how you can do it. All I tried to do was I tried to connect a a demo company called Northwind Health Plus. So they, this is an insurance company and they have learned a list of PDF documents behind. And these are few examples in which a patient or end customer will ask those questions. And in an in a traditional way, they might, they might call to a call center or they might ask the admin team from uh respect to insurance company, they get those details. But now A I can actually help you to find answers very easily. So for example, if there is a question around what is included in my plan, which is not standard. So this needs to combine from data from different like let's say you have 50 documents. It it will be sitting somewhere in one benefits document. So I can give you the summary and also give you the citation where the summary is being uh pulled from so that you can go to the right summary here. On the right hand side, you can actually get into the more details. I in case you want to read it, it also provides you some information around how the supporting content from where it is pulling the pulling the insights here. And also it will give you the thought process how it is trying to understand behind the screen as well. This is one simple example. And if you want to use um another question like what happens like what does product managers typically do? So yeah, again, like the same thing, right? So you in in your health care? Well, you can ask like what, why is this patient when the patient has not visited me? And what was his problem at that time? So I can give you the summary and it can actually pinpoint uh where the summary is coming from. So this is a laundry big 31 page document and the answer is coming from page 29 here and this is what it is. This is like a small demo. But now I will go back to uh a demo where you can see how a conversation between a agent and conversation between a customer will happen. Let me share my screen again. Yeah. Ok. So just to give you a background, this is a setup where um there's a conversation happening between a insurance company provider. So I am a, I'm taking a role of a call center agent and Ryan is a patient or you can think of a customer who is calling the insurance company for a problem. Ryan. Are you ready to start this demo? Yeah, my acting is ready to go. Ok, just so we'll be starting 123 go. Thank you for calling Coma Insurance. My name is James. How may I help you? Uh yeah, I've had an accident um and I'm calling to file a new claim. Oh, I'm so sorry to hear that was anyone injured in the accident? No, nobody was injured but there's some damage to my car. Good to hear that no one is injured. Can I get your name please? Yeah, my name is John. Can you verify your date of birth? Uh it's uh October the 29th 1996. Let me pull up your information please. Hold on. I see you live at 60 Peters Drive Manchester. Is that correct? Yeah, can you verify your phone number? Yeah, my number is uh 0712345678. Ok. Where did the accident happen? Uh it happened in a Tescos car park er in Manchester City Center. It was raining really heavily and I guess I didn't see the, uh, the other car when I was backing up. Ok. When did the accident happen? Uh, it happened, uh, Sunday morning at about 10, 8, 10 o'clock on March the fifth. Can you describe the damage to your car? Uh, yeah, my car's bumper is damaged on the right side and I've got some pictures of the car that, uh, I can send to you if that's useful. Ok. Let me create a new claim for this. Please hold on. I created a new claim for you. We'll be contacting you for scheduling repairs to your car. We will also send you a link via email so you can upload your pictures you are taking. Is there anything else I can help you with? No, that's all good. Thank you ever so much. My pleasure. Have a good day. Bye. OK. So here you can see on the left hand side. So this is what I think the transcription that happens in real time where you make a conversation on the right hand side, you, these are the li literally those entities that call center agent would like to fill. I'm not sure why it is not showing the mobile number and all here which should have come from Ryan. Uh OK, sorry for that. So this is where like from this conversation A I picks the keywords for all the list of questions. For example, the organization event, the address and what is the product? And why the call has happened. And all it's like what accident. So you get answers in real time on the right hand side and then this is like a custom prompt. So think about this is something uh agent we're following the call. He or she would like to send um he uh she would like to send or update the back of the patient uh register log so that he can submit the record in the system. So they can just type in here and they can see like, you know, what's the response. So here, the main reason for conversation was creating a new claim and sentiment is uh there is no one is injured and the accident location is Manchester. And then like we also have something called uh Jason file. So what does this, what does this do? So here, uh for example, uh there is a interoperability. So there is a medical system and then there is a call center system, both of the systems to talk to each other. And this is what the uh transcription that happens behind the screen. So A I can simply do the transcription, it can give you the output for the technical team in a Jason format. So that's what the A I can do. The ideal time it will take is a 15 to 20 minutes to combine all these activities. But with the help of A I, he can finish all these activities in a span of couple of minutes and this is a real time use case which we are done with one of the customer. So with that, I will take a pause and I think I will hand it over back to your end. Fantastic. Thank you. And er, I think, er, obviously it's, it's pretty clear to, I'm sure everyone in the audience, um, that whilst that's not a medical thing, you can imagine the future whereby we had telephone consultations that actually all the key points are brought out, allowing us instead of what many of us do at the moment, which is having to have our clinic and make notes simultaneously being able to um be pulled out. Er So I think, yeah, it's super exciting. Um Dina, I know you had a first question as a as host. Definitely get uh uh the kind of priority question first. Um Thank you so much to Taa. That was really, really interesting and what a whistle stop tour of what Microsoft offers, but also opening up the um potential um opportunities for healthcare and particularly for the NHS. So you made a really interesting point about 100 million being reached in 22 months. So hype to real reality, but hype to reality to realism. Where do you see that? What is, what is the, what is, how are we going to bridge that hype to reality and that reality into realism for, for a system like the NHS which is very complex, fragmented. And although we're one system, we're not really one system, we're multiple systems within one system. So how do you think you will get A I embedded into the NHS? And how do you think, what do you think we need for interoperability of all these various systems? So I know it's a big question to start with, but I thought I would kick start the conversation. Yeah, sure. Then that's it. I think that's what the question for Microsoft as well when we started the journey. So what we have done is we started uh using a from the back office point of view because that's where it is slowly uh uh it's not a highly regulated enrollment. It's more like how do you improve the productivity gains. So a couple of examples are number one is nuanced DEX with the interaction between a GP and a patient and where you can uh automatically transcribe the conversation into the text. That is one use case. And the second one is the admin task uh with the, with the GPS are like, you know, spends a lot of, not GP, the spends a lot of time. One of the example which I gave was prescriptions. So like uh NHS uh trust will typically receive prescriptions coming from all the pharmacies. And then they have to understand those prescriptions uh information from those prescriptions, then they have to dispatch the medicines, right? So there uh we are seeing a huge impact of A I. And then one of one of the other example was NHS Grampian, which I gave, I find the cancer uh from the list of radio mammograms and finding, acting as a copilot for the GPS. So that is another one. So it's still evolving then and there is an appetite in NHS plus now. So everyone is trying to see like, you know, what are the right rightful use cases, few of them which will uh run through. And those are the ones where the impact is right now happening at NH I, I think that's really um a really good point that you make actually. And I think often as doctors and surgeons and there's a lot of hype isn't there that A I is going to take over and we're, you know, there will be no radiologists left but clearly a lot of the emerging evidence that's being published is that like you say, it's copilot, so the best way of interpreting a mammogram is actually using a radiologist and A I in conjunction rather than to individual radiologists. And that's probably the most cost effective way. But I think that it's really interesting that you're using A I for those, like you say back of office tasks like administration. And I think that's really intelligent because that will help us create a digitally enforced, digitally enabled workforce, which is really important. So it's not just about the surgeons, but about everybody, you know, all the, the physicians and the clinicians, it's about all of healthcare and everybody who works with that. And I think that actually that's probably a really intelligent way of introducing A I within the NHS and then slowly building in the layers to, to become clinical facing. So II think that's really great and I think certainly it will be a much easier way of introducing um A I into, into the NHS. Um But I will stop hogging the, the the floor and we'll probably ask some questions, why. And do you want to take that? Just following on? I think the point that uh again to reiterate what you've just said that as, as doctors in clu and surgeons that we've got to remember that the role of A I in health care is not necessarily the bit that we're doing at the, at the front end, but there's a huge amount of work that happens to support the work we do. And actually there is probably a lot more low fruit for kind of high impact in that group than, than the kind of the stuff that automatically comes to mind. I think it's a super question from Oscar Wallace King. Er how do you ensure responsible A I principles are actually maintained and reached? So it's really impressive the um the the the slide you showed with the different steps, but uh what are the requirements for regular auditing of A I and how is that happening? Yeah. So what we are doing, Brian on that case is so from the models that Microsoft is currently publishing from opening through open, all the models will follow those principles that are shown on the screen. And, and at the same time, we also have having discussions with the customer data security legal teams to walk through them how Microsoft is trying to build the A I those responsible A I capabilities and then how do you want to play those uh responsible A capabilities within your limit as well? So as I, as I give one of the example I gave earlier in the talk was uh cutting someone's body in a normal language is killing someone. But from a health point of view, it's more like you're doing a surgery. So those are some of the nuances and hallucinations which uh the trust, for example, in the, in this case, some trust who wants to use it and they can um then they can create a responsible I framework that is what we are encouraging them to form building. Like one of the recent uh like offering that I'm working at Microsoft is how do we create a kind of responsibility committee and where we have different bodies regulation, body, data, body, security body and how everyone understands what A I is doing. What is the use case? What data is being fed into A I? So, so everything will go in a nutshell, right? Go by use case that you are applying a and for that, for that use case, what is the data and what level of security of data that is required? So we we'll be auditing the output I think is probably Oscar's question. Apologies Oscar if that wasn't quite but like, II understand the auditing and, and the governance around the start but how much of what's what the output is is being checked and audited? And yeah, so coming to the output, that's why I said uh it's like a A I will give you the head start from 0 to 80%. And then as a doctor in this case, your it's your owner like who is actually using that A I is responsible to make sure that the output is true, factually true. And then for example, A I is not always perfect and you might say that this is right and this is wrong. The beauty of this one is it will learn from this uh what you are trying to provide the input. So next time when it encounters a similar kind of a situation, it learns from it and it will give you the right response. So the efficacy will actually increase over a period of time. All right. Yeah. Fantastic di did you, I suppose data out is only as good as the data in. So as the more data in that you get and the more high quality data in that you get the more likely like you say, to tell you the A I algorithms will refine and become more intelligent and more accurate and then you know, the output will be better. So like you said, I suspect at the start of this journey, whatever that looks like, even though you'll have huge volumes of data, it will only be, you know, specific parts of the country specific subsets of patient cohorts, et cetera. But as the data in gets bigger, you'll be able to make it more generalis and hopefully it will improve outcomes overall. So, you know, II understand that regular audit is important. I suppose the other thing to know is that there's a huge amount of data bureaucracy in the NHS. Um and one, the things that we probably need to think about and it's probably not for Microsoft alone but for organizational bodies like the R CS in conjunction with Department of Health, NHS England and Industry is how do we streamline that process to introduce new technology in a safe and effective way in order to, you know, so that we don't miss out so that you don't have some trust that are very early adopters for whatever reason and some trusts that lag behind because ultimately that leads to disparity. Yeah, exactly. Yeah, there is a um a kind of a, a practical question, er, kind of combining a couple that have been put forward at NHS level at hospital level, how will how will we as clinical staff or nonclinical staff actually be able to use A I like? Is it, is it already within the kind of um office 365 set up? Um yeah, how practically how, how can we put it to use? OK, that's a good question Ryan. So so as I said, there are different ways of what we call is A I. So if you, for example, the M 365 like office PT Excel and all, so those are your day to day, those are like a tools that you use on a day to day basis. So for that is more about there is a subscription per user per month that uh like you know that you need to pay and then you can use A I like Copilot for that. So Copilot, I said there is nothing but A I engine behind that. And for the, for example, if you want to use the enterprise solution where you want to have a huge impact on your back office. In that case, uh we recommend to actually have an Azure subscription and then within the Azusa subscription, you actually uh what you do is you actually use A I and you bring your data from your systems into Azure and then actually you can uh leverage the A I capabilities inside that so that you are, you are rightfully governed, then you are sending the data to outside world. So that is how you can use it, but you can't use it directly on your laptop or your desktop. So you can't think about using open, for example, open air subscription and think that is safe I II would not to do that because open eye, open eye, whatever the charge is, more like a individual level, as I said in the beginning, it's not enterprise ready. You should not be passing any of your patient data or personal data into that because that is used for a training purpose. So that is a general purpose model which is being trained across the millions and mil petabytes of data coming across billions of population. So don't use that open A, you should come to an Azure or like you, if you want, you have also have the GCP aws. So they also start building the A I models. So I would recommend going to their respective infrastructure and start using it. But if you want use the advanced models, you have to see, you have to be on Azure to use all the GPT models. Yeah, and, and just absolutely repeat what you've just said of do not use patient data on uh on uh Yeah. Um and, and so, yeah, you can imagine actually, you, you'd hope there's a kind of national level um drive to integrate this because otherwise you could certainly see how the big trusts that have money or have some money can almost become more efficient. While the smallest trusts that are struggling that therefore wouldn't necessarily pay for um kind of the A I support kind of end up falling apart where you need that equality and equity um of, of the use. So I want to ask you a question. So, so not all of our, so we have lots of trainees who might not be part of, you know, who rotate around different trusts who get different access to different systems. But for somebody who is just interested in data science, in digital surgery, in innovation, which is what the Ihab represents. What what advice would you give to new, new pe people that are new in healthcare professionals that are new to A I that are new to the phenomenon of change. CPT or DAL three. What, what, how, where do you think? How can we use it effectively to help us educate ourselves better, train ourselves better? Should we be going on training courses? Should we, should we be learning how to put in? Right, correct prompts or prompts in the right way? What, what's your advice to us? Yeah. So II got this learning guide you can see on the screen. So be depends upon which level you are, if you are basic, you are intermed advanced. So I have pro I have I will share this link after the after this call and I will ask all the new be to actually use this one and also there is a certification path if you want to upscale yourself, like you can go through A I fundamentals, which I would recommend, I don't think you need to work on A I engineer or data scientists that is more like a technical level. So I would urge on the left hand side on the A I fundamental side. Yeah. OK. I think there'll be lots of people that will be uh wanting to sign up to that today. If me and Ryan included, I'm sure. So we'll definitely share the link with um all the delegates on the call. Um er The next question actually from one of our fellow I Huber colleagues. Er So we've talked about the idea of garbage in garbage out in terms of data. Um Do you could you envisage within healthcare where data is so vast and continually created that, that that new roles are becoming apparent of kind of quality data inputs? Yeah, that's what exactly then. So we call them as a prompt engineers. So uh I thought like maybe I would quickly use one small slide to give you the prompt engineering. So like think about, think about this is a small email. So hi John Day and this was written by Chris Holder and you want to pull the mailing address from this email, right? So the prompt is, this whole section is called prompt. Prompt is nothing but what you want to ask A I in combination with what is the text you are looking out for an answer from a particular question. So in this case, the entire text is called a prompt and you also have a response. So the response is 123 Microsoft wa this is the mailing address which a has found from this information. So this is called a completion. And then these are the individual text that is called token. So this is the latest uh I would say technical skill set. So how do you ask the right question here on the top that is extracting the mailing address and how do you get the right information? That is 123 Microsoft way. Uh W A. So this is really critical for A I where you ask the right question, you get the right response. That's really so and I think there's a super question that I'd imagine is gonna take us to the end that's just come in from lean. Um She says, or I think she or he um they say er has there been any cost benefit analysis for the introduction of A I into healthcare? What a stunning question potentially to end on? Yeah. So uh so it all depends on the use case, as I said. And then currently at Microsoft, what I'm doing is we are building a business case. So what it's called return on investment. So the typically would be S as I said, back office is savings in time savings and cost or saving in manpower and uh matching with what is A I is costing you. So if it doesn't make sense from the commercial angle, Cio doesn't sign that use case in your case because it doesn't invest in technology. So we want to make sure that it is met on the commercial angle. Number one, number two, it is meeting the regulatory needs. Number three, it is passing through all the security framework principles that I have shown earlier. So if the answer is three to all this checklist, then yeah, is the right solution for that. If one of them is not true. So we ask them not to use it because you can still use it but it will not add any value, commercial value or like security value. Yeah, I suppose the cost benefit for that for clinicians are slightly different though than in a commercial perspective. So of course, productivity efficiency utilization are all really key. But what we also want to see if you think about back end NHS. So we want to see those things but at the coalface, at the front end, we want to see, you know, safety um er in you know productivity in a number of patients that you see diagnostics, et cetera. So I think that what we probably haven't worked out and I don't know what your thoughts are Ryan on this is what does a cost effective, clinically effective A I system look like. And I think that's a much broader conversation to be had. Um But I think it's really interesting that perhaps we can get save cost in different ways that perhaps we haven't considered in terms of integration of A I into the healthcare system rather than in terms of clinically cost effective systems. Yeah, absolutely, absolutely agree. Um So that pretty much brings us to time. I think the, my final comment is to say that um er Titania and I have been talking ahead of time that, that our hope is that this is the first of a number of different events around er, surgery and A I including some more kind of hands on er, workshops ideas um which we're really excited. So please do er, if you were able to, to follow a link register, er, and we will keep you up to date with, with what um with what's going on. Er, and we hope to, that this platform will be a great bridge in terms of the clinical er er er frontline experience and input into the work that Microsoft and Open A I are, are doing. Um de we'd like to do the final. Thank yous and, and yes, thank you Ryan. Um So thank you to Ryan, who's our I hub lead, who always leads the way and allows us to um deliver so many of these events. Thank you to the R CS um webinars team, particularly Jane and Sean for supporting us through today. And big, thank you to Chetana. It's been really insightful and we've really, really enjoyed it. And thank you to everybody who's taken the time to spend, you know, 60 minutes with us this evening. We hope to see you at our future I hub events. Um And we hope that you continue engaging with us, please fill in your feedback form. Um We'd be delighted to hear from you and if you would like to reach out to us, we're all very approachable. Thank you so much. Everyone have a great evening.