Transforming Stress with Dr. Ash

From AI to Human Intelligence: Leading Healthcare with Purpose with Naiteek Sangani

• Dr. Ashish Kumar • Season 1 • Episode 44

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 53:08

Send us Fan Mail

Artificial Intelligence is transforming healthcare at an unprecedented pace—but technology alone is not enough.

In this thought-provoking episode, Dr. Ash sits down with Naiteek Sangani, AI Product Leader at Microsoft, board member, and advisor, to explore how AI can enhance healthcare while keeping clinicians and patients at the center of every innovation.

Drawing from years of experience building AI products, Naiteek shares why the future of healthcare depends not only on smarter algorithms but also on better leadership, thoughtful implementation, and a deeper understanding of human behavior.

The conversation moves beyond technology to discuss trust, decision-making, burnout, resilience, and why slowing down for a few extra seconds can dramatically improve the quality of our thinking and our lives.

Whether you're a healthcare professional, technology leader, entrepreneur, or simply curious about the future of AI, this episode offers practical insights on navigating rapid change without losing our humanity.

In This Episode

  •  Why AI is transforming healthcare faster than ever before 
  •  The biggest opportunities—and risks—of AI in medicine 
  •  Keeping humans at the center of technological innovation 
  •  How clinicians can work alongside AI instead of competing with it 
  •  The importance of explainable and trustworthy AI 
  •  Leadership lessons from building AI products at scale 
  •  Why healthcare needs systems thinking rather than isolated solutions 
  •  Balancing innovation with patient safety 
  •  Cognitive overload in the digital age 
  •  Why taking five extra seconds can lead to dramatically better decisions 
  •  The relationship between stress, decision-making, and AI 
  •  How curiosity and continuous learning prepare us for the future 
  •  Practical advice for healthcare leaders embracing AI transformation 

Key Quotes

"AI should augment human intelligence—not replace human compassion."
"Technology is only as powerful as the people who use it."
"Sometimes taking five extra seconds leads to significantly better decisions."
"Innovation without trust is difficult to sustain."


About Naiteek Sangani

Naiteek Sangani is an AI Product Leader at Microsoft, board member, and technology advisor with extensive experience building AI-powered products that create meaningful impact. His work focuses on responsible AI, healthcare innovation, product strategy, and helping organizations adopt emerging technologies while maintaining trust, safety, and human-centered design.

About Dr. Ash

Dr. Ash Kumar is a Consultant Physician, international speaker, executive coach, and author of The Boiling Frog: 21 Strategies to Transform Stress into Strength & Comfort. Through his podcast, he explores practical neuroscience, resilience, leadership, AI in healthcare, and stress transformation to help individuals and organizations thrive in rapidly changing environments. 

Connect

Naiteek Sangani

Dr. Ash Kumar

Ready to transform stress into strength? The journey starts here. 

SPEAKER_00

Join us every Friday at 5 p.m. And let's start at night together. Let's start with today's episode.

SPEAKER_02

Hello, friends. Welcome to the Transforming Stress with Dr. Ash podcast. And today we have got Natik Sangani with us. Natik is a principal product manager as at the Microsoft Health and Life Sciences. And we met last month last month at the healthcare AI conference in Anahim, California. And Natik was one of the speakers, and he was sharing how AI is actually transforming the healthcare space. So please help me welcome Natik.

SPEAKER_03

Thank you. Thank you, Ash. I think uh thank you for inviting me on this podcast. Um, it's my honor, and uh I'm looking forward to having this conversation with you. Uh, it was a wonderful session at the Health AI Summit that we had in Anaheim, and it was wonderful connecting with you. So, thank you again for this uh wonderful opportunity.

SPEAKER_02

So, Natik, uh uh when uh when you gave the presentation, which was so well received, you have been very much heavily involved in the space of AI in healthcare. But I want to start the journey with yourself. I know I very cordially remembered when you told us that uh you had a childhood passion of becoming a doctor, but that destiny had some other plans for you. And then how you completely close the circle again and you're back to healthcare, serving healthcare, and also now potentially making going to make an impact into millions of people in the healthcare AI space around the world. It is so inspiring, and it was a real real uh uh inspiration to meet you and know your know your ideas.

SPEAKER_03

Absolutely, absolutely. Yeah, I think as as a kid I was fascinated by the healthcare industry in general. I had lots of friends and um you know their parents were being doctors too. Um so for me it was a very natural fitting uh in some sense. But uh my cousins uh and um a couple of other friends from my school who were also into engineering. So that's where the natural inclination came from my parents. Um at the end of it, when I do a retrospective now and think of how my life is shaped, sometimes I feel I think my parents had the foresight of what's gonna happen. Uh and they were able to, you know, see a bit into the future than what I could have thought. Um but yeah, if I'm trying to understand even this detour, you know, trying to figure out how engineering has helped me um over the time too, I think that has played a pivotal role in my life. Um especially when I started working at Microsoft and I was working on the software systems in general, you know, I realized that we were building uh software architectures that could very securely process millions of data points per second. And when I was looking at my friends, uh, you know, my all my clinician friends, they were still manually faxing uh patient charts. You know, so if I had stayed in that clinical path, I probably would have been uh helping or treating patients which are like in my immediate waiting room. But uh the opportunity that I get with um uh Microsoft and the tech company is uh it gives me a very good vantage point uh to kind of see the forest for the trees. So basically, you know, we have the opportunity to build scalable AI solutions that can actually reach millions and millions of patients by essentially fixing the broken systems uh that our currently our doctors are trapped in.

SPEAKER_02

Really, really uh inspiring. Uh so Natik, just uh take us through the journey of your healthcare AI space, how it started, and how it has evolved over the last few years.

SPEAKER_03

Uh yeah, I think my first venture into uh health and life sciences work was when I was um work when I was studying at Columbia University, so right out of Columbia University, I got the opportunity to do my uh kind of internship. It was a freelance internship then, but I got it in the biomedical uh uh you know, the wing of the university. And my job there was to try and use technology to identify and distinguish between cancerous and non-cancerous cells, um, uh just with the help of you know an ultrasonics machine. So I spent about three months working on that. And uh towards the end of the internship, um, as a perk, I was uh told that now that you have successfully completed the internship, let's get you to one of the um you know uh operating rooms and have you dissect a mice and take a do a small biopsy of a white mouse that they had, a lab mouse that they had, and then biopsy that, put it under the ultrasonics machine, and then run it through your systems and see uh how do you you know differentiate between cancerous and non-cancerous cells. So that was how I got I got my first initial gig with tech and um uh bio uh you know the the biology side of things. Uh and then eventually I think we continue the path of uh computer vision and imaging and you know working on the engineering side of things until I got this opportunity here at Microsoft where you know back in 2024, I remember we were starting the era of foundation models. Uh everything was like, hey, let's create an AI model, you know, and instead of having these bespoke small models that would do you know targeted tasks, can we have one huge model that is capable of doing multiple tasks at a time? So that was the idea then. And um uh Microsoft had delved into this at a very large scale and not going into a particular industry per se. And um I took that opportunity to say, let's try into healthcare and see, you know, how this thing fits. We we bought a bunch of data, we trained the model with that data, and we saw some really promising results with that. And eventually, you know, um, I got uh I think I caught the eye of my uh very close friend and a mentor now uh by the name of Dr. Matthew Langren. Uh he's a renowned radiologist himself. So uh he offered to start a team within Microsoft. Um and uh he and I, with one additional person, uh uh a colleague of mine, we kind of co-started this team in the Health and Life Sciences organization. And um yeah, we have embarked a journey uh into the space since 2024 as a team, um evolving from a mere three you know product manager team to a good uh 2025 uh member team and uh working on models and now pushing it to the agent tick side of things as well.

SPEAKER_02

Thank you, Natik. Natik, uh if you don't mind me saying that most of the listeners, the healthcare professionals around the world, to them uh to them terms like data, uh LLMs, systems are pretty Greek and Latin. Still at this stage, we are at a very early stage. All we know is that we see patients, and many times uh we are accessing a lot of data, and a lot of hospitals have epic in different places, there are different systems, there is integration of uh computer systems in uh in a lot of countries, still not everywhere. So we are still at the preliminary stages. What we know not only in the US, in the UK and other places, that 60 to 70 percent healthcare professionals are having a huge amount of cognitive load and having a burnout. So we we know that uh there's a huge promise, and we are already saying that how AI is uh is really helping reduce the cognitive load in the healthcare professionals. So in today's episode, I really want you to go to the to the basics to help people understand that what is AI, how it is helping the systems. I know it might be a little difficult uh job for you because you are meeting people in the best centers, giving uh talks in the conferences, but in our listeners, there are people all over the world, all over the globe. But what they are having is a lot of data coming to them, a lot of information, a lot of uh challenges in the environment where they are making these decisions in the ground level. What will be the starting point for them to start understanding what is AI?

SPEAKER_03

Yeah, no, that's that's uh that's a really good question. Um, and to help understand, I think for any non-technical person, the way I would phrase this would be you think of AI as a tool that can help assemble a lot of information that is so widespread across the world and bring it to your fingertips, right? So one is uh coagulating and synthesizing information, that is one aspect of it, and after that, giving out pure sense as to what does this really mean, what does this information really mean, and how does it help take actionable or give actionable insights to doctors and radiologists and physicians, I think that is where the the strength of AI really lies, right? So, yes, there will be lots of EHR systems, there are lots of software systems that are being deployed in you know companies, but again, they're all built with varied tasks and you know, depending on some use cases and the asks from a particular institution, a particular hospital. But the the biggest challenge still exists, like even in US, you know, we we do see the healthcare system is a very broken system, right? Uh a particular institution or a particular hospital uh uses data, stores data in a particular format, which is not easily transferable or available, you know, across. So a simple example being let's say, you know, I did my master's uh from Columbia University in New York. So I spent about four years of my life in New York City. Uh, you know, I've gone through a you know a bunch of tests and uh given down my annual physicals over there, and then I you know switched over and said, let's move from New York, let's go to Austin, stayed there for six years, you know, ran a few tests there too, then came to Seattle. Now, when I'm doing this, and my doctor's like, okay, can you tell me your uh history? I have to at each and every instance, I had to repeat my history, you know, and transferring the data just became so difficult. Which I was like, how in this era of internet, why is it still so so difficult? There are times when I've still seen you know uh hospital systems use fax machines to kind of transport data, uh, you know, uh the the the patient charts itself. So that it's see these kinds of things are some things which there are still a lot of manual dependencies, and with the help of tools and service and with the help of AI, if I'm able to bridge that gap, make things, you know, even though it's available fragmented everywhere, but with the click of a button, if I'm uh able to assimilate, you know, query the right way and bring all that information in an actionable format, thereby reducing the you know cognitive overload for the um uh physician, that is the whole uh premise. Like we want to be we want to have AI to be used as a tool, as a think of it like a a highly talented intern which does all the pre-work for you, and you as an oncologist are now you know uh very well prepared with a concise list of the things in front of them and uh are able to be more human when it comes to a patient encounter.

SPEAKER_02

No, that's that's very insightful, very, very insightful. I will uh express this in my own uh uh language for uh the simplicity of the listeners. We have got uh smartphones, and we know that there are more smartphones in the world than people, so maybe maybe more than eight, nine billion smartphones, and also the same thing with internet connectivity. So the internet connectivity is also now pretty much reaching even the smallest villages. You mentioned that the American healthcare system is is a broken system. We all uh the all the statistics show that year after that, and we're talking on a global level now that there are so many countries around the world where the that the access to the physicians to the healthcare is very, very limited. But people have the phone and they have the internet, and that is where I feel that there are there are huge developments in the direction which we are moving that people will have access to at least a great starting point where they can be able to, where they are able to make sense of a lot of things around their health. Is that a reasonable way of summarizing it?

SPEAKER_03

Absolutely, I think it's spot on. I think the idea is to uh democratize the you know the the access of AI to everyone. So the use case that you bring up, like you know, I can I can go a tad bit more uh deeper also. Think of you know any rural community in a even in a tier one country or especially where you know where we come from, you know, places like India and uh Brazil, there are still uh you know, uh there is a lack or dearth of radiologists, that the space is wide, but the number of radiologists uh attending to patients, you know, that that ratio is very, very abysmal, right? So uh in in that scenario, you know, if a radiologist is not available for a span of a 200 miles of distance, but there are nurses that are there, and like you said, you know, if I'm able to you help and train and say, use a smartphone to take a particular picture, and if I have a model or if I have a particular product or uh AI running on the phone, you know, telling you here, point it this way, take the uh you know uh picture in a particular way. It does the analysis on the phone itself, saying, is the image quality you know appropriate enough that I can actually send it uh to a tele radiologist to kind of uh review and send it, you know, give the information back. I think that is something that becomes a really, really uh massive use case for all you know rural communities and trying to bring things uh much more closer and be uh overall as a system be a bit more impactful as well.

SPEAKER_02

Thank you, Natik. The other aspect which I would like to share that in healthcare, things have been happening for centuries, and the transitions are going to be slow. There is a there is a culture in medicine, and there is a lot of fear of the AI. If I recall correctly, is it reasonable to say that black box, the black box thinking, what is this mystery surrounding this particular thing? So, because most of the people are still in a very, very early stages, and also the mindsets have been honed for decades in practicing medicine in a very, very, very systematic way. So now it's the same thing is with the AI. So, how do we take the healthcare professionals and the clinicians from the black box to the glass box?

SPEAKER_03

Yeah, so let me put it this way. I think if you were driving a car and the GPS tells you, you know, go dive into the lake, what are you going to do? You're going to ignore it, right? So that is the whole premise over here, too, right? If I'm giving a you know a decision without any explainability to the doctor, then it's like, how do I trust it? Right. I think that that is why if AI becomes like a complete uh uh box where things have happened, some magic has happened, and spit, you know, it just spits out the final decision, then it becomes difficult, at least right now, because we're still in the era of figuring out how do we trust you know the software, right? So that is where the glass box piece came in. So when I mentioned, you know, we we there is a need for us to transition from the black box to the glass box, the idea was can I put in or give in the information that are needed if I can help the doctors understand the reasoning as to why the AI took a particular decision, I think that would you know enforce that um uh confidence in the system too. So something like, you know, uh there's a patient who's ready for a discharge, instead of a black a black box AI would basically say the patient is ready for discharge, you know, the doctor don't do anything else, just discharge the patient, as compared to me saying that um, you know, the AI system comes and says, I saw the patient uh notes, and even though the uh the white blood count cells, uh the white blood cells count um depreciated 30%, but it's over a period of 48 hours, which is well within you know the the uh regular safeguard limits of a clinical um uh uh uh outcome, then and hence I take the decision of saying, you know, it is or I suggest or recommend you to say uh discharge the patient. So that explainability will give the uh trust to the um uh doctors as compared to just saying discharge and um be okay because at the end of it the doctors are still legally and morally liable for if something goes wrong. So that explainability is why I say we need to move from glass box to uh from from black box to glass box, especially for AI and healthcare. And I totally understand, and there is a legitimate reason as to why there is a fear, uh, especially when it comes to an industry like healthcare, because even one issue, one fatality is just um takes us back to the drawing board, and at the end of it, you know, uh we are playing with human lives over here, so we have to be the bar is very, very high for uh AI and healthcare.

SPEAKER_02

Yeah, absolutely. I think the analogy you have uh shared is very apt. Uh that when when you're driving a car and the navigation tells you to drive into a lake, you don't do that, you use your sense. And I will use the uh example that uh that that when you're using the AI, embracing the AI, you are still the pilot. And let's use the example of co-pilot co-pilot because you are at Microsoft. So the AI is going to be the co-pilot. So the pilot is in the driving seat, is navigating the aircraft, whereas uh the co-pilot is just giving suggestions that maybe we take this, maybe this is the option, and this is the reason, this is the reason behind it. But the final decision still lies with the clinician.

SPEAKER_03

Absolutely. I mean it is pertinent that I think this is what we also refer to as human in the loop system. Um so uh we always I think I keep getting asked as a question saying, as a radiologist, as an oncologist, given how quickly you know AI is making streets in this industry, is my job at risk? I'm like, I can't say what's going to happen in the future, but at least for now. Uh no, we we are working at and we are always looking at making sure that there is always a human in the loop. You know, the human is the final decision uh maker. Everything that AI gives you is to be used as a tool. Everything that AI gives you is as a draft that you as a human need to review before making a final call.

SPEAKER_02

True. I think that is uh that is what uh of course it will depend the from speciality to speciality how much how much element of the human is required in a particular loop, but there are a lot of specialities where that element, the person-centered care, will always be very important because everything uh Natik, as you would agree, is about the context. And they are like if you go to a forest, there are a hundred trees, and the hundred trees are different. So even today, when practicing medicine, and that's the job of medicine that you are practicing, and that's why I say it's not the science of practicing medicine, it is also the art of practicing medicine because the treatment has to be absolutely contextualized to the patient and the family and what are the resources which are available to the patient and the family in front of you. And I if I see a hundred patients in a month, all a hundred are going to be different.

SPEAKER_03

Yeah, absolutely. And yeah, context absolutely does matter a lot, uh, you know, especially when it comes to healthcare. I I can tell you, I can I you know very uh vaguely remember uh an incident that has happened in the past as well where uh back uh a couple of years back uh you know I had a friend who had just gone through you know a regular um blood checkup an annual you know uh body checkup being done and there was this pathology report that have come in so we here we are on a Friday evening at like you know 11 p.m uh just having a jolly time and she gets a uh you know a pathology test results back on her phone she reads it and it's basically saying uh you know the report says infli uh infiltrating ductal carcinoma and uh so she suddenly just starts freaking out and we she doesn't understand what is infiltrating ductal but she knows it sounds really you know uh uh harsh and carcinoma the first thing is like cancer so now suddenly the entire weekend became like a weekend of you know terror and panic because we don't know what it is and you know doctors are off uh for the weekend no one's answering her until Monday but here's a scenario where I think if AI as a tool could help out you know explaining the layman terms and say that you know this is a common type of uh you know breast cancer but so uh here are the next actionable steps for you to go ask your doctors or take for yourself to take and at the same time next uh question uh actionable questions for you to ask doctors on Monday. So basically now you're managing that weekend of terror and panic and then converting that into saying you know this is a weekend of mental preparation you know and you're calming the patient down. So that releases a lot of stress too and that also reduces the fear of saying I don't know what the medical jargons are but you spit out something and now I've just completely lost it right so this is where I feel uh AI can really help bridge that gap between what medical jargons the doctors are used to talking or the EHR EMR systems are used to spitting out on their reports versus how it gets consumed by a layman consumer as well.

SPEAKER_02

Very very true.

SPEAKER_03

And I think uh going back to the same thing what we were talking about flying an aircraft and your example of uh um the navigation system so still early in the stage what do we think are the biggest mindset shifts the healthcare professionals uh will require to make to embrace AI that AI is a partner in their journey and uh a partner that is only going to enhance their own uh own um own presence and uh what we discussed when we met at the conference is augmented clinical intelligence and uh you mentioned about the human in the loop but the mindset shift is very important because we can be stuck in the old patterns of thinking right yeah I think there is there's a concept of um um you know how there is how there will be an over-reliance on the AI system versus how can I use the AI system to my benefit you know an over-reliance is basically you know you don't want the system to tell you nothing's wrong go over it skip it but uh you want to basically make the um uh system or the tool be um more useful or make the doctors be a bit more human and help them by saying hey while you're looking at this one particular XA or this one particular image you know uh ensure that you're looking at the top left corner also because you know I'm seeing something else so the idea is not to uh have the the the doctors or the geologists look away from stuff but kind of narrow while they're focusing on stuff make them add on and say I am covering for things for you that you may potentially miss so just ensuring that you know all of these things are being you know being being brought together as these things play out as doctors as radiologists as clinicians start using the system more and more the system again there is this term called uh a feedback loop in machine learning right the more feedback there are two kinds of feedback loop that are there that exist one is an active one where if something goes wrong you know the user kind of gives a feedback saying you messed this thing up you need to you know uh make make sure you don't do this again it's like training an intern. The advantage you have over here is that this is a machine this is not a human you're able to train with a lot of information you're able it doesn't tire out very quick you know easily and quickly it is very easy to up level the amount of data the amount of information you feed to the system and uh eventually it's like I have trained the system over time to ensure that it gives better and better and there will be a point again this no system is an ideal system so I don't expect a 100% accuracy rate at any system very soon anyway and until that happens we definitely need to work and you know work together with technology hand in hand and use it to our advantage rather than say I'm not gonna use it because it's being more of a detractor to me I'll wait for it to become you know uh like a God mode kind of a thing then I'll use everything once I have a hundred percent success rate then I'll use it but that 100% success rate will only happen when we get to you know using it in uh in a testing mode and we're able to give feedback to the system too. In our stuff and especially when it comes to you know um very um important uh verticals like in uh man uh like um healthcare and finance is where you know things are very very important what becomes very important for us is focusing on the accuracy of the model of the AI as compared to the speed I'd rather say I'm taking five seconds more to understand what is happening and give out the actual output than say everything needs to happen very very quickly so this way I can go over the patients uh you know uh more quickly and I have probably address more patients too so rather than speed I want to focus on the uh the accuracy of the model itself and for that in technology what we also use is something known as shadow testing right so while the you know for a particular scenario for example an AI system runs in the background and is doing stuff and it's just monitoring what uh the a human is doing it does not interact with the doctor it does not interact with the clinician at all but it eventually does like a uh um a uh retrospective of saying this was my output let me compare with what the uh doctor or you know the clinician had put in if it matches great I think I'm doing good if it does not then I need to update you know what went wrong and how it uh what appropriate steps needs to be uh be taken and once that shadow testing becomes you know more and more efficient and more and more um um accurate that is when we kind of want to take it out of that shadow testing and then put it into a proper diagnostic tool um that can be made available to the doctors thank you thank you but uh i mean the the process you have shared is uh i i i believe that that kind of a process is is true to any of the any process in in in when we embrace anything new we have to have both the support and the challenge and come with the feedback loops continue continuous improvement yeah absolutely the feedback loop is very very important uh both active and passive so like i said the active feedback loop is something which you know we as humans will give to the system and there's another piece which the system itself kind of monitors and sees what is working on I can give you a simple example of saying you asked me a question I responded to you did you continue asking me a question or were you satisfied with my answer or did I track and find out that you opened something else or you went onto the browser and did the same search again to kind of you know verify what I'm saying. So those are you know instances of passive feedback where my system can figure out what the your end user is doing and try to say yes whether my prompt was successful or did it really solve the intent that was put to uh that was put in play or did they did the user have to go through a lot of channel and that the uh the system itself needs to you know uh recalibrate thank you natik natik you have worked at the Microsoft for the last eight years now so my question to you is that over the last year last eight years how you have seen the AI transforming the healthcare the work the clinical workflows I know you've closely worked in the hospital as well so I want to see how the trajectory is going that's my first question and then the second question is where do you see the vision in the next five years? Yeah that's a pretty loaded question uh so let me take the first uh step um what have we what have we been doing at Microsoft uh you know especially when it comes to the health and life sciences right so funnily Microsoft I think has had a health and life sciences team for more than a decade now uh even being in the tech industry I was pleasantly surprised to say we have a health and life sciences system so I didn't know that uh until 2018 I think until 2018 when we started con uh having conversations about acquiring nuance now nuance is a big player uh in the radiology side of things for US healthcare so once that acquisition happened you know more of the health folks um started coming in and you know the widely used products like the nuance power share um the um nuance empower system and the power scribe um you know systems and then there's this nuance back so the dragon copilot uh dragon uh uh ambient dictophone that's there you know which does the speech to text uh kind of thing so these things already existed right but like I mentioned uh these were still not AI but they were still very targeted products that would do just one task um what I have seen internally at Microsoft is the pace of adoption of AI in general within the organization within the company has been really really great but there's still a lot of you know uh uh scope of improvement for us where we can say can we make this even more better can we make this even more widespread if you look at satya's recent uh you know um uh uh announcements yes co-pilot in general hasn't made uh that enough of a an adoption widespread uh so there is this new push within the org to say please use the um our co-pilot systems give us feedback let the you know let let let the uh product see what's working what's not working so there is a good push for us to get better so it is wonderful to see satya himself be so deeply involved uh in this it's taking you know uh the i i believe it's his top uh as a price zero as we call in uh tech stuff so it's his price zero to ensure that the adoption of AI increases because not just from a research standpoint but you know from a regular commercial enterprise and a consumer standpoint to the number of use cases uh that the AI can help solve is tremendous right and especially when it comes to healthcare we definitely know that there is a need for us given the scarcity given the uh the complexity in um you know uh some of the cases that the patients go through the the complexity or the the amount of knowledge that a doctor has and how wide the world is getting information from place A to place B. We have seen in the past that information either gets lost over time or takes decades to come out because you know people write journals and it depends on it on a particular oncologist who's anywhere very busy uh you know in a scale in his or her schedule I can't expect them to be aware of what's happening you know if they're if if there's an oncologist in US it's very difficult to keep track of what's happening you know at the other end of the world which is in Australia where similar cases might have come you know but things did not come here and we could have saved uh a patient over here but with the advent of internet with the advent of coming up with um you know um solutions where we can have gold data standards and you know a an easy way for us to transfer information and like I mentioned you know uh assimilate and come up with actionable insights for the oncologists or the radiologists or any of the uh genomic specialists too I think that will help speed up the um uh the process of uh patient care and uh I I particularly personally believe and I would love to get to a point where uh you know our doctors are spending more time listening and understanding the patient rather than you know listening and just typing away and not looking at the patient just typing away in notes and I've said I know firsthand information like firsthand experience rather where I've gone to a doctor I've you know told him this is what I'm you know going through and the doctor's not even looking at me he's just typing stuff and I don't know what he's typing right so maybe 30% of my stuff that I told him he's not seen it he's not understood it because our brains don't multitask very well so yes you might listen something you're processing something but you might type you know just a few things out and maybe some in important information gets missed out. So that is one of the um things which I would love for us to overcome and we do have some solutions like ambient listening which are there but it is still not the solution that I think is is the answer right now.

SPEAKER_02

I think the answer what's going to be there is going to be something known as ambient clinical intelligence where I can have multi modalities being taken in and by that I mean you know I have a camera in uh you know uh watching your patient encounter looking at you your facial expressions you know how you're expressing stuff if you're going through um you know any kind of uh emotion that that gets recorded and that gets noted as well there is the dictaphone or there is a listening device which is listening to your pitches trying to figure out how your breathing how your tone goes ups and downs notes those two and then at the same time the doctor is not even looking at a screen the doctor is asking you questions they are doing their cut check they're you know following stuff up they see their observations and add it to the notes that are there so if you're getting to that scenario I think that is the um uh world that we want to get in where doctors are more human the patient encounters them more like I'm talking to a human and I'm getting some feedback there and there is this clinical intelligence system which is taking these notes or taking all of this information transcribing them ordering labs making sure that these are the next actionable action items that you should you should be taking uh and truly kind of you know help uh get that uh the entire system work around the patient rather than patient working around the system I think we are making very uh we are making good progress this morning itself uh I was in uh John Hopkins program and we were discussing exactly the same thing and uh also I was reading an article the few hours back that we showed that the burnout in clinicians has gone from around 60% to 40% in just a few months at a center where uh what you're describing has been tested. So the trajectory looks very very positive and I agree with you that um the multimodal kind of uh input can pick up more nuances and things however it also comes uh there are also ethical considerations that when you know uh there used to be a physical uh sorry a law of physics which used to say that when you are under observation the behavior of what is observed changes remember what kind of law was that some but some uh some law of physics I remember very is it Heisenberg's uh uncertainty principle or something like that so the issue here is that when especially in mental health and psychiatric consultation you know a lot of a lot of health issues might be around mental and emotional health issues and if there is um that confidential patient confidentiality and again regulatory frameworks like in United States around HIPAA and other compliance issues also are very very important in this area and I know that that is a discussion for some other day because it's a huge topic in itself but again I see the optimistic side side of it Natik because there are far far more positives to this and then of course with anything new there will be challenges and we'll have to see how do we manage those challenges yeah no I think uh yeah data privacy confidential confidentiality and uh uh that itself is a whole forecast altogether right um but no I think whatever data we collect and whatever data we send to the model right I think there are three things which need to play out before all of these things happen.

SPEAKER_03

One is there's an end-to-end security right um second is that is end-to-end anonymity so rather than saying XYZ patient with the patient name and everything I'd rather say the patient this with these demographics and these these information go through this and the third thing is my models need to be trained on a diverse set of data. So you know the true cure will lie when I train my modules that comes from you know patients from Brazil from India from China from Europe from Australia from America all of these things with different race and different ethnicity all these things need to come together at a global scale and of course keeping the security pieces keeping in the confidential the the confidentiality pieces and keeping in place the uh anonymity pieces so that no one can reverse engineer and find out what's happening with uh X, Y, and Z person. I think that is going to be a very very very big thing that will be needed. So if I was ever given $100 million to kind of say go do this I will try to fill out a uh find a way to kind of create a global data trust which would do all of these things together because at the end of the day that this data is the fuel for all of these models. Models are nothing but you know uh just how do I connect these different pieces of information together and um bring it in front of you but the real oxygen for all of this thing lies in the data and um I can't yeah yeah if I need to get over any model biases or or ensure that I can correctly predict what's happening with a particular patient in a current scenario and or or predict what's going to happen later on as well then I need every possible side of the data in order to make those correct um uh pieces and at the same time you know while capturing this data is difficult there's another aspect which goes into this with this synthetic data generation right so similar things you know we also know when it comes to healthcare not all while we might have an OMOP data set which is like you know 70 100 uh columns data but at every visit I don't expect anyone to fill all 70 or 100 of them right so all those headers don't get filled but if there is a way for us to figure out how do we fill those information and how do we make the models more resilient to an actual real world data I think that's going to be very very key because right now all of these models are trained trained in like lab scenarios or they are these ideal data sets which don't uh exist uh in our world right so we need to ensure that these models are as resilient as possible and are able to either generate meaningful additional information or context to understand what's an actual and accurate context of what's uh uh what's going on um but that is something that is going to be a thing which goes post this particular wave of AI and uh education of models and agents yeah absolutely netik um what you had also mentioned that you know the data is the new is the new oil and but the oil is when is the

SPEAKER_02

Oil is screwed, it has to go through a lot of refining process, which you have been which you have just talked about the contextualizing and uh the refinement of the entire thing to be able to safely then uh deploy it.

SPEAKER_03

Yeah, I it's it's funny when we when you talk about this. I think very earlier on when we started this this team over here at Microsoft too, right? I think we very quickly understood, and this is one of the pictures which I keep telling all of the people that I work with. Like, you know, when when I'm working with partners, like why should I trust you? Or why should I partner with Microsoft? That's the first question that comes out, right? And my simple response to that has always been that I know there is never going to be one entity in the world which is going to uh be the clinical expert plus the technical expert plus a data expert. So if I was to solve a particular industry or a particular vertical, then I need these three players to play together. I can't have, I don't have that unlimited uh, you know, or uh infinite resources in terms of money, in terms of people, in terms of knowledge. But if I was to make this change and make this change impactful, then I need technical expertise comes from you know all of these hyperscalers like the uh Googles, the Microsoft, the Amazons, and of course the small startups too, right? They bring in a lot of knowledge. With big tech, they're turned taxes, but the this the speed of innovation, the speed of trying new things come off from these startups. These these tech startups are really, really great. In addition to that, we need to partner with hospitals, with doctors, with uh you know, uh clinicians, because that is where they have learned all of this information. They they see through you know patients on a regular basis, they had their gut check, they had their observations, and we have seen when it comes to uh a split decision between a human and a machine. I I don't know if you remember this movie called Sully or Captain Sully. Um, where there's this airplane, uh, you know, this pilot, and it's uh about to uh it's all about how he crash lands into the Hudson River, and there's this entire uh legal aspect as to the machine said you could have reached the nearest airport, but you know, all the simulations also prove otherwise. But turns out eventually that the machines end up doing only stuff that it has been trained to, but humans are able to bring in their gut their conscience, and that is that X factor that uh you know makes makes a huge difference, especially when it comes to these critical moments. So we need those specialities, we need that uh uh intelligence from the doctors as well. And of course, like I mentioned, data is the oil. I need uh non-biased and uh global data and varied data and diverse data set in order to make sure that all of these things are amalgamating together to uh to make sure the AI or this particular uh product or uh tech is impactful.

SPEAKER_02

Thank you, Natik. I'll have to uh the the movie you mentioned. I've heard the story, it's really inspiring and have to check it out. Now, Nathan we are coming to the top of the hour, and there's so much to cover. There's so many things I learned from you when I attended your talk at the at the healthcare AI conference at Anaheim, and uh but today we wanted to have more of a bigger picture for the healthcare professionals and for the um physicians around the world to understand what direction is the AI going. So I think you have given a great bigger picture today. Now, uh what is the message of hope you would share with uh with uh people that how the future looks like? What do you think is that in the next five to ten years? What is your vision?

SPEAKER_03

Yeah, uh I think I briefly touched upon this. I think for me, I would love uh for the industry in general to kind of I I probably am gonna sound counter uh to what the entire world wants to do, but I want us to slow down a bit. You know, I want us to focus on the accuracy uh of the work that we are doing rather than speed of innovation. Uh take a step back, make sure things are actually being useful, helpful, and impactful, and doing the right stuff, uh, and then eventually move over and pivot over to speed. So as long as the regulatory boards and businesses, uh, you know, they they are wanting to work and use AI to help patients and not businesses and not run it as a business, uh I think we are in good hands. Um so for me, I'm a very optimistic viewer, and I love to see, irrespective of every business is saying I'm here to make money, uh, we are still the primary goal in healthcare is patient-centric. Patients come first, businesses come next.

SPEAKER_02

Of transforming, of transforming healthcare and the challenges we mentioned earlier, not only in the United States, uh UK, but on a global scale, if AI is the new fire, and if it can help us to transform healthcare at a global level, I think we should make every possible take every positive step in that direction. Now, Nathan, it's been a real uh inspiration to meet you in person and have this conversation. I wanted to cover a lot of things, but in our first episode today, uh we just wanted to discuss the bigger picture, and you have certainly shared some remarkable insights. If uh and this is a work which will continue, it's the as you said, that we are in all this together. We are in this all together, and uh the engineers, the big corporations, the hospitals, um, the physicians, the data scientists, there's so many other specialities which I even don't know because I am new to this area, relatively new to this area. So if uh if listeners want to reach out to know more about your work, uh what would be the best avenue?

SPEAKER_03

Yeah, I think if you look at uh Microsoft's and just search saying Microsoft Health and Life Sciences uh platform, you should be uh able to reach out to us. And uh my team in particular is called AI Frontiers. So I focus primarily on uh AI model development and agent execution development. So every new feature or piece of tech when it comes to uh AI, that has been my bread and butter, and that has been something that uh I am very passionate about. Uh I work very closely with the research teams over here. We've got a wonderful pool of you know, uh more than what uh 2,000 uh researchers just working, you know, nonstop on uh the health and life sciences side of things. I yeah, I am taking an immense pride in working with them. And my job, which I uh try to tell everyone, is to accelerate what they do in research and bring it to productization as quickly as possible so that everyone can benefit out of this. And this is not just a uh archive or a uh Jama paper, and it's not just a publication, but um, in in true sense, I want to have uh these um uh scientific researches, uh discoveries uh be brought um uh in front of people and uh uh may make sure that they're able to utilize all of those um capabilities.

SPEAKER_02

Thank you, Natik. Uh we will really look forward to our next conversation with you, in which we'll make a much more deeper dive. And thank you so much for taking out your time, and I look forward to seeing you again.

SPEAKER_03

Thank you so much. It's been a pleasure. Thank you, uh Ash, for inviting me.

SPEAKER_00

Very welcome. If you enjoyed today's episode, we'd greatly appreciate it. If you could leave a five-star review, a like, or subscribe on Apple, Spotify, or wherever you listen to your podcast. Your support helps us reach more people. We'd love to hear your thoughts. Don't forget to comment up to check it out. We can't wait to help you with us next time. Remember, it's not the stress itself, but how we rise above it's definitely next time.