How AI ls Reshaping the Future of Healthcare

The Life Sciences DNA podcast is sponsored by Agilisium Labs, a collaborative space where Agilisium works with its clients ranging from early stage biotechs to pharmaceutical giants to co -develop and incubate POCs, products and solutions that improve patient outcomes and accelerate the development of therapies to the market. To learn how Agilisium Labs can use the power of its generative AI for life sciences analytics,

To help you turn your visionary ideas into realities, visit them at labs.agilisium.com.

You're tuned to Life Sciences DNA with Dr. Amar Drawid.

Amar, we've got Dr. Sabodha Kumar on the show today. For people not familiar with Sabodha, who is he? Dr. Sabodha Kumar is the professor of statistics, operations, and data science, and also the founding director of the Center for Business Analytics and Disruptive Technologies at Temple University's Fox School of Business. He has a secondary appointment in information systems.

He was awarded the Informed Information Systems Society Distinguished Fellow Award in 2023. He was also elected to become a Production and Operations Management Society Fellow in 2019. He's published over 240 papers and authored two books. And what are you hoping to hear from him today? He has written extensively on applying statistics, analytics, including AI and machine learning

to improve healthcare operations and outcomes. I'm hoping to get his perspective on how AI and machine learning can be used to improve patient outcomes. So I'd like to also get his perspective on areas of concern, particularly about the ethical use of generative AI. And a reminder to our audience before we begin, if you want to keep up on the latest Life Sciences DNA Podcast episodes, be sure to hit the subscribe button. If you enjoy the content,

we encourage you to hit the like button and let us know your thoughts in the comments section. And if you want to listen to this while you're on the move, an audio only version is available on most major podcast platforms. With that, let's welcome Subodha into the show.

Subodha, thanks for joining us. We're going to talk today about AI, how it can be leveraged to improve healthcare, and what precautions we might also need to consider. So you've mentioned that AI helps in healthcare in a lot of different ways, in assisting HCPs or healthcare professionals, replacing some tasks, giving solutions to people. Perhaps you can start by providing us with the different ways in which...

health care are using AI today? Sure. Thank you for having me, Amar. So, AI in health care is coming up in different ways. One is more on the operations side. So, AI is helping in improving health care operational efficiency, and we can talk more into specific examples, but you can think like, how can I serve my patients faster?

How can I utilize my surgery rooms better, right? More efficiently. So that's the one side of use of AI. The second side where we see a lot of noise and a lot of press releases is on the clinical side of use of AI. That how physicians are making the better use of AI to improve the healthcare delivery. But the third piece, which is also...

actually getting a lot of prominence is how to deal with customers, like customer service and getting answers quickly, right? Using a lot of chatbots and all. And AI is playing an extremely important role in that aspect as well. So we are seeing AI in almost every part of healthcare. And if somebody asked me today,

which is the one industry which will get most impacted by AI, then without any doubt, it is health care. It is at number one. And there's a good reason to say that, because we are seeing applications, and we see potential where we can make huge improvements. So it is doing in delivery of services, the better...

better assistance to physicians, as well as providing good customer service. But another important aspect in health care where AI is playing a very important role in pharmaceutical side, in drug discovery, and as well as making the whole research around pharmaceuticals more efficient. Yeah. Now, as AI is coming up and I mean,

what like we in general population we think about in the healthcare settings, it's a bit, they're a bit, it's slower to take on new technologies. or so. So what are your thoughts around that? And then to what extent do you see already AI being integrated in healthcare practice? So I will partly agree that they have been slow. In fact, I will say in some aspects of healthcare,

the AI is coming faster than any other industry. But I will also agree with you that some of the adoptions could be slower. Now that is by the nature of the industry. Healthcare has not been in the forefront of adopting IT in general. And the resistance coming from the system is complex. We have the whole physicians, hospital system, players and all. And many of the people, you know, some of the...

the people who are hard to change are physicians or sometimes even academicians. But these are the people who don't like to change the way they operate because they think they are the best in their field and they should do the way they want to do. So there's always a resistance to change. But the reason why some of the places it is faster than other places is very simple because there was so much inefficiency in health care

that we had a lot of room for improvement. And also here, we are not talking about just saving a few dollars here and there. We are talking about saving lives. So the implications are also higher. So just to give you an example, where we are seeing the real use of AI in health care is already AI is bringing health care closer to patients. Now,

In the US, the rural health care is not as bad, although still there are some challenges. But worldwide, there are many countries where the patients in the rural area are not getting the health care services at all or very low quality. Now, AI is playing a very important role. Like you can take pictures of your eye and it can give you the good solutions in the time. Telemedicines there, AI is helping us in analyzing that data in the real time.

So a lot of AI assist system or even during surgery, tele ICU is becoming a real thing, right? It started from the US, but it's gone to many countries. So many of these things we are already seeing a lot of AI, a lot of AI. And that's what we work with many of these companies who are on the top of the line in doing that. Now, where we are not seeing as much of AI is that still dealing with the...

patients like having a good portal is still not there. Healthcare systems are still struggling with having a good portal and having AI features there. So first, why AI is not there? Because we have to first have a good portals. So we are working with some of the systems like Penn Medicine. We have a big project. The whole goal here is that how we can improve the app system with access to both physician side and patient side. Right? But

when we started working on this, we realized that we are way behind in that. I'm not talking about the US, the healthcare industry in general worldwide. US is not advanced, I will say, in healthcare IT part. But because of that, we are not seeing a lot of AI because we don't have a basic system. But in some parts, we are seeing a lot of AI implementations. Now, when we think about AI, there's a lot of different aspects of that.

One thing that comes to my mind is the image analytics, right? So you have your pathology, images, radiology, right? Images. And this is like, I mean, a lot of the big revolution started in deep learning with really the image analytics, right? So do you, like, how do you see the progress in that area? And do you see over the next five to 10 years, this

entire field of pathology or so will be completely revolutionized where the AI will be doing a lot of the diagnosis or so? What are your thoughts there? Good point. And I will connect it with a few other things. Image analytics also comes close to voice analytics, but I'll come to first image analytics. So there are a couple of things happening in image. First of all, the answer to your question is yes, we already have a lot of things out there.

And we have a lot of solutions which is already being implemented or have been implemented. And in the next five to 10 years, we will see lots and lots of this image analytics coming, which is helping different parts of health care. What are these solutions? One of the projects that we did, and the whole idea was that, can we look at the images of the...

for diabetic patients. And one of the challenges we saw with these patients was that they will not go for diagnosis because it's a very intrusive process. So we did thermal imaging for that. And what we found is that the results were fantastic in predicting what kind of treatment the patients may need and also even helping in understanding whether they need to go through the intrusive process. Right?

So thermal imaging is there coming. It's already there. Many hospitals are experimenting with that. Second piece of image analytics is coming in with all the radiology reports and all. So how it is helping when generative AI came one and a half years back with Chat GPT into the limelight, now with GPT -4 and all, what hospitals are right now experimenting with is they take these medical images, they take the image analytics and fill up the forms automatically. OK?

with somebody who can go and then look at it to make sure. Still we have to look at it because generative AI is full of problems. We cannot directly put it in action. Will we fix it in a year or two? I don't think so. Five to 10 years, we'll have more confidence. But see, health care, we have to keep in mind. In our lifetime, we are not going to replace physicians or many of the tasks. This will mostly be AI assist, at least in the near future. Right? So, but...

they can reduce a lot of time in taking this data, filling up this forms, right? So a lot of that image analytics we are seeing, and that will reduce the load on healthcare system what we have, again, worldwide, but more so in the U US, right? We have a lot of staff shortages and all. So we are already seeing that. This is already there. This is not a futuristic thing anymore. What we are doing is that how to make it better. That's all we are working right now

because many of these things are not as good as we want it to be. And this is also helping in disease diagnosis as well as predictive maintenance or predictive treatment, I will say. So a lot of these things is already happening. In the future, we would like to see more biometric kind of systems for even the health care data and all. And we are seeing some of that.

But I think we can get better with that. Sure. And we talked about the image analytics. How do you see in terms of the machine learning or so coming into the clinical decision making at this point, or the algorithms becoming, the new algorithms coming up that will take inputs about a lot of

different health parameters of a specific patient and then think about what could be the prognosis or and what could be the potential drugs. So do you see that becoming a reality? Where do this kind of machine learning stand at this point? So where we are right now is that we have now systems, so for example, take an example of gastro, let's say, and

the systems are being done and a lot of that is done at my center and I collaborate with a lot of different healthcare practitioners and what we are trying to see that can we create a large language model based system which is trained on the specific data from that domain. That's very important, right? So we have general purpose large language models which is used by the generative AI. The problem is that I don't think there are any...

that are called to be good. They are quite mediocre right now. And the problem has been that they are trained on general purpose data. So one of the projects that started more than a decade back was that can we start collecting cancer data worldwide and try to understand that. Think of a system where oncologist is going for treatment and they get a list of recommendations from this system, which is based on the

data which can be a lot of unstructured data which has been analyzed by the system, which is almost impossible for even top oncologists to go through. So the cancer system come up with this kind of things and provide recommendations and that is already there. So again, the quality of recommendations can be better, but we are already seeing that this AI assist thing is coming in clinical decision making where they say that, well,

you may want to do this additional test. You may want to try this, right? And they're not going to replace physicians in the near future. No way. But the lot of decision making will be done. A lot of recommendation. Think of it like Amazon recommended system that surgeon is in the room. They have limited view. Now it comes with five different things that they might not have thought of. Right. So a lot of those kinds of systems are being done. And I think that's a very good starting point.

Like I gave you example of thermal imaging, right? So those kinds of things we will pursue, the pulse sensing algorithms are being developed. Even EEG data looking at the brainwave data. If the algorithm can see something and connect it with the past data and give some recommendations in the real time, that can make a huge difference. If I understand how brain wires are moving, that can make a difference.

And for our audience, when you say thermal imaging, it's basically the heat coming from the different parts of the organ. Yeah, yeah. So body part take that heat and then create that thermal image. And we have sophisticated technology to analyze that because different colors can indicate different things. Different, OK. OK. And so what you're describing is that more like a co -pilot, is that this is a co -pilot system that is assisting

the physicians, right? As the physicians are making decisions, the copilot is assisting them about, okay, well, have you thought about this? Maybe this makes sense. So they're kind of like, it's basically like a personal assistant that the HCPs have to make the right decisions. Is that a correct kind of thinking? Yeah, yeah, yeah. I hope you are not referring to Microsoft Copilot because their product is called Copilot. You are telling Copilot. Yeah, yeah, yeah. Absolutely right. Absolutely right. You know, this...

and they're not going to be pilot in the near future. They're going to stay as co -pilot, but they are very good co -pilots. And that's where the push is right now that there's a one side of AI machine learning or generative AI is that create good systems, which Google, Microsoft, they all are struggling to have that. But there's another side that can we create a domain specific models in healthcare, which can be a very good co -pilot. But...

you are absolutely correct. That's where we are heading. We already have some breakthroughs, but we plan to have more. And that will happen pretty fast. So next one or two years will be very interesting for that, for this co -pilot -based systems. Yeah. So it's basically, these are the agents that we're creating. These are digital workers. And so there could be like an oncologist digital worker, right? Someone who has a lot of knowledge in oncology

who is then assisting the oncologist in the clinical practice, right? Or there could be like a gastroenterologist, right? So like these digital workers, we really get trained into a lot of the data and a lot of the knowledge in that specific therapeutic area, right? You're absolutely right. And one more thing I want to add here, which is important to understand. So one side of these AI systems is that they will help physicians

in giving you very specific guidance on, you know, you can do this, you can do this you can do this right. That's the one side of it but can we somehow approach it differently altogether so that that I'm talking about more holistic changes in health care and let me give you what do I mean by holistic change. So right now if you go to a treat breast cancer patient yes so you have a list of standardized doses you will. pick the

You go with the dose type one, dose type two, or dose type three. And we see very similar thing if you go to take loan from a bank. What they do? They create a bucket for you. So they put you in a bucket that you will get 3 .25, 3 .5, or 3 .75. They don't tell you you will get 3.394321. They don't give you. Why does that happen? That's the limitation of data and algorithm.

What AI and especially generative AI is going in the direction is that can we create mass customization in healthcare? What it means is that a breast cancer patient comes and then rather than telling them that you will be part of these doses, I create a personalized treatment plan for you, which is only for you. Now we actually did one of those works. This is a published work. Anybody can find the search, right? It's on breast cancer.

So we work with some leading hospitals and what we have created there is a personalized treatment plan for breast cancer patients. We have created a framework for how these cancer hospitals can take that and build based on the data. You need a lot of good data for that. More recently, I work, my center work with a team from MIT. We got a grant from MIT, which includes people from Sloan Business School and

the science team from MIT, and we have done the similar thing for ovarian cancer. And there we are taking a lot of genome data to train our model. So we just finished that work. And the idea is very similar there, that can we take all the genome data and create personalized treatment plan for ovarian cancer patients? Our goal is that, and we are working very closely with oncologists on that, because we have limited knowledge, all of them trying to learn.

But you know, we can't learn without going to medical school as much as they know and their experience. But one thing we have realized is that they agree, they all agree that there is a gap. They are not satisfied with what they are doing. And we say that, can we come in and help you in doing that? So it is not only going to help us in giving assistance to physicians in helping what they are doing, it can also help in changing the way we do treatment.

And this personalized treatment or mass customization is the way to go. If the fintech companies can find the exact loan for you, if Amazon can give you exact recommendation of what you should buy, if Netflix can tell you what movie you should watch tonight, why can't we try to bring some of that in health care? Absolutely. So the promise of personalized medicine that we have been talking about for two decades now, I believe.

Do you think that is going to get realized soon with the generative AI and now with these new technologies that are coming up? It's already being realized. I will not say that we will see that in full flow that everything is being personalized. No, that will take several decades. But in pockets, we have started seeing it. So what we were talking about two decades or so is becoming a reality.

And thanks to what all happened in last five, six years, or even last one and a half, two years with large language models, that is becoming... See, we knew this is important. We were just not able to get good models. But now we are able to do that. So you're right. We will see that as a reality. We are already seeing that as a reality in pockets. Based on what you've been saying, it looks like generative AI has really penetrated

quite a bit in the last one and a half years in a lot of the healthcare aspects, it looks like. Of course, you know, this is if you look at all the Gartner, Hive curve or any other thing, no other technology has gone this fast. Actually, one of the senior guys from Microsoft AI, he was in my class and he was showing one of the graphs. And I think he told I could share that, that things that internet took 18 years,

this generative AI took three months to achieve the same thing. And even with the more sophisticated thing, I will call traditional AI took several years, it took a few weeks. So, and what we are seeing is phenomenal. We have not seen anything like this. We thought we saw worldwide web and social media and that's it. I think we have seen everything. No, this is tip of the iceberg and...

please do not forget it is just one and a half years since it all came. But what we will see next few years will be phenomenal and healthcare will be I will say, the most beneficiary of these kinds of things among everyone else. So let's imagine, right? Like let's say you have a crystal ball and you're looking at it, you know, healthcare 10 years down the road.

And I know it's very hard to imagine, but where do you see like the greatest change in terms of, so you talked about the operations, you talked about the clinical aspect right? So how, like, can you paint a picture of how the healthcare of tomorrow is going to look like with all of this AI? Yeah, I love to do that. And I love to be wrong also, but...

I think I can give you a pretty good picture of what I have in mind. See, when we go to the doctors right now, one thing because of electronic medical records and other things, a lot of time goes into putting that data into place. But what we have achieved in the last 10 years or so is that data entry has been a lot smoother. We have systems like

it can transcribe when doctors are talking, right? I was with one neurologist, he was showing me how good his transcription system is. And he was telling the editing he needs to do is minimal, right? He needs to still do some editing. So those things are there. So we will see a lot more of that in next 10 years. So we'll see a data entry part getting quite smoother. Okay. Another piece that we will see in next 10 years is that both

access of accurate data and timely data to patients as well as different hospitals. So for example, my data with Virtual can be much easily shared from the University of Washington Medical School Hospital, which is sitting on the other corner of the country. So those things will happen next 10 years because what happened is that we started this data sharing part in, I will say mid -2000.

So we have been around 20 years doing that. But we have lots and lots of challenges, which was very good. Actually, a lot of of PhD students graduated doing research on all this health data sharing. So it was great. But what we realized is that there were some of the basic things we were not able to do, either how to share it following all the HIPAA rules, or how to get everybody on board to do that. I think we have...

gone past those obstacles. And where we are right now is that we are seeing the fruits of that. So now once we reach this point, we have reached the inflection point in data sharing, and we will see a huge growth in that in the next few years. So that's what you will see. You know I have lived in the world. Even five, six years back, I had to fax all my data from one hospital to another. But those things are already gone or going away. So we'll see a lot more of that.

The third thing we will see is that some of the basic tasks of recording your healthcare information or even analyzing them, that could be done by variables and other things. So we'll see a lot more of that happening in next 10 years. So a lot more data collection coming directly from the patients. You don't have to go to somebody to do that-- done by the automated system. We will see a huge jump on that as well. It's already happening. So next 10 years, you will see that.

So this will be all great. What we will see where we will not see the complete fruit of what I'm talking about AI assists, right? Now there, there is a pushback on that from physicians. They say that, do they really know better than me and so on, right? So it is not just about technology. It's a lot of cultural aspect, right? Now that takes a little longer. So in next 10 years, I hope to see a lot more of that, but I don't expect that it will be totally

transformed or changed. What we will really see is that more and more hospitals pushing for these kind of systems to their physicians. So larger hospitals are looking at it very differently. But the smaller ones will have little challenge. So we will see the gap actually increasing on that. So I will say that we will start seeing a lot of these AI assist systems at the decision -making side. But it will not totally transform in next five to

10 years, it will take a little longer than that. But the customer facing side, the access of data, the ease of use of these facilities, and some of the basic documentation, image analysis, or the radiology reports, those things are going to be pretty solid in the next 10 years. OK. We've talked a lot about all the great things about AI. I want to now

go to a bit of the darker side. So there are concerns about generative AI potentially replacing human healthcare providers. And you mentioned that in the near future, you don't see them replacing, right? But can you talk a bit more about that? Like why it will not replace, there will be forces to replace that, there will be forces against it, right? So it would be great to get your perspective there.

We have to understand when I say that it will not be replaced, I'm referring to the physicians are not going to be replaced. And when I say near future, I'm talking about maybe our lifetime. I don't see any way physicians will be replaced by this system. However, we already have this temperature measuring robots, right? So you go to the hospital, nobody has to take your temperature or your weight, right? So what we will see,

some of the replacement is in some of those tasks, like some of the tasks done by nurse practitioners or even some of the staff. We will see some of that being replaced. But I will not call it a replacement. In fact, I always feel, especially in healthcare, we have so much shortage of the good quality people that this will not be really replacement. They will be doing higher quality things.

They should not be doing some of the things they are doing right now. There's no reason we are in 2024 and they do things which machines can easily do. So the concern should be how to retool and retrain our health care staff and health care practitioners rather than worrying about replacement. We don't have that problem. It does not exist. And we have some recent studies and research which have shown...

that this is mostly in our mind. But if we do not start retraining, we do not change the process in which we train in our medical school, nursing school, that would be a problem. So we need to worry about that rather than worrying. So we should be proactive on what should I be teaching to a nurse who will be working 10 years down the line and what could, what should be the right skills for them to have. If we have that mindset, I think we are fine. Okay.

Do you think that there should be like a clear line to be drawn between using generative AI as an assisting tool versus relying on it to make critical medical decisions? Or do you think we don't really need to draw the line? It's not really necessary. No, I think there should be a line. See, healthcare is one area where regulations are for good reasons. And I think...

if we don't draw a line, it can create lots of problems. Number one, there was a study done by University of Maryland, and I also participated in part of that, that if you go and ask Chat GPT, so in this experiment, they asked Chat GPT about breast cancers. And then they validated that with experts. And they showed that it was 94 % accurate. Now, problem is that if we don't draw the line,

and people say it is 94 % accurate, then, you know, why do I need to go to see oncologist? I can rely a lot, right? There was an interesting article in Wall Street Journal or New York Times. And they were talking about how Dr. Google is being replaced by Dr. Chat GPT, right? And I think that's a serious concern, right? Why? Well, you say it's 94 % accurate. Aren't oncologists also not a hundred percent accurate? Why do we need to worry? There should be need to worry.

The problem is that some of those 6 % errors any oncologist will never make. They hallucinate. They will tell you something that is totally nonsense, and that can lead to serious implications. So in the settings like health care or financial sectors, we need to draw a line where these models should be used, can be used. In fact, I will say that they should be even draw a line on what kind of results

the general purpose systems would even give to the patients or consumers. I think if they don't, it can have a lot of serious societal and ethical implications. Now on the doctor side, even though some doctors and physicians will be hesitant in using, some are very excited about that and they may go overboard on that. So again, I think we should have some boundaries. How to draw the line that is a huge challenge even though,

you know, telling this is easy, defining those rules are harder, but at least for next few years is better to be conservative without tempering the innovation. You know, and I'm telling this because half of my research is in healthcare AI right now. And if I say this, then that will hurt my research as well, because some of those things I have to be more careful. But since I have seen this, I can say that we need to be a little watchful and careful. And,

it's okay if you get a bad recommendation for a movie, but it's not okay if even one thing can take your life, right? So I think drawing a line is critical and this should come from policymakers, hospitals, healthcare practitioners and general users. Absolutely. Now, what guidelines or best practices should healthcare providers follow

when using generative AI to ensure responsible and ethical use? So the very first thing is that they should be very clear on the explainability . I think that part is very important. It should be very clear to the health care practitioner or to the system is how this model is giving this result, what kind of data they use. And to give you a simple example,

we have an NIH grant right now. There we are working on pulse oximeter. And it's a very interesting project. It is well known that pulse oximeter gives inaccurate result for non -white skins. And it has been known for quite some time. And the reason was that because the model was trained on certain skin colors. So in this project, actually, we are trying to

create a system where it takes their skin color and then, you know, do some algorithm based on that and so on, right? So the thing is, the point here is many of these things, if we do not know how the model is trained, if let's say I'm treating a patient from a certain reason, but the model is trained based on the people from different reasons, and the outcome can be very different, I think that can be disastrous, right? So...

I think explainability of the model is very, very critical in health care before we start using poll decision making or assistance in decision making. That's the number one part. The second part is that also we have to worry about what kind of data is being used. Although it's very good to train them with a lot of data and all, but we should not lose sight of privacy of data.

Because if we are not worrying about that, that can lead to a lot lot biases in the system. So how can I use data in a certain way that did not add already a little bias to system? I will say that our systems are not unbiased. And we should use AI to remove those biases rather than adding to those biases. So how those data is collected, how the models are created, it should be very clear. Then, as I told earlier,

very important aspect is that how we use this. There should be strictly recommendations for the time being. And time being could be a very long time period until we get a lot of stability. We should never ever give an impression to the physicians that this is it. This is the God sent message and you should take it for granted. Because...

some of the physicians may get influenced by that if you give those kind. So that messaging and education and training is extremely important. That's how to rightly use it. What is the purpose of this? Purpose is to whatever you are doing to give you another eye rather than replacing with your eyes. And that messaging has to be repetitive. It has to consistently go out. Otherwise, 10 years down the line,

we may see physicians using the system that can have disastrous results. So unless we do that, I think we will be in big trouble. So I think these are how to use data, explainability of these models, and then the training for the users of these models. That is very, very important. And please do not forget, I think, that we still have to worry about accountability.

I think if something goes wrong, the accountability should be well defined. well.

Yeah, that's a great point. And how, like, so you brought up this issue about the data security and the privacy, right? So especially with the HIPAA data that these models will have access to, right? And then we know that these large language models don't behave exactly right. There is some probabilistic behavior even if you try to control the parameters, right?

So how do you see that? I mean, that can be a big issue if the HIPAA, using some prompt engineering, if the HIPAA information can be taken out of them, right? So we need to relook at many of these rules, by the way, when we go through these models, we should be careful when physicians are using this open access Chat GPT model,

where their data is going. And I know, I know for a fact that some of the physicians are going and putting questions there. I don't think it's a problem because just by putting questions there does not mean they are violating any rules because it may just go to train models. But we need to relook at how HIPAA is defined. Also, we need to have guidelines about who should be putting data at what place. And how you know this problem can be solved,

if these healthcare systems have their own systems, which can be trained on general purpose data, but also your data stays with you. Like, let's say you are Penn medicine and you sign a contract with Microsoft right now, Microsoft will not take your data to train their other models. They are not allowed to do that. Right. Now those kinds of things should be well defined. Otherwise,

things will get muddy and we have to relook at our HIPAA things as well, right? We can't just work on the old world because these are the new things. We have never seen this. So a lot of redefinition and rethinking is needed in the data part, keeping in mind that we still want to secure the privacy, which we have been doing well, but we have to redesign the system. Do you see some parallels here between

the regulation for these and the regulation regarding this autonomous cars that, you know, it's kind of the wild west, not really sure what's going on. And so like, do you see any parallels about how, especially the policymakers think about these? I'm glad you brought this up because I was going to say, bring that example because you know, right now Tesla can drive for you, right? And,

I was trying this out last month actually with trying on their own completely. I usually do like partly auto with this test lab but I thought okay let me give it a try and I could say that I was only 30 % pleased because some of the lane changes and all I don't think were that safe. Even some of the stops were not safe. So there are certainly a lot of parallels because both are dealing with lives.

The difference here, I will say, it's your choice in case of cars whether you want to put them in auto or not. In health care, many times as a patient, you may not have a choice. You are being treated by a physician who is deciding to rely a lot on technology. So a lot may not be in your hand. So the regulations need to be stricter than what you have for autonomous vehicles.

Because still the Tesla can come back and say, well, if you don't want, don't do any of this. I'm not forcing you to do that. That's your choice, right? But it may not be as simple in healthcare. So what I will say that there are a a of panels, parallels, the regulations need to be stricter in healthcare. Yeah, and also the data is at stake as well, right? The private data is. And how do you see the government,

and the policymakers, do they really understand this? Are they taking the right education from professors like you, to really understand this, to make this like, where is the government and where are the policymakers? And do you think they're making the right decisions? So this is where I will say US and partly UK are ahead in the game. So because of the culture of the countries, so they have

done a much better job in collaborating with researchers, practitioners and creating this committee. It's like I told you about that grant that we are working, right? Many such things are happening. I have also talked to some White House committee on some of these policy things. And I know some of the leaders in the academia are working very closely on this team. So yes, if you ask me

do all the senators know about it? No, they are struggling with how Google gives ads, some of them. But some of them are pretty well advanced on that. But I think the right thing that is happening, they are taking help from right people and they have a good committee set up. And that is a positive direction of this happening. But even researchers don't have complete picture yet, you have to realize. But I will say that there are a lot of good committees I'm aware of.

They are looking at it in a very different perspective and serious perspective. But many large countries like China and India, they need to think a little bit more carefully on how they are dealing with health care. And I get involved in some of the discussions in these two. A lot of my health care research is in China and India. And what I find is that they are getting very advanced in AI, by the way. Where they are lacking is having right policies. That's not there.

Even cybersecurity policies, right? They are very relaxed. So what needs to happen is that I think US will take a lead on the policy side in this one for sure. And they are doing a lot of right things. They need to continue doing it. I don't think, you know, one thing people get worried always in research, what happens if different government comes? What happens this? I think this part, the AI

use in healthcare should be separate from anything that happens in the country and we have to continue on that path. And US can set a role model for the whole world if they are able to do that. And I think they are in the right track. They are on the right track. And I also heard the news a few days ago about EU passing some laws regarding the use of generative AI as well, right? So there's also efforts. Yeah, so EU has been proactive in that and in fact,

last year, late last year, US Vice President and UK Prime Minister had a meeting. And in fact, they met in the place where Alan Turing developed his core, the whole thing, the imitation game and the machine to break the enigma and all. So it was truly very psychological from psychological perspective, it was very interesting. But yes, you can use it, you care,

working closely to have a common set of rules. The only problem is that we don't even know how to regulate AI in general. Yes. Yes. And I don't think we will ever know. We have to experiment. We have to fail. And we have to improve. And some other AI, we can be more relaxed. But AI in health care, we have to be more conservative in taking this. That's how it will happen. But we will see a lot of

action going on in the next six months to two years when we will get a little more stability. Professor Subodha Kumar, thank you for your time and insights today. Yeah, thank you. Thank you for having me. Okay.

Amar, so far on the show, we've talked a lot about AI and drug discovery and development. This was interesting because it was really, I think, our first opportunity to talk about AI's use and potential with health care providers. What did you think? Absolutely. It was definitely a very different episode than what we have usually, which is a focus about the pharmaceuticals, right? Like the pharmaceutical and biotechs and what's happening in that area.

This was a bit of a different topic. So when you think about overall life sciences and health care sectors, so there are three different areas of that. So one is the drug development, which is the pharmaceuticals and biotechs. There is the other is more about the medical devices and diagnostics. And then the third area is health care in which this is where there's clinics and then also the hospital systems, right? And so today I wanted to talk to professor Kumar

about the healthcare settings so we got a very different view today. And it was fascinating about what are the different ways in which AI is being used in the healthcare settings today. And there's been tremendous progress there. Well, you talked about holistic changes in healthcare and the potential for AI to enable mass customization in healthcare. He used

dosing of patients with breast cancer as one example of this. Is AI the missing piece for realizing precision medicine and personalized medicine? Will the data be in electronic health records that's gonna be needed to really enable this? I think it's definitely gonna push personalized medicine or precision medicine much further. Again, we have talked more before about the personalized medicine or precision medicine

from the drug development point of view and developing drugs that are more suitable based on the genetic makeup of the patient. But now here we're talking about taking all that data and then in the clinical practice setting, making decisions based on that, right? And then, we know that the doctors, they practice precision medicine and they have been practicing precision medicine for decades where...

even, you know, whatever whatever label is, the drug label is, they will prescribe to the patients what they think is the right thing for the patients, right? So I've had this argument with a lot of physicians who say, well, I already practice precision medicine. It's already personalized to the patient that I'm providing. And that is true. Now this is going to be enabling them to do this further, right? Because now, see, any doctor, they have limited amount of memory

and then they have limited amount of processing power. So the way Dr. Kumar described this, you're going to have a co -pilot who has a lot of this data and that co -pilot is also going to have training in your therapeutic area. And they're going to be able to come up with some recommendations about, well, how do we really customize and personalize the medicine for this particular patient? And definitely going to be able to see this, yeah, he's going to be able to think through a lot of different angles

the physician may not be able to do at the spur of the moment. So that is definitely going to add a lot of more personalization to the treatment. Now, is that it? I don't know that. Again, as we've discussed many times, it depends on the amount of data there is, the amount of access that the AI is going to have to the data, because we discussed a lot about the HIPAA and the security and all of that. So we have to really think about how much, what data can...

the AI really process, right? But then also, see, the AI is also dependent on the algorithms that it has and in terms of like, so the, its capacity, right, is going to be based on some of the algorithms. So there are some limitation about what it can prescribe or what it cannot and making sure without doing these hallucinations, like where it's like coming up with stuff that never exists, right? So there's going to be limitations, but I think this is definitely going to be a really very, very helpful

tool for the HCPs to personalize the medications much, much more. And I would say not only the physicians, but also the nurses across the board, it's going to be very helpful for all of them. He was a bit low key at times, but he actually envisions a radical reworking of healthcare through AI and that it will be a major beneficiary of this technology. He sees this happening relatively fast, is he right? Yeah, absolutely.

And that's why it is very hard to really predict what's going to happen in 10 years down the road, right? Because, I mean, two years ago, hardly anyone knew what was generative AI. And now, in the last one and a half years, all the examples that he was talking about included generative AI in healthcare. I mean, we're talking about this progress in a year and a half. And as he gave some examples, right, like,

the amount of pickup that's, you know, the the of generative AI has been tremendous. And he also described how it has been really changing a lot of things in healthcare as well, right? So it has been pretty rapid and I'm seeing that everywhere, right? And as I'm talking to a lot of people, I'm seeing that. And then what I was surprised about is that healthcare, we talked a bit, sometimes,

with some of the technologies, it can be slow to adopt, but that doesn't look like the case because there are areas where it needs to be there and it's rapid. I just think it's just going to continue to be rapid. So I completely agree with them that it's a rapid change that we have already seen, but it's going to be rapid continuing. See, so far, this is May of 2024. So far, a lot of the institutions are doing some

proof of concept studies or so, we are going to be moving more toward industrializing AI and generative AI. And so that's going to be change at scale. And I believe that we will be seeing that quite a bit over the next two to three years. One of the things you asked him about was how much of a threat AI may pose to physicians and other healthcare workers. He sees it freeing them up more than it replacing them. And he talked instead about the need to retool training and

the skills that healthcare workers will need. That sound about right? Yeah, yeah. And that makes sense, right? So yeah, there are a lot of countries right now, there's shortage of healthcare professionals. So here, instead of doing some basic things like you talked about, you know, why do you need a human being to take your temperature? Their AI can do that. And he can then put that information in the system. And then if there are any signals, AI can flag those, right? mean,

These are some of the basic things that a human being doesn't necessarily need to do if an AI can do a decent job. And there are limits in which AI can really go wrong. I mean, what's the worst that can happen in measuring temperature? Well, it was wrong temperature. All right, so if it's the wrong temperature, you measure it once again and then see. So these are some of the, I would say, easier things for AI to do. These are also low risk things that AI can do.

And that's something that the AI can focus on. But then you have, let's say, an oncologist. Let the oncologist focus on, like, instead of trying to find data here and there, what happened, and blah, blah, blah-- having all the information, all the recommendations, and all the rationale behind the recommendations at their fingertips. And then they're making the highest value decisions about how to treat the patient, right? Let them focus on that. So what that's going to happen, what that's going to cause is,

the humans are going to be making the high value decisions. The doctors will be able to treat more patients actually, which is fantastic, right? And so we will be able to actually expand the healthcare or so. There's always going to be like some people saying, okay, well, can AI replace it? And he gave a fantastic example. Yes, the doctors make errors too, but then the errors that the AI makes

are not the same errors that doctors make. And so that's why you need to have both, right? So that you can actually, these are two very different type of decision makers, or if we put them together, then it's just gonna improve the outcome for patients. So yeah, I mean, it's a fascinating world and we're always seeing that, you know, whenever there's this revolution happens with the industrial revolution and stuff.

Some of the old things that people were doing, they are gone, but then it's more about the new thing that has come up. How can we use these new tools much better for our advancement is going to be the key. And as optimistic as he was, he did talk about the need about drawing a line and how these systems are used. And the fact that we don't know how to regulate AI in healthcare and that it will be an experiment and will also fail along the way. Do you agree with that?

Yes, we still don't know what's coming. It is very hard to predict. But as he said, we don't even know how to regulate this, right? These are all unknowns. Things in generative AI are changing on a weekly basis. It is very hard to keep up with all the new stuff that is happening. So what he says is, let's be conservative, right?

It's better to be safe than sorry. So let's not make too hasty decisions because these are the things we don't necessarily understand 100 % how they work, especially the new things that are coming up. We don't necessarily know all the implications of using those and what are some of the long -term effects, right? So we need to really study those. We have to be careful and be extra cautious. So I completely agree with him on that aspect.

Yeah, it's fantastic, but at the same time, it is something we need to keep an eye on because if you use it wrongly, there could be really bad consequences. I mean, you can think about, let's say, you know, we talked about this co -pilot for physicians. Think about a health co -pilot for a patient. The patient is entering their own data and then this co -pilot is then giving them suggestions about, do this, take this medication. And then the patient then goes and does that.

That is not a good way of this, right? Because neither of these is a doctor. They should not be making decisions based off of this, right? I mean, these are some of the downfalls that we see and we need to regulate this kind of stuff. We cannot have patients thinking that the acupilot is as good as a doctor and making decisions about that. And that is a scenario that we definitely want to avoid. So yeah, we need to regulate this. The question is, how do we regulate this?

Well, it was another great discussion and I look forward to our next one. Sounds good. Thank you, Danny.

Thanks again to our sponsor, Agilisium Labs. Life Sciences DNA is a bi -monthly podcast produced by the Levine Media Group with production support from Fullview Media. Be sure to follow us on your preferred podcast platform. Music for this podcast is provided courtesy of the Jonah Levine Collective. We'd love to hear from you. Pop us a note at danny at levinemediagroup .com.

For Life Sciences DNA and Dr. Amar Drawid, I'm Daniel Levine. Thanks for joining us.