Generative AI has been all over the news in recent months and is already making waves in the world of healthcare, especially when it comes to data analysis and clinical diagnosis. But how helpful AI actually is and how quickly it’s adopted depends greatly on its transparency, usability, and the specificity with which it’s deployed. Without clear explanation of its process, clinicians will (understandably) be hesitant to lean on any AI, let alone act on its diagnoses. For AI to make the full positive impact on healthcare that it’s capable of, engineers will need to work alongside medical professionals, allowing either party to remain the experts of their respective fields and in so doing create solutions that work with and for providers and their patients.
In this episode of the In Network podcast feature Designing for Health, CEO of Jiva.ai Dr. Manish Patel sits down with Nordic Head of Thought Leadership Dr. Jerome Pagani and Chief Medical Officer Dr. Craig Joseph to talk about AI, the founding of Jiva.ai, and how AI can and will revolutionize the healthcare industry. They also discuss communicating complex ideas to the average user, making AI accessible and trustworthy to clinicians and other health professionals, and how AI can bridge the gaps in life sciences.
Listen here:
In Network's Designing for Health podcast feature is available on all major podcasting platforms, including Apple Podcasts, Amazon Music, Google, iHeart, Pandora, Spotify, Stitcher, and more. Search for 'In Network' and subscribe for updates on future episodes. Like what you hear? Make sure to leave a 5-star rating and write a review to help others find the podcast.
Show Notes:
[00:00] Intros
[01:34] Dr. Patel’s background
[04:19] Starting Jiva.ai
[06:23] Representing technical things to “normal” people
[10:30] Working with designers as a go between
[12:50] Reconciling the goals of the manager, the developer, the designer, and the user
[17:45] Canva vs. Photoshop
[20:44] Designing for different levels of the same system
[23:19] How AI models will be developed in the future
[26:07] Where healthcare is heading in respects to AI
[30:36] How we will interact with AI in the future
[33:00] Using AI to put tools in people’s hands
[39:04] Over-alerting and alert fatigue in the NICU
[43:56] Things so well designed, they bring Dr. Patel joy
Transcript:
Dr. Craig Joseph: Dr. Manni Patel, welcome to the podcast.
Dr. Manish Patel: Many thanks.
Dr. Craig Joseph: How are you doing this fine evening?
Dr. Manish Patel: This fine afternoon in London. It's not too bad. Not too sunny. But it’s okay.
Dr. Craig Joseph: I was under the impression you were 27 hours behind us here in the United States. Is that…
Dr. Manish Patel: Completely possible.
Dr. Craig Joseph: Or ahead. I'm not sure it actually matters when we get to that number. Well, let me begin by asking you a question. And since we've now established that you do not live in the United States, I am going to try to translate this question into English for you.
Dr. Manish Patel: Yeah. Go on.
Dr. Craig Joseph: So I understand that ever since you were in reception, that's the pregnant pause, which I think is what we call kindergarten.
Dr. Manish Patel: Yes. Very good.
Dr. Craig Joseph: Ever since you were in reception, you've wanted to be involved with a startups and that you were talking to your teachers back then about artificial intelligence. Is that accurate?
Dr. Manish Patel: 100% true.
Dr. Craig Joseph: Okay.
Dr. Manish Patel: Back in 1981. Yeah.
Dr. Craig Joseph: So in 1981, you were telling your teacher that you were going to co-found an AI based company. And did you say that you were going to get a PhD as well? Was that also predicted?
Dr. Manish Patel: No. You know the truth is, there's only one reason I did a PhD. When I finished my master's, my mom said I had to get married. I'm an Indian, right? Indian origin. My mom said I had to get married, and the only thing I could think of that would prevent her from pursuing is to say, I can't. I'm still studying. And so I did a PhD.
Dr. Craig Joseph: That, Dr. Pagani, is that the normal route?
Dr. Jerome Pagani: I have heard the matrimony or matriculation before. Yeah. That checks out.
Dr. Craig Joseph: Excellent. Well, why don't you tell us the whole complete story of how you became the co-founder of Jiva?
Dr. Manish Patel: Sure. So I was never that much of an academic. Actually, I wasn't talking about AI in 1981, I was actually crying my eyes out because I always wanted to come home when I was in kindergarten or reception, as we say here. So I started off, well, as I say, as a bit of a non-academic, kind of went into university as a molecular geneticist around the time when the Human Genome Project was finishing. So maths and computing was coming into biology at a really big way. And that's how I introduced to machine learning back in 2000, 2001. Did my master's, and as I say, my PhD in that particular subject, and built the fundamental basis of what we have at Jiva at the moment. Very quickly found I can't afford a house in London as a postdoc, so I gave up being an academic and sold my soul to the devil and worked for banks and hedge funds and the algorithmic trading space, you know, that, as everyone does. I was actually at Lehman Brothers. I was at Lehman Brothers when it was peak share price, the day. And then it started falling and then I was there when it collapsed, which was fun. Ended up doing a couple of hedge funds banks and then, yeah, got back into healthcare, advising tech startups, health tech startups in particular, and then started Jiva in 2019.
Dr. Craig Joseph: Well, that's great. I think, as you say, everyone has gone and worked for Lehman Brothers.
Dr. Manish Patel: Absolutely everyone.
Dr. Craig Joseph: I know certainly I have and, and Jerome has. So that's basically how you kind of came to Jiva. How did you, what did you see as a need to start Jiva.ai?
Dr. Manish Patel: So we started Jiva in 2019 along with two other co-founders, Chetan and Sarah. Now Chetan is a doctor. He's a clinician, owns a couple of private practices in London, very successful. He at one point asked me to do some analysis of, of data, and I, I gave him a two double barreled kind of word that I'm not allowed to say on his podcast because I didn't have the time. But therein actually lies the problem. There are absolutely tons of clinicians, healthcare professionals, people related within healthcare and life sciences, who are experts in their field but don't have a clue how to do AI and machine learning or data science. They don't even know how to string a piece of code together. Why would they? That's not their expertise, they don’t need too. So what we wanted to do was create a system where people like yourselves who might not be coders yourselves, but you'll be able to describe your problem in natural language on our platform and describe what you want to do. And the system takes you through step by step in creating that solution with you. So you remain the expert in whatever you do. Say you're a pediatrician and you're trying to diagnose the onset of a collapsed lung in a premature baby or something, and you don't know how to interpret the signals from all the machines and write some code that might be able to predict such a thing. But you could actually tell our system, well, this is what I want to do. These are the types of data that I have. How would I go about analyzing that data? And we can take you through step by step, note by note, how to actually do that. So in that equation, you remain the expert in your field. We remain the expert in AI and machine learning, and together we create something amazing.
Dr. Craig Joseph: That's great. I'm glad you've done it. When we were preparing for this episode, you had mentioned that one of the more difficult parts of this endeavor for you, pretty much with any software company actually, is that the software company is filled with technical people who don't understand how normal people think. And you have to represent some very technical things sometimes to quote unquote, a normal person. And to me, that sounds like design. And that's one of the things that we like to focus on here is design and its overlap with healthcare. So how do you do that? How did you get a bunch of PhD type people like yourselves to be able to translate the needs of the common person?
Dr. Manish Patel: So there's millions of clever people out there, but actually relatively few have the tools. So the mental tools to be able to articulate something that's incredibly complicated in the expert field that they’re in in words that a seven year old can understand. And I know you have about the mental age of a seven year old, so I'll try to dumb it down for you.
Dr. Craig Joseph: Thank you.
Dr. Manish Patel: As I speak about AI. So this is a really difficult skill to learn. It is really hard, but actually anyone can do it so long as you understand what you're actually talking about. That's the main thing, is you firstly understand the complexity of your own world, whatever it might be, say it's quantum physics or AI or whatever it might be, and you're able to articulate that and string it together in a way that someone in everyday circumstances can relate to. So when I explained AI to my son, he was ten years old at the time, I explained it in a way that he understood from school. He goes to school, he learns, his brain is able to pick up signals from his teacher and that's how he trains his brain to remember things. And actually AI is not that different from that. It's just more artificial and it runs inside a computer. The brain is the CPU and the memory combined. So putting it in those particular, in that particular way, is a skill that I think is learned, I don't think you're born with or anything, it's just a matter of understanding how to do it. Now finding those people who already have it is actually hard because there's not that many of them, but easy in that when you do spot it in a in a potential employee, you get it straight away. So some of my employees will attest to the fact that we, in our interview process, will go through a set of questions like, you know, how would you write a piece of code that does X? And some of them really struggled and some of them just say, Well, we just do this, this, this and this and this and explain it really, really simply. And the guys that can do it simply are the ones that actually understand what they're talking about, the ones that waffle and talk a bit too much, maybe they’re nervous or maybe I’m being a bit unkind. But you know, the ones that talk a little bit too much are probably trying to pad a little bit of blank space in the in the interview to get to a certain point. So simplicity, I think, is actually the key to making things simple. Like I say, it's a tough art, tough art to master on that. On that note, remember how simple the first version of Android, what, I don't know if you're an Android user or an Apple user, but one of the first versions of Android was absolutely rubbish. It looked like a really rubbish version of genome, which people who use Linux will remember. Horrible, horrible interface. I mean, why would you ever use it? It was simple, but it was highly unintuitive. So actually simplicity, while that is the key, it has to be kind of tuned into intuition. How does a human actually interact with something in an intuitive way? And that's pretty fundamental actually to what we do at Jiva in making AI easy for everyone to access.
Dr. Craig Joseph: Well, so some groups might give you a counterpoint and say, well, you know, it's okay to have the developer be super techie and not think like a human. As long as I have a designer who kind of can sit between them and the and the code that they produce, you know, might achieve the same effect. Have you, have you gone along those lines or do you just say, no, no, no, we're going to look for the person who gets it? So I don't need to have that intermediary?
Dr. Manish Patel: So I've been on both sides of the fence here. So I was a deep technical person and, as the CTO of another company I was at, built what I thought was something that was awesome. It was a messaging system for doctors and patients to share data securely and for that data to make its way up to the EMR system, you know, all those kind of things. And I thought it was great. And then we showed it to a couple of nurses and showed it to a couple of patients and they were just scratching their heads thinking how do we, we don't get it, how do we, how do we get to this point or that point. And at that point we actually got a UX developer involved. He kind of did his bit of research and he said, Well, why don’t we just make it simple? Like this, this and this? So I've seen that work. I've seen having a designer or UX person in between to translate between the people who are using the application to the people who are watching, developing the application and, and make that thing easier to use. It doesn't always work, I got to say. So that has, I’ve been in a couple of situations where that's just been an absolute disaster. Maybe we just had the wrong designers, I suppose, but I think in general, if you have someone that understands the user and knows what makes the user happy, right. What is happy equate to? Ease of use. Get to do the things that they want to do quickly without, you know, clicking 3 million times on 20 different buttons. Then if you can get to that point then great, then you should definitely put a designer in place to translate to developers. Developers, their brains are wired completely differently. They're very logical people that, you know, talk to my wife. She's an actuary. She likes complexity on the things that she does on her spreadsheets. She would never be able to develop something that's easy for anyone to use. She can't even explain her job.
Dr. Jerome Pagani: Manni, is part of the problem that the developer and the end user are conceptualizing the problem differently, in your in your experience, or is it that they are focused on a different end state? And what is the role of design that really, where do you get sort of the most bang for your buck?
Dr. Manish Patel: Yeah, so there's probably two or three different things happening that is actually outside of the developer and the designer and the user. And actually as always, ther’s probably something to do with management. So management have some idea of what they want to do. They tell developers what they want. Developers interpret that in their own way. There might be some kind of exchange about, you know, what is our acceptance criteria for this particular project, acceptance criteria tend to be very stepwise, very logical, don't tend to imbibe a sense of intuition in the application creation process. And then, what, and then we’re surprised we find things that are very kind of, you know, computerized. I can't really think of the word, you know, do you know what I mean? Kind of jagged edge rather than soft edged type of user interface. And so I think the fundamental thing is, in that communication between manager and developer, the designer has to be there right in the middle. So don't exclude the designer from that entire process. And I understand, like for us, I say from experience, we didn't do that to begin with because we couldn't afford it. We were a startup, we were bootstrapped, right? So we didn't do it. But eventually, put that designer in place in between so that whatever the manager thinks or the managers think should be the solution, that is actually passing through a filter where the designer, or the UX person, has actually gone out to market. They've actually gone to ask the users, interviewed them about what they think is important and imbibe that into the application. That's all well and good in theory. I'll go a bit further with a little bit, a little bit sort of off-piste, but in startup world, the companies that have done really, really well, the unicorns, have often thought of things that the user didn't even know they wanted. That becomes really difficult then, because actually you don't have someone to go out to market. Steve Jobs never went out to markets and Oh, what would you like in a, in your all in one iPod phone web browser? Obviously, you know, they might have done things in sporadically, but they had, they were visionaries. They created a visionary product, but they understood, critically, they understood simplicity and they understood making things intuitive. And that's why that thing was so successful. No one actually really needed it. If you think about it, they only needed it after we knew we needed it.
Dr. Craig Joseph: Yeah, I think that's quite interesting. So just to summarize what I heard, management is the problem.
Dr. Manish Patel: I can say that, I’m the CEO.
Dr. Craig Joseph: And if you get rid of management, everything can flow quite nicely. I like that. And that's, one of the key things is, you know, how can you, how can you, how can we all become Steve Jobs? We can't. Right? But oftentimes, you know, your users might tell you that they want something, but it's different than what they need and they don't take into account how others might be needing something as well. Right? So, you know, if you ask a nurse in the intensive care unit what they need, it might be, you might get very different answers than a nurse in the emergency department or a nurse that works in an outpatient clinic. So that's another aspect that I suspect that you have to take into consideration. You ask the question, but then you might have to ignore the answer.
Dr. Manish Patel: Yeah, exactly. Yeah. And there's no right way of filtering that complexity out. You're always going to get noise, right? You're never going to get a hundred people saying the same thing. Entirely different scenario here. I've gone through the startup journey. As part of that startup journey, I have to create a pitch deck, usually 12 to 14 pages of what the company is and does and where we want to go. And I can guarantee I can ask 100 different investors and advisors on feedback on the deck, and I will get a hundred different responses because those kind of things are highly subjective, right? So you've got to somehow rise above the noise and just say, okay, on average, what do people need right now? Actually, the true visionaries ask, what do those guys need tomorrow? Because those are the ones I got to do really, really well. And that's a that's a very difficult thing to translate into design because you can't actually ask anyone to validate it. You have to imagine it.
Dr. Craig Joseph: Right. And that's great and it's a great segway into another question comparing two pieces of software and I think you mentioned this before, Canva and Photoshop. Those are probably words that most of our listeners did not expect to hear us discussing on our Designing for Health podcast. But so those are two different pieces of software with kind of two different visions and two different user bases. And I'd love for you to kind of compare and contrast the two about how you use them and how you see them being, you know, fitting into our world.
Dr. Manish Patel: Absolutely. So, so the very first time I used an Adobe product was in 2001 or 2, I can’t remember what I was doing. I was doing some graphic-sy type of thing for the department, the university department that I was at. I was completely blown away by how complex that application was, even back then. I look at it now, I look I look at things like Illustrator and I'm sure there's, you know, obviously people out there that use it and with great effect because it's got so much functionality in there. But for a newcomer, entirely untrained, it's really difficult to approach that. Now, they are two products, as you say, that are aimed at two different types of people, the people that really know what they're doing in design and have probably been trained to use those kind of things. And then Canva for the more general populace. But when you think about it, I mean, I know designers who actually use Canva to do quick mock ups. Why? Because it's really fast. It's really easy to use, right? And the more easy you make something to use, the more popular it’s going to be. So if your objective is, you know, a number of users uptake, possibly revenue associated with that and popularity, then yeah, ease of use is the way to go. Of course, there's always going to be, you know, requirements to go the more complex route where you have to cater to a more complex audience. But that's going to be a relatively small population compared to the massive one over there. You know, Canva vs Illustrator or, you know, all the other products that Adobe have that I have no idea what the difference is between them all. That's kind of the same thing that we're trying to do in Jiva. Most of the applications out there are like the Adobe type applications. They're really hard to use, really difficult to put together an AI and data solution. For a newcomer, for someone who's not untrained, we're trying to make Jiva like a Canva. You come to it, you don't need to be trained. You just basically start experimenting, click and traggin, and you'll get to a point where you can actually create something that's really amazing. ChatGPT, done exactly the same thing. What did they do? They didn't, they didn't add loads of buttons. They didn't add loads of dropdowns. So, yeah, let's see. I mean, you know, there's lots of different ways you can actually change GPT, temperature and things like this. They didn't expose any of that. There was literally text box, put your language in there and your history on the left hand side. That’s it. How many users do they have now? 100 million?
Dr. Jerome Pagani: Manny, that is a great point and ties into something that we've written about in the book, which is that you have to design for different levels of the system. So some things end up being designed on a project level, but then they don't scale within the organization or they end up being designed for the organization as a whole, but they don't help that organization fit into the larger health ecosystem. So similar to the Canva/Adobe problem, and in consulting we talk about this as not trying to boil the ocean, right? So how do you pick the level of analysis or level of design that you want to aim for to make it maximally useful, not just for sort of the individual stakeholders that you're able to get around the table at that particular point in time. But as you said, as a visionary, you’re really designing for a larger group and for a bigger purpose then and just what's sitting on the table in front of them?
Dr. Manish Patel: Yeah, that's that's a really great question. That's a really difficult one to answer generically. But let's start here. So to get to a point where you're addressing the audience that you're targeting, first, you actually need to understand what that target actually is. So very first step is who is my user? Who is my customer? Try to understand who they are. Forget about how how big that population is. Adobe didn't care that their applications would have only appealed to 0.000001% of the population. It was that part of the population that they really wanted to wow. And they have. They’ve completely monopolized that part of the market for those types of users, for graphic designers and things like this. So understand who you're actually trying to target and don't worry about the magnitude of that target. Ask them what they want to do. Ask them what they feel is necessary. Don't necessarily take that as gospel because sometimes they are not visionaries. They can tell you what they want right now, like you were saying, but they can't tell you what they want tomorrow because they can't see it yet until you show them. So that's, there is a mix of your own imagination, your own creativity in that whole conversation which comes up with something that, you know, no AI could have come up with, at least at the moment, no AI can come up with a design, an idea that formulates in your mind that no one else thought of before and you test it and you'll probably do that about a gazillion times before you get it right. But that's just the name of the game.
Dr. Jerome Pagani: So let's change gears a little bit and talk about AI and what it looks like today and sort of how it's going to evolve. So the way that AI models are built today are largely, there are clearly defined languages and sort of structures for the models that we build today, but we can easily see this beginning to shift into sort of a low code, no code kind of AI development. What does that look like and how will that sort of change the way AI is built and deployed in the future?
Dr. Manish Patel: Sure, so rewind 20 years. Do you remember around the dot.com peak where all these big companies were hiring HTML developers?
Dr. Jerome Pagani: Yeah. Yeah.
Dr. Manish Patel: Do you want to search for HTML developer in LinkedIn and see how many results come up?
Dr. Jerome Pagani: Got to be almost nothing.
Dr. Manish Patel: Right? Disappeared. Why? Because people like, companies like WordPress and Wix and different types of companies, they highly commoditized access to HTML development through a no code interface, right, they made it easier, which actually meant the people that were doing HTML coding before and maybe even made a career out of it had to switch to CSS or design or something about something else, something more related that wasn't covered by that kind of application. So I see the same thing happening with AI, we’re flush with data scientist and AI engineers. Or, well, a lot of people say that data scientists are not actually data scientists but you know what I mean. There's lots of those type of people and I think they're going to find in the next few years, actually, only the guys that are as say, guys in a in a non-gender sense, by the way, only the guys that are really deep, deep coders, the guys that do the nitty gritty of a machine learning on a very low level. I think only those guys will survive and the rest of them that, the fat will just be skimmed off and they'll move on to something else because there'll be applications like Jiva who can highly commoditize the access to the isolations without you having to write any code at all. It’s what happens with almost every technology. At some point there's going to be lots of people who are absolute geniuses at doing these things, but then it very quickly it disappears because someone else comes along and says well actually, you don't need to do that, you can just do it like this. It made it easier. I mean, human beings, I mean, we're actually pretty lazy people when you think about it. We always take the easiest route, right? We never learn anything, at school we're told, you know, choose the harder path, you know, you'll learn more. But no, that's not, that hardly ever happens. Everyone takes the easier road because it's easier and that always wins.
Dr. Craig Joseph: So Manni, you deal with healthcare mostly from, or exclusively. Let me ask that actually. Which, which is it? Answer the question.
Dr. Manish Patel: Most, mostly. Mostly healthcare.
Dr. Craig Joseph: All right. And so being at the at the forefront of AI in healthcare, any predictions from you about where you see healthcare going with reference to AI? We've established that all developers are going to be out of jobs in a couple of years. I think you said, you just said that. What about physicians? When do I have to have my retirement set? If I'm a radiologist or a pathologist?
Dr. Manish Patel: Oh, I don't think you need to worry about that. So, so I think healthcare is going to be, well, you don't have to worry about that due to AI. There might be other things, I suppose, but at least with respect to AI, I think in healthcare we'll have the highest amount of excitement and hype, but probably one of the lowest uptakes in industry. And the reasons for that is actually, they're actually very sound. Firstly, I don't know of any AI company in healthcare or any company, health tech company that is using AI in healthcare, that is touting the replacement of clinicians in any shape or form. If anything, it's an empowerment mechanism. So almost everyone, I mean, I’m gonna say almost everyone, only because I can't think of anyone, just one or two. Everyone is talking about having a human in the loop, whatever it might be. So, at Jiva we've created a prostate cancer diagnostic, the idea being that we want to reduce the number of biopsies like get done after MRI scanning, and that is all about making the radiologist more accurate in predicting whether or not there is a clinically relevant tumor in the scan. Now, at no point are we saying replace the radiologist with the AI. What we're saying is use this as a tool in much the same way you use a DICOM viewer or an EMR system rather than paper to make yourself more efficient, get through more them, do them more accurately for the benefit of the patient. That's essentially where all of this is going. No one is, absolutely no one, as far as I know, is talking about replacing clinicians in any shape or form, making healthcare systems more efficient is actually I mean, I guess it means something different in the US as it does in the UK, given that we have a national health service. But at least for the National Health Service, which is basically heating over at the moment for a number of reasons, there could not be a stronger argument to augment the current healthcare system with automation and AI supervised with humans, you know, so that we don't, so that we can mitigate the risk as much as possible so that they can be more efficient, so they can get through more patients, so they can get through the backlogs and get it back on its feet again. That's really, really important. We're not talking about a similar, there's a similar argument going on in the UK that, there’s recently been rail strikes as well as nursing strikes and with a rail strikes it’s a bit different. They’re talking about having automation in trains to drive them without effectively a driver, but actually no one is saying you shouldn't have a fully qualified driver on standby on the train just in case they need to do something. They’re just saying actually let’s just make that more efficient. Because drivers get distracted sometimes, that kind of thing. Right? That's okay. I think that's okay. So long as they’re not saying replace an entire workforce. And just to clarify, I wasn't, before when I was saying or talking about data scientists and machine learning engineers, I wasn't talking about like getting rid of an entire workforce. I was just saying they are just going to get displaced and do something else, just like everyone else did. Even in the Industrial Revolution, people moved on to something else. The guys, the guys that were working on doing manual work, they ended up learning how to use machines. So humans move on. They move on to do other things. So we're not talking about replacing anyone so long as we have that mindset. I think in the next ten years, to answer your question directly, so long as we have that mindset, we'll see AI being adopted in good volume. I'm not going to say across the spectrum, but in good volume, and I'm afraid to say some of that is because some of the older generation is going out and some of the skepticism going out and the new generation is coming in. But a lot of that is also because the technology is cutting back. Good regulations are kind of catching up in some countries and people understand that, okay, if you're going to use an AI as a diagnostic, as an example, that's got to be properly clinically validated. If you're going to use an AI to predict bed occupancy, okay, that's not very highly regulated. There shouldn't be many barriers to entry to that particular type of scenario.
Dr. Jerome Pagani: So Manni, you said, and I think very rightly, that it's not just the AI that is important, it's the way that, like all technologies, people interact with it. Today that's very manual keyboard and mouse, but that's going to change. So what is that going to look like in the near to intermediate term future? And how is that going to influence uptake and usage within healthcare?
Dr. Manish Patel: Yeah, so, so taking a step back away from healthcare just for a second, I think the interactivity with machines are human to machine interfaces is just going to change forever the next ten years. We're so used to using the keyboard and mouse, we only just started using voice five or six years ago in any big way, right? I think those things are just gonna, I mean, I saw something, some really amazing stuff just the other day on LinkedIn where someone had to quickly sort of very generically put together some code that could read some ECG, not ECG, sorry, brain information, what are they called, waves of some kind?
Dr. Craig Joseph: EEG.
Dr. Manish Patel: Yeah, yeah. And, and he was able to get words onto the screen by thinking about certain things. I mean, cool is that? Elon Musk is working on something similar. No idea how far they've got. Neuralink, as I think it's called. Those things, those interfaces are going to change. That's the first thing. Within healthcare, the one interface that is never going to change is human doctor or nurse or healthcare professional with a human patient. That's always going to be that no matter how much technology you have, maybe up until the point we have Star Wars like type of, you know, doctors delivering babies who are actually droids, that type of stuff. We’re, we’re obviously many years away from that, but that part is never going to change. So AI in whatever shape or form it takes in whatever sub profession within healthcare it finds itself in is always going to be an augmentation tool. It's going to be something that's either sits in the background and tells the healthcare professionals as to, you know, upcoming risks, maybe or diagnoses or is going to be something that is going to hit screening. That screening is a big, big thing here in the UK as well. Can we actually select patients that are susceptible to certain diseases before actually it happens and therefore not cost the healthcare system further down the line? That's that's very important in countries where there is a system that is paid for by the state. Right? So, so those two areas I think are the main drivers, main uptake.
Dr. Jerome Pagani: So Manni, when I was an undergrad, there were two courses that were, were super popular. One of them was called rocks for jocks and the other was physics for poets. The whole idea was to take those really complex ideas that are very important, but boil them down to something that a very non-technical audience could understand and make use of. And it seems like Jiva AI does that, but actually goes beyond that. So you're not just talking about putting concepts in people's hands, you're creating tools that help everyday people in healthcare do things in a new way and extend their functionality. Can you tell us a bit about that?
Dr. Manish Patel: Yeah, sure. So I said a little bit beforehand that it's important that we put an application in Jiva essentially in front of users where we have a clear separation of what the expertise is of the user and what the expertise is of Jiva. So you guys are doctors, you guys have a specialty, you know what you're doing in your clinical environments. Jiva does not know this and it doesn't have to. You guys remain the experts in that particular field, but you don't know how to do data science and AI and that's where Jiva comes in. And so that interface is really important that we make the complexity of data science really easy for you to access to the point where you can just interact with it with natural language, you can describe what you want and the system starts creating the solution with you. But it's really important as you're going through those steps that Jiva explains back to you why certain things are being done, right? So as an example, you could create a model. You know, in general, artificial neural networks are very, we call them black box models, very difficult for you to understand what's actually going on under the hood. Now, we might understand kind of what's going on under the hood and how we built it, but you certainly don't. What we would try to do in that particular circumstance is try to make it a little bit more gray box, a little bit more white box. So we try to tell the clinician, okay, we predicted, this model predicted, that this was the cancer, and these are the reasons why. These are the inputs that you gave us. And these are the things we think are really high indicators of whatever it is the type of cancer you're trying to predict, because we've seen this in history. And wouldn't it be cool if our model could also say, well, out of the thousands of training datasets that you gave me to train on, I can pick these ten and look how similar they are to your test set, to you test patient. And because they're so similar and these were all cancer, we think this guy's cancer. And so having that explainability, that white boxing, is really important, that's kind of in the in the enriched model sets. But also when you actually interact with Jiva trying to build a model, we want this to be kind of going to be like Excel, right? So in Excel, you don't know how the, well, you might know, but you don't know how a T test is coded under the hood in VB, or whatever it is in Excel, you don't know how the spreadsheet looks when you're doing a V lookup or an H look up and all these kind of amazing functions. It just makes, Excel, just makes it magically easy. Stick the formula in and it just works most of the time. It’s a funny story about that actually about Excel, which I'll tell you about later, but usually it just works. So we want to make Jiva like this, that the technical magic that actually really doesn't matter is hidden away. We expose the simplicity, but where it matters, where it's actually really important to tell the clinician why certain things are happening and how they're happening, that's the things that we expose. So not the, you know, how do how do I explain this as an analogy? If you have a glass of water, you don't need to understand the quantum mechanics of the protons and neutrons and oxygen and hydrogen atoms, the perfect positioning of everything, even if you could, you just need to know that it’s a glass of water.
Dr. Craig Joseph: I love that. And if you could explain to me why the water molecules bent, that would be great.
Dr. Manish Patel: That’s a whole journal.
Dr. Craig Joseph: Because apparently, apparently that's important, Manni, that H2O is bent.
Dr. Manish Patel: Yeah, I actually know what you’re talking about but the way, I don't know. But I do know there is a whole journal dedicated to the chemical properties of water that I found at Kings College when I was there which I find very interesting.
Dr. Craig Joseph: But we'll have to find a link to that because I know our listeners really want to understand molecular biology.
Dr. Manish Patel: It’s pretty fundamental, I would say. So yeah, yeah.
Dr. Craig Joseph: What is the story of Excel that, you said there was a humorous story of Excel and I believe I have heard zero humorous stories of Excel.
Dr. Manish Patel: I said, I see, well, humorous, but maybe it's not so humorous. So when I was at, when I was doing my PhD at UCL, I was in the oncology department for part of my time there, and one of the medical physicists was building his dosimetry models, so dosimetry being, you know, how you figure out how much radiology to give to a patient for a particular cancer in Excel. And what he found actually was, because it was quite a complex set of equations, what he found was that Excel was introducing a tiny decimal error for every calculation, and it was adding up as it was going along. And actually that caused it to be quite dramatically wrong towards the end, he recorded everything in R as we did back in the day as so it was completely different. So, so those things do kind of happen when you hide away complexity, I guess, that’s more to do with testing your technical capability more than anything else. I say that’s what, it's not actually that funny. I'm kind of hoping no one died from that.
Dr. Craig Joseph: Well, I was going to laugh until you said that last part.
Dr. Manish Patel: I'm sure no one did. I’m sure no one did. They were all terminal, don't worry about.
Dr. Craig Joseph: Wow, Manni, you're really taking this to a dark place. All right. Well, let's talk about something a little bit more, with a happier ending. You had a daughter who was born several months premature and found yourself along with her in the neonatal intensive care unit at the hospital. And one of the things that you've mentioned in the past is that you heard a lot of beeping noises. And then sometimes the, you know, what nurses did in response to those beeping noises was a little upsetting. So, yeah, tell us that story and what were the takeaways for you?
Dr. Manish Patel: The whole birth of my daughter is a very long story, I won’t take you through that one. But it was a bit of a traumatic experience, obviously, for both me and my wife being, you know, her being born three months early, being this tiny little thing kind of looked like this, you know how birds, baby birds fall out of a nest kind of like this with their hands out. That's what kind of what she looked like. But what was actually really distressing, at least for new parents, is you see all these kids and I can't remember who was NICU or PICU. I can't remember the difference. But anyway, in this department you have these six or seven kids in each ward in the incubator, and there’s this bleeping sounds really loud. Everything's like bleeping. And you got these all these machines connected up to the connected up to the incubator and connected up to my daughter and the, you know, the chest and the oximeter and, you know, all that kind of stuff. Everything's beeping. Even when her oxygen levels were, they kept on dipping below 80, 80%. And we're told that was really bad. We had no idea what was good or bad. I mean, it was still orange on the screen. So I though orange is not as bad as red. So it's not too bad. But actually, 80 is bad. So but all they do is just press the button. Silence it. Just kept silencing, kept on silencing these alarms and what's the point of setting an alarm at a certain point? It goes below 70 or 75 or whatever it is. I can't remember what the threshold was and it just alarms for no intelligent reason. What if you were temporarily below the threshold and came back up again? Is that an alarm? Is that, is that a positive signal? And there's a lot of data science there. Well, there's two things. There's a lot of data science signal processing there, but there's also a lot of design flaw there. I think that what happens when there's actually something that's really important, something's alarming. And these nurses, I mean, they’re doing a great job, absolutely fantastic job, but they get tired, they’re humans, and at some point they're just going to on autopilot, they're going to switch something off without realizing it's actually important. And that's what really scared me is when, how can, how is it that in, what was it, 2014? How is it that we're at that point where we can't even think of something more intelligent than just beep when it goes through a threshold? That's a design flaw for me. Another application I think of AI machine learning in that area, although I say it as a parent, probably a risky one, you have to get that one really right. Alarm when it's actually only useful to know.
Dr. Craig Joseph: Yeah. To your point, really, when you talk about risk, it's more complicated because it's, is the risk “well it can never be wrong,” or is the risk “it can only be wrong less frequently or less importantly than when, you know, versus a human.”
Dr. Manish Patel: Yeah.
Dr. Craig Joseph: And I think the same is, the same argument’s made about self-driving cars. Can they have zero accidents or are they allowed to have accidents but far, far fewer than humans would have? Is that okay?
Dr. Manish Patel: Have you heard of the Daily Mail test?
Dr. Craig Joseph: I have not.
Dr. Manish Patel: Okay, I suppose it's a British thing because the Daily Mail is a British newspaper, but the Daily Mail, popularly known as a very accurate (ha) newspaper in the UK, a little bit, you know, right wing, a little bit anti anti-immigration, blah, blah, blah, blah, blah. And they often print stuff that gets people into trouble, especially politicians of all sides. And so we say, does this pass the Daily Mail test? And self driving cars is probably one of them, right? Does it pass the Daily Mail test? Is it, is it that you can have a car that is statistically much better than a human at avoiding accidents, but when that when it's that one accident that happens, that will blow up in the media or blow up and you could effectively kill off an innovation because of one incident, even though the truth is that it’s statistically better. And I think that's more of a societal thing for us that we have to we have to adhere to a little bit more to the truth and actual fact rather than anecdotal evidence. That's basically the Daily Mail test, does it pass it? Yeah.
Dr. Jerome Pagani: We've talked a lot about design today, Manni, and we like to end the podcast the same way with everyone and ask them about design that they encounter in their everyday lives. So what are two or three things, and they can be outside of healthcare, but two or three things that are so well designed that they bring you joy to interact with?
Dr. Manish Patel: Right. I'm going to be a little bit geeky and a little bit esoteric for the two that I can think of. So the first one is one that I'm pretty sure no one who's listening has ever interacted with but I used to interact with every day of my life. It was actually a programing language called Q. And it was, you know, this is not, you know, it's not designed in the traditional sense, you don't see, you know, something nice and pretty on the screen. And fact is, if you've never been trained in it, it looks absolutely awful. And you'll feel like blowing your brains out when you're trying to read it. But it is such a well-designed language designed to do exactly what it's supposed to be doing, fast computation, vector mathematics or vector logic, I should say, and in ways that are fully functional. So they, you know, you could say it was funny, it was a relatively small community, less safe because relatively niche language. And there was internal competition, I say internal but global, of which there was only probably only 100 people, but there was a competition of who could write the shortest Sudoku solver in the language. And someone got to 37 characters. And that's how amazing this language is, is that you can write an entire Sudoku solver, a program in 37 characters. That is awesome programmatic design. It’s not the type of design you guys are used to, so I'm cheating a little. But for a programmer, that was an absolute delight, language to work with that you can do such amazing things. That was really geeky, I'm sorry.
Dr. Craig Joseph: I love that. I love that.
Dr. Manish Patel: The second one, I only have two, but the second one is also a bit of cheating. Again, not human made design, I'm going to say design of ant colony behavior. Stick with me on this one. It does go somewhere. So think about you see ants kind of colonies all the time. They're all over your kitchen and you kill them all. And now it's like you don't really think about what's actually happening under the hood. Ants have maybe three or four or five, depending on the species, different types of ants inside that colony. Actually, each ant only really interacts with its environment. It doesn't know what's happening, you know miles away. It can only detect the chemical signals in its immediate vicinity of what's in the air, and therefore it only operates on a very small set of rules. So if you were to computerize this, you could say you're going to have a very small set of rules with a very limited sense of, very limited set of sense organs, if you like. So sense of smell essentially. And from that you can create something so organized as a colony where it builds ventilation systems, it builds defense mechanisms, it even puts a graveyard on one side, on diametrically opposite sides of where it actually puts its food, where it harvests it, it always puts the queen and its and its eggs somewhere in the center down below where it's nice and cool at a very precise temperature. All of these things come out because there are millions of agents that are interacting with a very limited set of rules. That is really interesting for me because if you think about it, you could create very complex systems in machines, in computers, in software by understanding how to create those rules. That design that nature has there, I think that that just blows me away, that nature was able to come up, I say able to come up, you know what I mean? Evolved to created design that was that beautiful.
Dr. Craig Joseph: You're making me tear up in talking about the Q programing language and ant colonies.
Dr. Manish Patel: I’m going to make you cry even more, look. Liverpool.
Dr. Craig Joseph: Oh, wow. Wow. We really wanted to get through this, through this episode without any English football references, but we were unable to do so. And for the record, I acknowledge that the team that you support, Liverpool Football Club, is slightly better than the team I support, Tottenham Hotspur, and that's on tape now and that will never go away.
Dr. Manish Patel: Thank you.
Dr. Craig Joseph: But it will change next, in the fall when we start playing again.
Dr. Manish Patel: Let's see.
Dr. Craig Joseph: Awesome. Well, thank you so much, Dr. Manni Patel, for joining us and talking about healthcare and AI and design. We really appreciated it.
Dr. Manish Patel: Thanks so much for having me guys, appreciate it.