Artificial Intelligence in Healthcare with Bulent Kiziltan, Chief AI Officer, Chief Data Scientist at Stealth Mode Startup (transcribed by Sonix)

Welcome to the Outcomes Rocket podcast where we inspire collaborative thinking, improved outcomes, and business success with today’s most successful and inspiring healthcare leaders and influencers. And now your host, Saul Marquez.

Saul Marquez: And welcome back to the podcast. Saul Marquez here. It’s a pleasure to have our guest today. His name is Bulent Kiziltan. He’s a chief A.I. Officer (artificial intelligence) former head of Deep Learning Chief Data Scientist, Astrophysicist… yep he’s all those things. He’s a leader in disruptive innovation in the industry for 10 plus years working with the pioneers in the field at the top institutions pushing cutting edge RND. He’s implementing company wide A.I. strategies to effectively leverage continuous growing data, building diverse cross-functional teams to identify business opportunities something that we’re all up to and we’ve got some great ideas free here during this discussion. He enjoys extracting low signal to noise information from heterogeneous data at any scale, delivering data driven insights to business stakeholders and creating immediate value because that’s what today’s businesses are about. And so it’s a pleasure to have Bulent on the podcast. Welcome my friend.

Bulent Kiziltan: Thank you Saul, it’s a pleasure.

Saul Marquez: Did I leave anything in that intro that you want to chat with?

Bulent Kiziltan: That was more than what it was supposed to be. You captured everything.

Saul Marquez: Love it, love it man. Well no it’s a a pleasure to have you on man and you know really grateful that you carve some time out. What is it that got you in the medical sector? I know you’ve been in A.I. for quite some time and your entry into the medical sector is probably a little over a year now, but what got you in?

Bulent Kiziltan: So I am an Astrophysicist by training, I come from a physics math and some engineering background and in that domain for a very long time I’ve operated at the end where we looked at complicated data sets, trying to extract information from low signal to noise data as you just mentioned and tried to understand the data with our domain expertise and you know collaborating with astrophysicists that have expertise in different sub domains and so essentially what we were doing we were extracting information from data. And this is one of the premises of A.I. sometime before machine learning became so popular, I worked with experts in the field of Bayesian statistics some liked math and through that interaction we’ve been able to publish some papers, make some discoveries. And after I joined Harvard and M.I.T. later on, I was fortunate enough to collaborate with some of the really pioneers in the field of A.I. and deep learning and what worked with them closely to learn what that domain has produced and try to implement some of those tools in my own domain of astrophysics and through that interaction also made some interesting discoveries which we published in journals but through that transition what I’ve noticed in academia at large is that because of its nature and how science operates it’s somewhat conservative when it comes to implementing cutting edge know-how and A.I. at the moment from a scientist’s perspective is considered not science but good engineering because the science is still not too well understood. So implementing it using that technology to do fundamental science has certain challenges and it takes time moving forward and implementing them. So I decided to transition to the industry where the dynamics of implementing those cutting edge tools and technologies is very different. There is more space to fail whereas in academia you know you have to be very conservative when you’re moving forward. You cannot just be a maverick about whatever you’re doing you have to be sure when you apply certain technologies. So in the industry the lifecycle of things is much quicker which is essential in A.I. currently because things are changing on a weekly basis. So you have to apply and you have to fail to a certain extent and then move forward. And this is how technology especially in the A.I. space is moving for. So and when I was trying to decide which domain I wanted to be I wanted to be in a space which would have some meaning that comes with the impact. I mean the business call the business schools of healthcare companies is not mutually exclusive with increasing the health of individual members. So if you can work in that direction and motivate yourself with the idea that you can help members get healthier you are partially achieving your business school. So that was very meaningful to me it’s something that I could relate to better than other domain. So I wanted to be in that space. That’s reason number one and another reason is the healthcare space is kind of broken especially in the US potentially being able to contribute to make that ecosystem better is also something that I could relate to. So I wanted to kind of bring in some of the machine learning A.I. strategies into that domain and see whether I could contribute to help members live healthier lives and also make the ecosystem work better, morph… be more functional. So it’s something that I related to really strongly.

Saul Marquez: Yeah Bulent I think it’s great. And you know it’s such a such a great opportunity for somebody of your mind. You’re a you know the things that you’ve been able to achieve and and understand. And now to have your mind here in healthcare to apply it in a way that could really help the system is a privilege. What would you say is a hot topic that needs to be on today’s health leader agenda as it relates to A.I. and how are you approaching it?

Bulent Kiziltan: Sure. I mean we could as an option go into specific use cases and talk about those use cases individually and there. There are a lot of them but I would rather prefer to talk about A.I. in general and how it can affect a whole ecosystem. And that’s not only based on use cases but it can help optimize the whole process of delivering health care. It can help make the ecosystem and hospitals much better. We today know that there are approximately a hundred thousand people dying from mistakes in hospitals. So can A.I. contribute to predict those mistakes ahead of time and prevent them? Can A.I. help making predictions about a flu? Can A.I. help us changing the behavior of members? We can talk about that in a more general sense. And I think also something that relates to A.I. is the data governance in healthcare. It’s very important if we want to move forward in that direction and if we want A.I. to deliver value we basically have to reform the current system in a direction that is conducive to A..I implementations in the long term Saul I’ll be happy to talk about those as well in a general sense and we can maybe pick some use cases and talk about those as well.

Saul Marquez: Sure yeah. We’d love to dive in and maybe this is a good opportunity to talk to park and and say folks so Bulent spend some time at Aetna doing some some work there with A.I. and optimizing things but not necessarily only optimization but taking a look at what this technology could do for members being covered by insurance. I’d love to hear from you Bulent maybe some examples that you did either there or or elsewhere and how you improved outcomes or improved results through it?

Bulent Kiziltan: Sure. So I will talk about some projects in general that I’ve worked on in the past and in different domains with different companies. And one of the powerful things that A.I. provides us is to make predictions based on data. And if a company or operations has access to some interesting data and if you have an A.I. team or data science team that has the creative capacity to join information streams coming from different domains there is a lot of interesting data in the public domain that unfortunately has not been used and utilized by many companies for a very long time. If those are combined some interesting insights can be produced using A.I. and producing those insights is actually not sufficient to produce the value that A.I. promises and realizing its full potential, one has to use that insight and turn it into actionable items. So that’s another layer on top of what the data science teams and A.I. teams are delivering. So this is why I think it’s essential that the A.I. operations are led by domain experts and really supported in a close collaborative manner by business leaders because there are different layers to the whole A.I. ecosystem and if it’s run by one or the other. There are interesting pitfalls that one can get stuck into and A.I. will produced the full potential value that it’s promising. So one specific area is for instance insurance companies and many companies in the healthcare space are trying to modify member behavior in a positive direction. So in order to do that you have to identify what these psychological barriers are for a particular member to not have healthy habits and that is a very kind of a vague or gray area. And data is not plentiful in that domain especially when it comes to member behavior. So how do you go about this? So there is there is a lot that I can offer in this space. For instance if you want to make members more drug adherent, if you want to if you want members to stick to their specific drug regimen, you have to understand what psychological barriers are in play here. In order to understand that there are different models that domain experts and academics have come up with. One of them is the PDC value which is a metric that gives you how drug adherent that particular patient is and you have to make predictions for the future of that particular member. And A.I. has made contributions in that area and we’ve been able to make predictions two or three weeks ahead of time for a particular member specifically if we have seen certain signals that are coming through that are indicative of that particular member to be not drug adherent in the future. We can customize the means to reach out to that particular member and give them an additional nudge that they need and even that customization that additional psychological nudge is being customized by machine learning algorithm so that this is a an interesting area where if a person has different chronic diseases at the same time the psychology is very different from two different members with the same chronic diseases. So it’s a complicated problem. I don’t think it’s a fully solved problem but there is some progress being made in that direction.

Saul Marquez: And when you when you expand this is a great example Bulent. When you expand this across say 20,000 people, you’re definitely going to find clusters though right. And then you’ll be able to address these different clusters in a certain way.

Bulent Kiziltan: Yes.

Saul Marquez: And there is the power.

Bulent Kiziltan: This is one aspect of what you can do you can do things in bulk and you can build your business strategy, your healthcare strategy based on those clusters as you put it. You can see certain trends in certain directions for different cohorts but then I think something that should not be overlooked is how individually you can really reach out to the member. So every member is different from each other. Their zip code gives a lot of information about some of this sociological obstacles for a particular member to reach the level of health that we would like the members to have. And so you have to really think of this problem you know far from a generic view where you look at clusters and then you build a strategy on that but also one can actually look into individual member profiles and come up with individually customized approaches and strategies to help those members to become healthier individuals. And I think A.I. can deliver on both ends.

Saul Marquez: Fascinating. How about from the from the perspective of a of a medical device company and I’ve got a good amount of med device folks listening in. What do you say to them?

Bulent Kiziltan: Yeah very important. I mean it’s a very active area. I have collaborators and friends in the academic world who are working on sensory data how to interpret them, how to analyze them, and there’s some effort in the industry especially the startup space to use sensory data. Bigger companies are building partnerships and I help foster some of those bigger partnerships with companies that have those devices already in place. We have obviously cell phones that produce a lot of sensory data, there are watches, there are healthcare gadgets that individuals are using and just tapping into that space of information is very interesting. I mean from a scientist perspective, I sometimes feel like a kid in a candy store but then tapping into that space of information brings an interesting obstacles and problems that need to be solved and by means I mean there’s so much more that can be explored in that space. But it’s one of the areas where A.I. on top of information flowing in from those devices will make an impact. It will be disruptive it will completely change the midterm or long term healthcare ecosystem, how we approach healthcare, how we approach individual healthcare, how we approach, how we can reach out to individual members, and how we can customize things. I mean it’s crucially important.

Saul Marquez: For sure. And so you’ve obviously been through a lot of different projects. Can you share a time when things didn’t work out. Maybe you had a setback. What did you learn from it and how did it make you better?

Bulent Kiziltan: Sure. I mean science in general and data science is not an exception is a trial and failure process. And you know it’s just through the process and depending on which domain you are. How much buffer you have for failure. In academia, that that space is not huge in the industry depending on the bit the company culture that space for failure can be more flexible. And then what you deliver may or may not have to be perfect depending on a watch which use case you are particularly working on… some of the successes we had we have been able to build some very interesting models in which I use some of my physics and astrophysics background to predict disease behavior how it’s spreading in continental U.S. for instance how humans interact which was very interesting for me and we’ve been able to come up with predictive models that that were very powerful in terms of predictive power but then A.I. in particular is an ever growing dynamically changing domain. And even when you talk with some of the pioneers of the field and ask you know which architecture which approach would you prefer for this particular problem, most often you don’t get a straight answer because it’s non-intuitive process. So you have to go through a trial and error process and sometimes if the information is just not in the data there’s nothing to extract right. So sometimes you just are trying different models trying different approaches you’re trying to come up with very creative ideas to implement and integrate into the process, some public data. But if the signal is not there there’s only so much you can do. And so we have sometimes not been able to extract some useful insights through that process. But what I see more often than not in the healthcare space specific producing value with A.I. is most of the value that I produced in the past have not come from cutting edge applications of new algorithms or a new A.I. tools. It came from a holistic perspective of the whole deliver process which I alluded to by saying you know producing insight from data is one thing but executing based on that insight is a totally different layer in which you have to take into account the whole ecosystem the internal dynamics, the silos, you know the you have to work very closely with the marketing team and with product managers which is something that scientists may not be too much used to. So having a kind of a holistic perspective of the whole process has been very valuable in producing the value that we’ve created in the past so it’s been an important metric in the successes we’ve had and some of the failures in the past also have not been always related to the technical process but it’s been about the culture and the cultural barriers within a certain company. For instance if you don’t communicate your results in an effective manner to the stakeholders internally you won’t be able to deliver and execute on the insights that you’ve produced. So I think communication is as paramount especially for an A.I. leader when you produce some insights to produce the value that you would like to do.

Saul Marquez: I think that’s a great call out and what would you say at the other end of that spectrum is something that that you’ve experienced success with one of your most proud moments?

Bulent Kiziltan: Are you asking for particular use case or should we be…keep it more generic?

Saul Marquez: You know you choose. I mean if you want to talk to us about something in particular that I think that would probably resonate more. Yeah let’s do that.

Bulent Kiziltan: Yeah. So let me give you my what one of the things that I have been thinking a lot about as a company culture as opposed to building a strategy and tried to implement the A.I. strategy I think I’ve come to the realization that a culture eats strategy for breakfast. I mean if your company culture is not conducive to implementing the strategy that you’re building, it will not happen it will not produce the results that you wanted to produce. So while in the technical aspect coming up with interesting, ideas extracting information is very very important and that delivery on debt and is easier in startup space as opposed to bigger companies in midsize and bigger companies. It’s often somewhat more complicated to execute on an insight that you produce. So as an A.I. leader when you’re moving into an operation while you’re building your talent pool, you’re optimizing your process, you’re trying to nurture your data science teams in a test to be a continual process because they had a science moving very quickly changing on a weekly basis so you have to really invest into the nurturing and educating and training of the data science teams. You have to also work on incrementally changing the culture and what I mean by that is that in the long term in order to sustain the value that A.I. creates, you have to create a collaborative environment and at least this is my subjective perspective on things. This is what I prefer. Certainly not all companies have this culture. It might just work in their favor for whatever their business goals are. But I prefer to build a collaborative culture internally at least that is continually trained and I take on mentorship individually on my own as well. But also I built peer mentoring programs internally where we have more senior data scientists helping out the Junior get a sense that are coming in the field and this is additionally important in a field like data science mainly because most of the data scientists including myself are not formally trained in data science because as a domain it didn’t exist and even the curriculum is not something that we have settled on in the academic world. So people are coming in from diverse backgrounds which I think brings in a lot of value but also it creates a lot of challenges internally because you have people that are coming in with different backgrounds who speak different languages. Technically speaking that is so in order to create a very productive process and a team it falls on to the leader to create that kind of nurturing environment which brings in diverse ideas, diverse talent from the outside, but also helps them to work effectively and productively internally so that there are challenges that comes with the whole process.

Saul Marquez: Culture is key.

Bulent Kiziltan: Culture is key and you can be in one of the two ends of this as a company you can be either act like a predatory company where you hire really top talent you attract them and then you eliminate them with a process and then you have a big turn that turn out in coming in and going out where you hire people and then you fire people. So this is one way to go about this and this is kind of an environment that I don’t find essentially conducive to A.I. itself but it might produce a short term value that a company is targeting for. But I think companies and industry at large has to work with the academic world in order to nurture the incoming data scientist. I mean you cannot just expect academia to produce all that talent and then you basically go through the list of talented people you extract whatever you want and then go to the next in the line. So I think there have been really interesting examples of industry leaders that are investing into building that hybrid institutions with academia, with universities, or invest in some other means by creating grants in order to kind of create an environment that is conducive to producing the talent that the industry will need. So it’s not a totally idealistic thing. It’s for the business value in the long term. So so culture is important. Continually training and mentoring is very very important. I think a collaborative internal culture in the long term is better for attracting talent and more importantly retaining the talent. So a if you, if a company wants to be recognized in the industry is in the industry for the long term I think there is no way around this collaborative culture. You just have to build that culture and it can be difficult for bigger companies to change and revolutionize or reform that culture internally.

Saul Marquez: Yeah especially publicly traded companies that are on a quarter over quarter basis. It’s short term over long term and an A.I. is at the stage right now folks where it’s going to need more long term nurturing as Bulent explained to us. Great call out there. Tell us about an exciting project you’re working on today.

Bulent Kiziltan: The most exciting one I cannot talk about unfortunately. But in the past. Just one of the two that I just mentioned working on different psychological barriers and how we can overcome them was an interesting project where we looked into the complicated and sophisticated interplay of different chronic diseases and how they play a role in making choices and what I found it was very interesting. I used some of the public data from Census and some other government institutions generically speaking in A.I. a one plus one is often more than two if you’re doing your job right and if the information is there so sometimes when you look at the certain information stream that does not have or apparently does not have the information content that you’re looking for at the level of strength that you’re looking for, once you combine that information stream with another information and if there is an interesting interplay between the two it can give you that edge that you required to extract inside from that information. So sometimes you know you can look into zip codes, you can extract information about the average income, you can look at the socioeconomic level, you can look at some other metrics and sometimes companies or data scientists may go about manually eliminating some of those some of the columns or features that are coming with that data. I would strongly suggest otherwise to use everything possible and see at the end whether in conjunction to other data when you combine them in a careful manner and come up with a creative idea that you want to extract sometimes what you will see is you have you can get a contribution from an interesting feature that apparently may not have an impact on your insight on its own. But once it’s combined with other things you’ll see that it might actually produce some some interesting information that you require. So I think that that aspect in that strategy has produced interesting insights in the past in particular, looking into extracting information about psychological barriers for individual members and also in other project was to understand flu in general there is a lot of waste during winter times when members get flu like symptoms they go to the E.R. as opposed to urgent care and you know you want to prevent flu like symptoms or you want to have the symptoms you want them to get the proper care and not go into going in a particular direction that will cost a lot of waste before the companies for the individuals and in turn they won’t necessarily sometimes get the best service that they deserve. So you know again even in that area when we looked into the whole process of predicting the flu whether we can make predictions about certain zip codes in town so when the flu season will peak we made some progress using A.I. and some physics actually.

Saul Marquez: Oh Bulent I think it’s it’s fascinating right. And as soon as we start to take a look at how we can apply A.I. technology and theory into practice I think the call out here is one plus one doesn’t always equal two it’s when you combine those two data streams that we could potentially start to see those insights that then you can take the next challenge which is make them actionable. It’s definitely hard work folks. So you’re going to get into a giant and hit it out of the park with little effort. Just try something different. If you’re not all and don’t get in right.

Saul Marquez: I will be more encouraging. But it’s true though right. I mean… you don’t want to falsely encourage.

Bulent Kiziltan: Well it’s not a walk in the park that’s for sure. Yeah I would encourage anybody with backgrounds that may not seem very appropriate for A.I. they can get some proper training and bring in the diverse background that other data scientists might not have. And most of the time when I conduct interviews well there’s a minimum level of skill set that one has to look for. But what I look for more importantly is the creative aspect of an individual because data science is largely a creative process, it is not something that you can pick you know a black box and feed in your Excel sheet and come up with insights. I mean that’s not A.I. Unfortunately we’ve seen this over and over that many companies think that they can buy off the shelf, software, and feed in the data that they have in an excel sheet form and create some interesting insights that way. You may but you know that type of value will not be sustainable in the long term. So what is very important is the creative process of the data science effort and that requires really a diverse talent. I think this is very important so I would encourage anybody to you know start learning and there is so much information out there on the Internet. So one has to really learn to drink from the fire hose sometimes and that takes also an effort mainly because universities and institutions still with few exceptions. There is no there is not a formal way and curriculum to produce great data scientists. I mean there are certain skill sets that you have to gain through the process but you can do that online as well.

Saul Marquez: Love it. So Bulent getting close to the end here, it’s time for our lightning round. We’re going to produce a little mini syllabus for our listeners five points here lightning round style followed by a book you recommend to them. You ready?

Bulent Kiziltan: Wow. Okay.

Saul Marquez: All right. What’s the best way to improve healthcare outcomes?

Bulent Kiziltan: Well invest into A.I. definitely investing into A.I. in general but also when you’re investing into A.I. make sure that you have domain leaders that are driving the strategy and not only business leaders.

Saul Marquez: What’s the biggest mistake or pitfall to avoid?

Bulent Kiziltan: Is to have either a business leader only or in a person that comes from a purely academic background to run the whole show. I think it is essential to have both perspectives in a balanced manner fashion to run the A.I. process and strategy for a company.

Saul Marquez: How do you stay relevant despite constant change?

Bulent Kiziltan: You have to keep reading the latest literature. I still read or at least go through the titles and abstracts of almost all articles that are coming out not only in machine learning A.I. from the relevant websites like archive but also from my own domain astrophysics so I invest quite a bit of time in reading the literature and also taking on projects on my own in addition to my leadership roles.

Saul Marquez: What’s one area of focus that should drive everything in A.I. efforts?

Bulent Kiziltan: I think transfer learning will be an interesting area in other areas of technology. It will be may create a disruptive value maybe in 2019 in healthcare mainly because the data governance has not been that great. It might take a little more time but I think a transfer learning will become a very important topic in a I mean because you don’t want to spend the whole computational time on continual training and reinvent the wheel.

Saul Marquez: I love that. And Bulen and last one here. What would you say your number one success habit is?

Bulent Kiziltan: Well that’s a tough one I really don’t know what that might be but my approach to managing and mentoring larger teams is I am a part of the team and if I cannot do the job that a member of my team is doing I wouldn’t be a leader that I want to be so I take on projects on my own I make sure that I know everything at least I do everything at least once in the whole process and it’s a continual learning process. So this has been my approach to academia. This is my approach to data science in A.I. in general. You have to be a constant learner and a good student and I want to remain that student.

Saul Marquez: I love that. What book would you recommend to the listeners Bulent?

Bulent Kiziltan: There are few that I really like and have played an important role in developing into the person that I am and you know it’s a continual learning process and evolution but I think what is very important in everything that we do including academia leadership and A.I. is communication and everybody has a very different style very different background and they bring in that personal intrinsic bias into their communication. So I think investing into communication is critically important to produce the business value that we want to produce or the academic impact that we want to create. In the business setting you know business people they speak a completely different language than people who are coming into the field of data science and that creates a toxic environment in order to alleviate that problem I think a nonviolent communication is essential and there is actually an older book that’s called Nonviolent Communication: A Language of Life by psychologist Marshall Rosenberg, which I find very valuable in terms of giving the insights into techniques into how to communicate in a nonviolent fashion especially if you’re a leader, in the academic setting I think also this is important. But the training that I have gone through at Harvard specifically for communication was somewhat different. But you know putting the two together being a public speaker, being an educator, and then reading this book about nonviolent communication was really a valuable for me.

Saul Marquez: Love that. Great great example. And folks you go to outcomesrocket.health type in bulent in the search bar. You will find the show notes an entire transcript and this mini syllabus that we just constructed with you along with links to all the things that we’ve discussed about, go there outcomesrocket.health type in bulent in the search bar. This has been a blast Bulent, I love if you could just share a closing thought and then the best place for the listeners could get in touch with and follow your work.

Bulent Kiziltan: Sure. It’s been a pleasure Saul thank you for the invitation. Closing thought you know talking about A.I… A.I. has so much to offer for almost every domain in the industry for academia and I want really companies to not only keep focusing on the short term outcomes but also focus on the mid-term and long term deliverables that A.I. has to offer. A.I. is not an optimization tool only it can optimize really well but there’s so much more that A.I. can offer, it can help with reducing the number of deaths in hospitals into healthcare by mistakes, you can produce models and make predictions when those errors might happen and the cost for those errors and mistakes is the life of a person. So it’s something that has to be taken seriously and in order to keep that vision for the long term I think it’s essential that this company strategies are set by domain experts in conjunction with business leaders I think hitting that balance point and building that culture that ecosystem around it in a collaborative fashion I think will help the companies to create the value that they want and also for the future, I think it’s gonna be an exciting year ahead of us not only in healthcare but in general where A.I. is going to be disruptive in almost any domain. So I would imagine that any leader in any business sector has to learn and become educated in A.I. to a certain level it will be kind of the computer knowledge of the future is if you have to learn A.I. to a certain level in order to make informed decisions for the future of your company.

Saul Marquez: Get to know that A.I. folks and Bulent finally what would you say the best place listeners could get in touch with you at or follow your work?

Bulent Kiziltan: Through my LinkedIn profile I have a website that I kind of try to keep up to date but I’m not too good at it but if they type in my first name and last name or just my last name I have a personal website where I post every now and then it has links to my LinkedIn profile and my other social media links.

Saul Marquez: Beautiful so folks bulentkiziltan is it .com?

Bulent Kiziltan: .org.

Saul Marquez: .org? So bulentkiziltan.org, we’ll leave a link there in the show notes as well so if you’re curious dig deeper sharpen your skills stay with it folks and Bulent really appreciate you spending time with us today and have a great one.

Bulent Kiziltan: It’s been a pleasure. Thank you.

Thanks for listening to the Outcomes Rocket podcast. Be sure to visit us on the web at www.outcomesrocket.com for the show notes, resources, inspiration, and so much more.

Sonix is the best online audio transcription software in 2019.

The above audio transcript of “Artificial Intelligence in Healthcare with Bulent Kiziltan, Chief AI Officer, Chief Data Scientist at Stealth Mode Startup” was transcribed by the best audio transcription service called Sonix. If you have to convert audio to text in 2019, then you should try Sonix. Transcribing audio files is painful. Sonix makes it fast, easy, and affordable. I love using Sonix to transcribe my audio files.