The Intuition Behind Generative AI

James BenhamOn this episode of the InsurTech Geek podcast, we're talking about the intuition behind generative AI with Frank Schmid from GenRe. The InsurTech Geek podcast is all about technology that's transforming and disrupting the insurance world. We'll be interviewing guests and doing deep dives into specific tech we see changing the industry. We're taking you on a journey through insurance tech, so enjoy the ride and geek out. Terra, this week's episode sponsor, is revolutionizing the world of workers' compensation claims and policy management. Terra's cutting edge platform empowers insurers with comprehensive benchmarking tools, robust management systems, and an integrated ancillary services marketplace.

Discover how Terra is driving innovation and efficiency in workers' comp at terra.insure. That's terra.insure. And we are back with another wonderful episode of the InsurTech Geek podcast. I'm James Benham, your InsurTech Geek live and in charge from beautiful College Station, Texas. With me joining from Stanford, Connecticut, the man, the myth, the legend, man who, has has has had a lot of hats in the insurance industry. I'm very excited to have, Frank Schmid on the show with us today. Frank, how are you doing?

Frank A. SchmidI'm great. How are you, James? Having a great day. It's

James Benhamsummer temperatures in Texas. It's gonna cap a 100 degrees today. It's, unusually warm, so I am inside enjoying the air conditioning. But, having a having a great day, and I'm just geeked out to talk to you. How are things up in Connecticut? Is it I'm guessing a little bit cooler up there near the near the coast?

Frank A. SchmidYeah. Gladly. Actually, I'm a runner, so I I'm averse to high temperatures. Now we have had quite a bit of rain, and today I was running in the rain. It's a rare exception, actually.

James BenhamYeah. Nice. You know, I I it was when I went through the ROTC military training at Texas A and M, I was in the corps of cadets there. We ran every day, rain or shine. Didn't matter. And at first, I was really averse to running in the rain and then actually kinda got to enjoy it. You know, it's a a natural cooling system considering how hot it gets here.

Frank A. SchmidThat's right. Frankly, after the first mile, you you don't realize it's actually raining.

James BenhamYeah. Yeah. Because you're you have the sweat and the rain kind of all mixed together, and it's it is what it is. I I I I never did super long distances. I think the longest distance I ever ran was 13 miles. Now the longest distance I ran is I I run is probably five kilometers. I've I've I've definitely shrunk my running portfolio down to about a five k on a regular basis, but I really enjoy it. Are you a long super long distance runner or do you sprint?

Like, what are your what what do you like to I'm

Frank A. Schmida marathoner. I actually have the Brooklyn half marathon coming up on this Saturday, and I just returned from London. I ran the London marathon.

James BenhamWow. That's awesome. So how did it go? Well,

Frank A. Schmidthe London marathon is a bit congested, quite a number of runners, but it's clearly a memorable event and I'm happy with my time.

James BenhamGood. Good. That's great. What's your favorite marathon?

Frank A. SchmidFavorite marathon is when I can run at a steady pace the course is not too crowded. These are usually smaller marathons. Berlin, I think, one of my better marathons. It was quite a while ago, actually. Okay. It's a flat course, and it's just wonderful.

James BenhamYeah. Me and Hills are not good friends. Same thing with wind. If it's a really windy day, I'm not super motivated to go running. I just don't like getting running directly into the wind. It's exhausting. But, I'm not a I'm not a marathoner. I've never ever done. I've only done half that distance and and, you know, not not my, not exactly my cup of tea, but I'm excited that, that you're you're you're an avid runner. Let's talk about you a little bit more. You grew up in Germany.

And what did you go you you have a really great education that you went through, Heidelberg University, and you got a a master's in economics. You got a bachelor's in economics. I mean, you really dove down the rabbit hole, then you got a doctorate in economics from Lufana, and then you got a postdoc in corporate finance. Then you went to Wharton and studied there too. So, I mean, you you really poured a lot of time and attention into your education.

Frank A. SchmidYeah, James. Well, I'm interested in many things. And, you know, my mind keeps traveling. And I'm also physically relocated a bunch of times, as you indicated. I have always been interested in questions of society, right? How society is organized or how it could be organized. That got me into history. And when you're in modern history, get into the Great Depression. And so I got into economics, similar to Ben Bernanke, who spent his academic years on the Great Depression early on. And yeah, so I studied economics and increasingly, I got interested in finance.

So I kind of migrated from economics to finance. And, yeah, with that, I joined the Fed at some point. And at that time, finance was not yet what it is today at the Federal Reserve. It was more traditional economics. And so it was quite an interesting time in the 1990s.

James BenhamAbsolutely. Well, I like economics because fundamentally, it's the study of human behavior. Right? That's right. You know, it's how do humans respond to different stimulus, right? The lowering of a price, the scarcity of supply, the the excessiveness of supply, or the the the fear of uncertainty, right, which, of course, drives a lot of decisions. And so I I did not go into a I I I got an undergrad and a master's in business, and I obviously was required to take economics as part of that. But my I love history books.

I love reading history books. And so I think it's quite useful to be in in particular in our industry and insurance to be a student of history because history tends to repeat itself all too often, in particular to those who don't study it.

Frank A. SchmidYeah, I think it's helpful when there's something new that you observe in economics that you actually go back in history. And, you know, for instance, the arrival of generative AI as a general purpose technology is such an event where you want to go back in history and see how technology is being adopted.

James BenhamYeah. And and certainly, like, I like to go back and look at the Luddites and the industrial industrial revolution in England and look at how they responded to the automation of textile plants and, you know, the what happened to the unemployment rate, which uptick, you know, several percentage points and then actually went down nine years later to lower than before the industrial revolution started. You know, there's some really interesting things I think that happened that are that are interesting parallels. You know, mechanical automation, and, obviously, right now, you've got, you know, intelligence automation.

And so it's, it's it's fascinating. Let's let's kind of put put a bow on your your history because you you went to work in The United States after going to Wharton, for the Federal Reserve Bank of Saint Louis as a senior economist, and that was a good seven years before you jumped into insurance. What did you learn there that led you to want to join the ranks of the insurance professionals?

Frank A. SchmidYeah, James, at the time, I had a great deal of interest in Bayesian modeling, Bayesian economet econometrics. And so the actuarial world is one where you can really build powerful applications of Bayesian statistics and Bayesian econometrics. Also, I had an interest in labor economics, which brought me to workers' comp as a choice, as a line of insurance. And frankly, it was a great time. I was able to mentor junior actuaries. I was able to do a fair amount of modeling and get to know the insurance business.

James BenhamAnd you worked at, certainly, an organization I have a lot of interaction with NCCI as a senior economist and then as a director and senior economist. They they had their conference this week, and I I'm super deep in work comp. Probably 70% of the technology we build is for workers' comp organizations. So you can imagine I have a lot of intersectionality with intersection points within CCI. You know, and and I I really like the what what you said kind of about the the labor markets and NCCI being being connected.

Did that lead to further jobs and work comp or did you jump into other lines of business when you went to AIG?

Frank A. SchmidWell, at AIG, I was recruited to build the Federal Reserve stress test model for long tail liabilities or for P and C liabilities in general and work comp. Clearly, these liabilities are long tailed and particularly complex in nature. So I did get to know other lines of insurance. But in terms of like long tail lines, work comp really stands out as particularly complex and, challenging.

James BenhamYeah. I agree. That's why I like building software for it. There's there's a lot of challenges. When a claim can last, eighty years, when, you know, policy and underwriting can be really complex, setting rates. I mean, it's a it's a very dynamic space. Isn't it interesting right now that workers' comp is the kind of the golden goose right now in insurance? It's got the some of the best combined ratios in the industry, and people are really chasing it. It's a big change from fifteen years ago.

Frank A. SchmidYeah, exactly. It has been quite attractive for insurers for quite some time now. There were challenges with the opioid epidemic, but that has been overcome largely through various measures. It's also a fairly steady space in terms of regulation. Yeah, I think the labor market also helps, right? We have had a rather healthy labor market. And so the incentive, you mentioned economics is about incentives, the incentive to actually return to work if you can get wage increases and have career opportunities is just greater than when the labor market has a lot of slack.

James BenhamAbsolutely. Well, let's let's jump in and talk about your current role. You're in it looks like you've reinvented yourself a few times in your career. Yeah. I I would say you're in, like, version four point o right now. I if I could put version numbers on your on your iterations, I'd call this Frank four point o. You're the chief technology officer at Genre. Can you give us a glimpse into, like, what a typical day looks like in that job and what kind of projects you get to work on?

Frank A. SchmidOh, a typical day has a fair number of meetings. I have a great team. So, you know, in a modern IT department, the hierarchy is pretty flat, right? And I'm leading engineers without being an engineer. So it's a different management style. I would say, attention to detail, get to trust your people, but know their weaknesses where you have to compliment them. So I make decisions only clearly if, you know, my direct reports are not able or not authorized to make the decision. Right? But other than that, it's great because we can have time to think about things at a conceptual level in technology, and I'm quite grateful for that.

James BenhamAwesome. And what what are the big projects right now that are that are really getting some attention? You don't have to give specifics so you can talk about, you know, generalities if it's confidential, but just help me understand the kind of projects that you're that you're working on.

Frank A. SchmidYeah. I think security is always an area that that keeps evolving so naturally. We are modernizing our edge infrastructure, also from a security perspective, moving away from, let us say, tools to more platforms that are more powerful than are modular. We have completed our cloud migration internationally. We have built a global data architecture, and now it's really about building AI as a layer on top of the data architecture. So clearly, I'm spending a lot of my time with my chief data officer on technology adoption or AI adoption in particular.

James BenhamSo what's the interaction like between your CISO, your chief information security officer, and your chief data officer in you? Like, how do y'all are y'all parallel? Do you divide duties? Like, how does that function for y'all?

Frank A. SchmidYeah, I mean, they are clearly at the same level. I would say where intersect is data governance, clearly, right? And cloud engineering. So we have built DevOps team so that the chief data officer has his own infrastructure person and his own security person. What he builds is secure by design. And also I did merge the infrastructure tower and the security tower. So my SSO is also leading the infrastructure team. And so, I mean, that that was a major that really helps coordinating between infrastructure and security, which has been a bit of a challenge, in the past.

James BenhamVery cool. Well, let's jump in and talk about AI. You've mentioned it a couple times. Certainly, I'm using AI tooling across our organization. We're a we're a 300 employee insurance software company. We build proprietary solutions for carriers and TPAs, brokers, ancillary providers. We also have our own off the shelf product. A separate team of mine has their own off the shelf product. We spend all day, I mean, all day buried in application development and have seen a a bunch of really amazing gains around using AI for technology development.

We've also seen a lot of gains around using it for customer support. We've seen a lot of gains around using it for specs and BAs and QAs. I mean, there it's really been a multi tool for a lot of the different job titles for us. But then, you know, you have to take it a step farther, which we've been doing, I'd say, in earnest for the last few years of baking it into the product line so that our users can actually use an embedded AI tools and not not copy and paste or scrape data or copy a screen and then use Chat2PT.

They can actually use an embedded AI tool. So talk to me about like, you you recently wrote an article around generative AI and as an abstraction layer that makes powerful predictions. So, I mean, tell me about what that means. And then tell me if this shift changes the way insurers should think about predictive modeling and the decisions they make from it.

Frank A. SchmidYeah, James, predictive modeling is is is one aspect in the AI adoption. Well, we have seen the rise of generalized linear models right in the 80s and 90s, and they are quite powerful. And the insurance industry uses it extensively and these models, they essentially rooted in Newtonian mechanics in the sense that they try to uncover causal relations and they give us regression coefficients that tell us the magnitude of the effects and the effects of these x's that we have on the right hand side of the regression equation.

And then came machine learning, right? And it's just, I hesitate to call it a black box.

James BenhamFeels like that.

Frank A. SchmidIs a box, right? And it gives you predictions and we don't fully understand the inner workings. And then we go from traditional machine learning to deep learning, which is the engine behind generative AI. And now we have very powerful prediction machines, but we don't understand the machines, right? We don't understand the inner workings. We no longer have these simple linear relations that we can point to, these effects that we can point to, and we can talk about the magnitude of these effects. Instead, we have an input on the right hand side and we have predictions on the left hand side.

And so these predictions are, I mean, this type of modeling does not lend itself to narratives in the way that generalized linear models lend themselves to, right? And we know how important narratives are for decision making in corporations, right? This is the way communicate, in a cohesive way and make the case for our decisions. And so the question is, how do we deal with these prediction machines? And you mentioned the abstraction layer. So one way is just to think of generative AI as an abstraction layer where that is hiding actually an abstraction layer hides in a working, right?

We also have this in software development. We just evaluate the model or the prediction machine essentially based on its predictive benefit, right? Does it deliver what we expect from it?

James BenhamYeah. I I've been struggling with the same thing that you described. I mean, typically, I've been writing software since 1991 when I built my first GW basic application that used a flat file data structure through object oriented programming and structured databases through web based application development. Certainly, mathematics is a core part of computer science. And, you know, if you understand predictive analytics, you understand it's really linear, you know, regression models. This it's mathematics. Right? You you're we're using we're using math formulas to look for the the the big two things that I think anybody in insurance should be looking for correlation and causality.

Right? And and that is that has been my challenge with machine learning. And then you know, these complex large language models from the start is back tracing and showing the work so that we can prove the outcome rather than just trust the outcome. Right? So are you seeing AI models get better at showing their work and proving the outcome?

Frank A. SchmidJames, I think we can develop an understanding of these models over time, and we can actually use AI to build that understanding. But there may always be a layer of understanding that we will not have. And, the analogy I use is when we learned how to drive a car, right? So you don't have to understand what is going on under the hood to drive a car. Instead, you use, driving as, you know, as an abstraction layer. So, you know, when you turn the wheel to the right, that the car will veer to the right.

And if it doesn't, you will have to correct your,

James Benhamabstraction. You're right, though. And we deal in abstraction layers in computer science all the time. Right? When we you really remove yourself every innovation that has been invented. And AI is like the ultimate abstraction layer to me because you're you're so far from the machine code. Like, you're you're you're miles away from the zeros and ones.

Frank A. SchmidExactly.

James BenhamAnd and that's challenging. I mean, the first thing I learned was GW basic. The second thing I learned was assembly. Like and, like, we got down into the the root code, like, getting into machine code assembly at the at the at the raw hardware level. And and what you're dealing with now is so far abstracted from the actual computation that we have to do a lot of thinking about how much we can trust this, in particular in insurance when you're trying to use this to make underwriting recommendations.

Of course, I'm a big believer in human in the loop. I'm sure you are too. That, you know, software and reporting and data are there to inform decision making, but not to replace human decision making. They're they're there to augment the decision making process to make it easier to collect, sort, and analyze data. Right? Correct. That's exactly our approach. Yeah. Because we have really smart people. We wanna keep them around and leverage their experience. We want them to stop doing the boring menial work of data aggregation, collection, and analysis and start making recommendations and presenting the facts and then letting them make the final call.

Do you think you know, you're a CTO. You do tool selection. You look at reporting. Like, are you guys a Power BI shop? Or what what are you using for for BI as a tool?

Frank A. SchmidYeah. We we are a Microsoft Azure shop. So, yeah, Power BI is an important reporting tool that we have, and we consume all our AI capabilities, at least those that we built through the Azure pipeline. Clearly, AI is embedded now in in applications or platforms like ServiceNow, etcetera. But what we build, we consume through the Azure pipeline.

James BenhamYeah. We do too, by the way. Totally Azure shop. I started this company in my dorm room in 2001 writing code with ASP and SQL 97. ASP one point o and SQL 97, and we're still building it. And we are we built our first insurance application, I think, in ASP two point o and SQL 2,000 was the first web based insurance app in o four. So it's been a while. I'm a big fan of the Microsoft stack. I think they've made it a lot easier for us to deploy and leverage AI through cognitive services than a lot of the other providers.

And to me, in a higher trust environment, like Microsoft really does their their EAs kind of run the their enterprise agreements kind of run the show. So I think it's a it's a good deal. What what's been your favorite use of AI so far that's actually made it into production? Not the stuff we're experimenting with. Like, what's the stuff that you're fired up about that actually made it I have mine ready, so I'm gonna tell you mine after you tell me yours. Your favorite, like, one that's actually made it in.

Frank A. SchmidWell, mine, James, is, clearly the copilot, for, you know, rewriting or drafting emails, summarizing documents. But that's maybe not the most important application that's in production at January. Our coders use various copilots, especially the Databricks assistant quite extensively. And this was also the first, really, first AI systems that went into production was really for the coders.

James BenhamOh, yeah. The first thing coders did was teach AI how to write code. That's right. What could possibly go wrong? Right? Code writing code? Oh, jeez. Hey, Frank, do you think that AI has already written the singularity and just didn't tell us?

Frank A. SchmidMaybe we wouldn't know.

James BenhamI don't think we would. I think they'd hide it. Just little side note. Just little little sci fi worries. It's not like we have a global satellite network that they could tap into and use to communicate with each other. Oh oh, wait. We do. Sky does here. Okay. So let let's let's let's keep it by by the way, my favorite, work comp. We use a set of AI tooling to recommend treatments, predict the ROI on the treatments, make recommendations on reserves, and summarize the medical notes.

That's been, like, super it's in production, and it's super useful. Dramatically shortens the amount of medical notes and record notes that adjusters have to read. So Yeah.

Frank A. SchmidAbsolutely. And I mean, summarizing medical records is clearly one of the applications that are greatly beneficial for claims team. Yeah. So

James Benhamlet's make let's kinda wrap with Frank's recommendations for other insurance companies. You you're you're leading look. An incredible business. Genre is respected all over the world as a major leader in reinsurance. You're doing some cool stuff. You've had a bunch of stops in your career. What steps should insurance companies take right now to prepare for and adapt to the growing influence, not just of generative AI, but all the tool sets AI has to offer? What should they be doing? What should they be investing in right now?

Frank A. SchmidWell, James, the way we look at technology, that's it's highly lexicographic, right? So first comes to cloud. The cloud enables a modern data architecture, and a modern data architecture enables an AI adoption at scale. So that tells you what you need to do, right? If you're not in the cloud, start your cloud journey. If you're in the cloud, but you haven't built your data architecture, work on that. Because generative AI sits at the top. Now, the generative AI adoption process, I like to compare it to electricity, will take many years, right?

It will be faster than electricity adoption, but it will take many years. So let's be patient. And depending on the size of the firm, the larger you are, the more you want to build rather than buy. And clearly, lot of AI comes embedded in applications, but I think we have to increasingly think data and data flow rather than applications. For us, generative AI and building AI systems is really orchestrating the Azure capabilities or more generically the capabilities made available by our cloud service provider. And so the cloud, cloud engineering, data architecture are really the foundations.

James BenhamI agree with that a 100%. We went we decommissioned our last on prem server, jeez, four years ago, which which was a lot. You know, we're a multi we're not. Okay? We're not your size. But we work for a lot of companies your size, and we have a lot of data that we had to we had to really rearchitect. We didn't just do a lift and shift physical to virtual. We actually architected all we rearchitected all of our solutions to be cloud native. So we're serverless.

We don't have virtual servers, don't log in to servers. We use SQL Azure. We use a bunch of other tooling. We use dot net core, you know, and app we we we really have leveraged the cloud. And it's amazing how much easier it made it for us to leverage all of cognitive services. It really, really was a critical first step. Like, if there if there's on prem you you can run cloud you can run AI models on on on prem solutions, but it's not nearly as easy.

And I I totally agree. I totally agree. That's a that's a big deal. And then, of course, the middle step that you mentioned with with a modern data architecture, you know, really rethinking how you store data and where, where you leverage, you know, relational and structured data burst basis versus, you know, nonrelational databases and and flat data structures, which are really, you know, super viable now. So I I I agree totally spot on. Okay. Well, look. Last thing, quick question. What personal and I'll give you my answer as well.

What electronic or software, not for insurance, available for you personally are you most excited about that you already already purchased or want to purchase?

Frank A. SchmidWell, I'm excited about my Garmin watch and the app that comes with it. So it's quite popular among marathoners and it gives me a dynamic workout schedule five days ahead and adjusts to my bodily condition, how much sleep I get, how easy or how the workouts were for me, how well I have recovered. So that's frankly what I'm most excited about.

James BenhamThat's I didn't know it could do that. That's incredible. I can start being very useful. As a I fly airplanes. I'm a pilot. So I use all Garmin devices in in aviation. I love their suite of avionics. It is awesome, and they have an Aviator watch that literally connects to your airplane that syncs all of your data with your watch and your phone. So it it they do a really neat job. I'll tell you this. My most recent purchase that I'm most excited about that I'm using the heck out of is my new Tesla model y.

I got full I got full self driving. I can drive from here to Houston, which is an hour and a half, here to Dallas, which is two and a half hours, and never touch the steering wheel. It is absolutely remarkable how good full self driving is. Elon said they're five years ahead. I'd peg him even more than that. I think he's even more than five years ahead. It is absolutely awesome. An unsupervised mode comes out next year for the new model wise because we have new hardware new hardware package.

So I must say I didn't anticipate to love it this much because I'm a German I'm a German. You'd you'd appreciate that I'm a German car guy. Is my is my favorite, but their self driving doesn't even hold a candle to Teslas. It is absolutely incredible. So I am I'm most excited about that. I think it's I think the future's now when I drive around, I feel like I'm, like, I came back into the past in a time machine.

Frank A. SchmidYeah. I think my daughter agrees with you. She also likes the new Model y. Doesn't own it yet, but that's what she would like to drive. Yeah. It's it's incredible.

James BenhamThe the new y, specifically that one is incredible. Yeah. So well, look. It's been a great conversation, Frank. I I really appreciate your time. Appreciate your thoughts, your guidance, and your leadership, the Federal Reserve, you know, all the companies before Jen Re, and, of course, now as chief technology officer of Jen Re. Thank you for coming on the show, and I really appreciate your insights.

Frank A. SchmidThank you, James. My pleasure.

James BenhamAnd thank you in listener land for tuning in today to geek out for our interview with Frank Schmidt from Jen Re. This is your InsurTech Geek, James Benham. See you next time. The InsurTech Geek podcast is all about technology that's transforming and disrupting the insurance world. I'm your host, James Benham. That's jamesbenham.com with co host Rob Galbraith at endofinsurance dot com. We're taking you on a journey through insurance tech, so enjoy the ride and geek out.