Transcript
-
0:09
Venky Ananth:
Hello and welcome to the next episode of PaceSetters. Venky Anant. I lead the healthcare business for Infosys. Today I have a special guest, Paul Hlivko, Executive Vice President and CIO, Wellmark. Look, this episode is going to be very different for me and for PaceSetters because we're not going to cover healthcare, but AI in healthcare. So it's going to be interesting. And before I say AI and before you roll your eyes, I know there's a lot of fatigue on AI, but this one, I promise you, is going to be very different. Different because Paul is a contrarian. In fact, he says that AI doesn't hallucinate VR. So it's going to be a very interesting journey and let's dive right in. Welcome, Paul.
-
1:01
Paul Hlivko:
Great. Thank you. I'm happy to be here.
-
1:03
Venky Ananth:
Awesome. AI is everywhere. As you know, it's on a hype cycle. And before we dive deep into AI, I want to really understand your background, Paul. I know your MIT Sloan grad, but then you start as a startup guy, move into finance Wall Street, and then you end up in healthcare as a CIO of Wellmark. Just walk us through your journey and you know, this spectrum of experience that you've had and especially you've gone through many cycles of, you know, the 2000.com bubble burst and then 2008, the financial crisis. So you also kind of weathered through many of these crisis. So just give us a sense of, you know, your own background and how do you, how has that influenced your, your leadership style today?
-
1:55
Paul Hlivko:
Yeah, sure. So technical progress has always fascinated me. So at a very early age, I was like kind of through the Internet cycle and the PC cycle, just fascinated with how that can impact business, impact society. And I lean pretty quick into the startup world. And the first, the first venture was trying to bring technology in an emerging market. It was right before the.com boom into the commercial and residential real estate race. And I think what what I learned through that effort was interpreting a market cycle and interpreting emerging technology and then applying it in a business environment are very different. And seeing kind of how the markets reacted to the boom, how they reacted to the bust, really helped me crystallize the effort that it takes to actually take something from an emerging tech or an invention at the time, which in that case it was the Internet into actual business where you're generating margins, you're operating in a broader economy and a market. You have customers, you have people involved in your organization. And I think that crystallized kind of my view of technology and the impact it can have both in a company, but also the impact it can have on society and customers as well. I think the other thing that was formative for me in my career, and you mentioned it, running a startup really gives you a broad aperture. Like I had the opportunity to add a small scale, be the CHRO, be the CFO, be the chief legal officer, like you have to fill almost every role running a small business, especially in an early stage. And that gives me great perspective now thinking about the challenges of new emerging tech. And we're, I know we're going to talk about AI today and the implications it has on professions, but also a lot of the barriers that are going to exist with adoption cycles, which we see in all all forms of emerging tech. There's all these barriers that put up kind of resistance to the diffusion of technology in an economy or in a, in a business. Finance, I think was my, the second journey of my career and then healthcare. Now I've always been fascinated with markets where there isn't a physical product, right. So in as a health insurer or in, in, in Wall Street and corporate finance, like technology was an engine that drove a lot of the product cycles and the innovation of the markets because I think we then have an outsized role in helping a company compete and ultimately helping customers.
-
4:26
Venky Ananth:
Let's dive in with that background into AI. You say that there is a big misunderstanding about AI and it's not even technical, it's human kind of what blind spots are we not seeing here? What do you mean by the fact that there is a misunderstanding that and it's not technical?
-
4:47
Paul Hlivko:
Let me start with, there's a variety of different ways to think about AI. But I, I do want to first say that I think it's going to change the world for better. Like I think technical progress in general has really improved society. It's levelled the playing field. It's distributed information across the world more evenly. It's created more economic opportunities for everyone. So I think generally it's a positive. I think where we often struggle is different biases we don't necessarily account for. And so we focused a lot on the bias in a model. Like when ChatGPT came out, there was a lot of talk about bias in models, which is still something we need to pay attention to, but we didn't necessarily think about all the human biases and how we think about the adoption and diffusion of tech. It's like recency bias is a common one. So we saw the utility of ChatGPT 2 1/2 years ago and then we saw, we heard about all the earnings calls and CEOs talking about AI adoption, which is great. We then saw almost anyone on a Business Media channel having to talk about AI to be relevant. It's like this perpetual narrative around AI is really driving a lot of recency bias in our assumptions that we're actually going to see immediate value tomorrow. And then humans tend to have a forecasting error, like we assume change happens faster than it actually does. I mean, that is something that is built into us. And then loss aversion's another another thing that we're all a bit dealing with on the flip side of the equation. Like when you think about AI showing up in your workplace or showing up in your personal lives, you're much more likely to protect yourself from a loss than you are to adopt something that is new. Like that is a human nature issue. So like, we're dealing with all these human biases and we're not necessarily talking about them. I think when you think in a structured way about AI adoption and you go seek out information that is structured in its analysis, you find different results. Like Daron Acemoglu, MIT economy, He's an economist at MIT, recently won the Nobel Prize. Like his research a couple of months ago showed that about 5% of tasks in a 10-year time frame are going to be profitably automated by AI. I mean, that's a far cry from like what we read from McKinsey as an example. And he did in a very structured way, thinking about like individual roles and individual tasks, which are all documented in in an O net database and doing that analysis to understand like what should we expect in the future? So I think we need to be more structured and how we think about AI adoption and not necessarily just assume that it is the hype cycle that's going to drive the entire story.
-
7:39
Venky Ananth:
That's fascinating. You mentioned McKinsey. I was reading an article recently which said that McKinsey is predicting $17 to $25 trillion in actual added economy annually. That's like, you know, on the other side of the Chinese economy annually for the global GDP. So where is this disconnect between, you know, this entire AI hype that everybody is obsessed about, right to enterprise reality? You know, where is the disconnect here in your view, Paul?
-
8:16
Paul Hlivko:
Well, I think there's possibly 2 disconnects in thinking about where AI is at now and then how it shows up in an enterprise. The first is understanding diffusion. And diffusion is kind of what the the term is for causal models of adoption of tech. And this has been looked at over the course of twenty, 30-40 years, everything from like how fast people adopted microwaves to how fast people adopted PCs. And there's two sets of forces. There are forces that accelerate adoption. So like for AI, word of mouth is really strong, right? And the utility of the invention is also very strong. But there's also, and we like to focus on those things, but they're also forces that are slow down progress, whether it's upskilling that needs to happen, whether it's procurement cycles, whether it's how data is set up inside companies and whether that data is ready for AI to be applied to it. I mean, there's a whole host of forces that actually slow down adoption as well as in another example, like this is the first time we're dealing with a technology that is not deterministic. when there's an issue, you're not necessarily just fixing the issue. Like it like loss errors are actually built into artificial intelligence. Like that's something different that we now have to deal with to understand risk as we're deploying AI. So I think that's the first thing. I think the second, the second thing we need to consider is there's a pretty big gap between a pilot and production. Like a a pilot's easy to tell a story about. It's easy to test the technology. You can use a small user base and then you can write how successful you are and tell stories about that, which is good because you want to build momentum in organization, you want to build momentum in society. But like getting that from a pilot to production scale in an industry or even in a company, there's there's a, there's a chasm. Like it's the size of the Grand Canyon. If you ask me, you have to think about, is it going to economically perform? What's the actual business value you're going to realize? Are people going to adopt it? Is there going to be rejection cycles? Are you going to find issues after you go live? It's like there's a lot of work that goes on between those two steps.
-
10:35
Venky Ananth:
So you're saying it's going to be much slower, it's going to diffuse everywhere, but it's going to be much slower than what everybody expects.
-
10:45
Paul Hlivko:
I mean, the short answer is yes. I mean, if the expectations, if you pay attention to almost any media cycle right now, the expectations is it's going to change the world tomorrow, right, right. I mean, look at, I think it was Anthropic CEO just a couple of days ago made a statement that in five years a large portion of jobs are going to be replaced. I forget the exact ratio, but it was not inconsequential. Yes, I understand someone in that capacity has to raise money, has to generate the hype cycle, has to continue to keep investors happy. Like there's truth and that it's going to be disruptive. I don't, I'm not debating that. Even the economic professor MIT I mentioned stated it. It is going to be disruptive. how long is that going to take is really a factor of what are all of the forces that balance progress and how strong are those forces, which are the ones I mentioned previously.
-
11:40
Venky Ananth:
So you're really saying that the Silicon Valley speed and then there is an enterprise speed. So let's let's take a punt. What, what is the timeline that we're looking at? When you think that enterprises are really going to, you know, you can say we have arrived kind of a timeline, right? Because you're saying that it's going to be slower, it's going to be much messier than we expect. What's your sense on this?
-
12:08
Paul Hlivko:
I think the enterprise timeline is, is already happening now. I don't think it's a question of should an enterprise wait, the answer is no. Like I think an enterprise should already lean into the changes they need to make today in order to prepare for what use cases make sense for AI right now in an enterprise. And I'll, I can go through a few of those. But like preparing your workforce for the change, thinking through like what upskilling is going to be required for your workforce, thinking about what, what governance is required so that it actually moves quickly into your organization in a risk managed way. But when you're, when you're balancing even enterprise adoption and Silicon Valley, there's the innovators and then there's really the imitator group, right? In the imitator group, and this is true even in the models that look at diffusion, the imitator group is 80 to 90% of the market, right? Like I'm not debating that most enterprises are doing the actual initiation or actually being innovators now. I think we are too, like we're interested in an adoption and we have a whole host of use cases, probably two dozen that are active right now. So we need to do that. All enterprises need to do that. But scaling it throughout all industries in a market, making sure that it actually has a return for each individual company, but a return for society, that's what takes time. I mean, if you look at most other technical diffusions, they've taken decades of time, right? And the class of technology that AI is in is called a GPT. And I'm not talking about Transformers, I'm talking about general purpose technology. And general purpose technology is an economics term, and electricity is the best. It's probably an easy example to think about. When electricity was invented, factories didn't change tomorrow, and it wasn't electricity that changed factories. What changed factories is people going through and analyzing. All right, what are the benefits that I now have electricity? Well, I can completely redesign the factory floor. I can bring in machines that actually reorganize how people work alongside machines. That's what actually changed manufacturing. And that happened 40 years after electricity showed up.
-
14:24
Venky Ananth:
So you always say that, you know, the game is all about execution, 90% of it is execution and any transformation. So where do strategies fall apart in your experience? You know, you've gone through startups, you've gone through the finance industry, you know, in the healthcare industry. In your view and wisdom, where do this fall apart in, especially in the context of AI? Now that there is so much hype going on and like you said, everybody expecting something magical to happen tomorrow, Where are you saying this is going to take its own adoption cycle? But where are the the gotchas in terms of what you need to look for so that he doesn't fall apart?
-
15:07
Paul Hlivko:
For it to not fall apart like early on in my career, maybe I'll, I'll, I'll start with this. I was an engineer. I went to school to be an engineer. I enjoyed coding. What got me to a point to switch from writing code to leading teams is the actual change management with people like that is incredibly difficult. And just if you think about what AI is doing, it's replicating A neural network like, so it's replicating a human brain like it is more difficult of an engineering problem than building software for the Internet back in the day. It's like, what fascinates me about where execution fails is it's, it's largely people. The technology does exactly what the technology is advertised as. Now, the advertisements can also be sometimes misleading, right? So being able to understand and interpret like where the technology is at on the frontier is difficult. I think the other thing that people tend to struggle with is focusing on scientific progress over innovation. Like scientific progress is, I've had the latest breakthrough on a foundational model that's scientific progress, which is great. Like we should celebrate scientific progress all day long. But innovation is hard, and innovation is what really keeps me up and keeps me going. Wanting to come into the office, wanting to do better for our members and improve healthcare. Innovation is taking scientific progress and making it work like in a market, making it work for customers, making sure that margin actually exists when you're adopting the technology so you actually have a sustainable business, right. So I think the people side of the equation is much more challenging and frankly, it's more invigorating as well.
-
16:55
Venky Ananth:
That's a good segue to my next question, Paul, which is, you know, there's an incredible rush on this whole model building and you know, every day we seem to get a new model and everybody's obsessed about who is on the top of the leadership board, right? But then, and this is something that I also hold very close to my heart and I believe in it is the real values in the application layer, the keys to the Kingdom to unlock value for business or an enterprise is really an application layer. But just curious to know why you say that. You know, what if your perspective can share.
-
17:37
Paul Hlivko:
It sort of leads back to the difference between scientific progress and invention and innovation. But I can add on, I can add on to that a little bit. Where software meets a business process is where value is created, right? And models don't generally meet a business process like they tend to involve a person, someone in the workforce and a process and then software on top. So there's this app, the application layer that you mentioned is ultimately where I think most of the transformations going to occur. And if you look at the size of the market opportunities, it's even true when you compare like cloud infrastructure, which we've all generally been adopting and adopting and have adopted and we've been one of the leaders in our industry. But the size of the cloud computing market, which is kind of like a foundational model in AI, is much smaller than the software as a service market. So like the application layer is where most of the value is probably going to occur. And thinking about companies that show up in your consumer lives like Apple or Google, or companies that show up in the enterprise, like your work days at Microsoft's. I mean, they have a lot of the relationship with enterprises today and control a lot of the application layer. So understanding those partnerships is how a lot of organizations are going to extract value.
-
19:00
Venky Ananth:
Perfect. I love that answer because I mean, I smile because I just, you know, to me, the way I think about it is even if you go back-to-back to the 70s and 80s, it's like you're building an operating system on which you start providing applications specific to enterprises. So the operating systems are really the models for you and the applications are aware it serves your customer better, your product design better, etc. So yeah, I'm going to agree more with you.
-
19:29
Paul Hlivko:
No, I totally agree. I'll add one more thing on since you just brought it up. And I remember you mentioned these, the foundation model developers chasing these benchmarks. I find it so the benchmarks are fascinating. We all, one of the reasons why we are obsessed with ChatGPT in the early days is the number of standardized tests that it could pass. I'm sure, I'm sure you remember that which I, I was equally amazed. But look at that a little bit different, right? So who what people can actually pass those same standardized tests all the time. Well, often times it's like grad students or the top of their specific profession. How use it's like when they show up in the workforce, they've been studying, studying, studying, they pass tests, they show up in the workforce on day one. How useful is that individual in the workforce? Like when they're working in a company around business processes around other people, Most of a company is a team sport of some sort. Like it's not just passing an intelligence test. So I think we need to start thinking about foundation models and AI in a similar way. Like, so a better way of benchmarking progress on a foundation model is what percentage of economically viable tasks in a market can it perform? Like, I don't know about you, but like we're 2 1/2 years later after ChatGPT has been released. Almost like the percentage of economically viable tasks that these models can perform today, it's probably pretty low today. Like I that is definitely going to grow, but it's a different way of framing it rather than thinking about just passing a standardized test.
-
21:04
Venky Ananth:
Awesome. It's a great take on on the state of affairs. Let me switch gears. And you kind of mentioned this about talent and people. Just curious what's harder to change the systems or the mindsets in an organization?
-
21:20
Paul Hlivko:
I mean, I would say mindsets for sure. And it goes back to what I, what I shared earlier to some extent, like coding, it either works or it doesn't work. Like people, like the neural network that's arguably in their brains is much more, well, it's much more complex, right? So like we're constantly evolving. I mentioned companies are a team sport. So it's not necessarily changing a person. It's also thinking about the team or the team that operates within a broader organizational context. So I think you need to get past just thinking about like the tech, like you need to spend a lot of time thinking about the mindsets of the individuals. And one way that I've I think is effective right now is looking for early hacks. And for me, a copilot. So like use Microsoft Copilot as an example is an early hack like many people are looking at that as well. Am I getting productivity out of every individual person using Copilot? Probably not. And if whether you're using Microsoft or you've got Anthropic in your enterprise Oregon, you have ChatGPT, it's roughly the same answer. What you really need to be looking at those as they're an L&D tool right now. Like they're for learning and development because the runway is going to be so long. Reframe those investments in Copilot or reframe the access to these foundation models in the chat interface. Like reframe them as a learning and development tool for your workforce so that you can focus on mindset because it's both the mindset of the person and their individual tasks and they know their tasks best. SO shifting their mind is important. But then they're adjacent team and then also the teams surrounding their teams that you have to win over the entire stack.
-
23:08
Venky Ananth:
So I want to logically move to the next point. Paul, you, you say that you know, early hacks like Copilot are LND turning and development for for organizational talent, but you earlier kind of talked about invention and innovation. I just want to go a little deeper on that. How do you what is the first of all, what is the difference between invention and innovation? And kind of unpack that for us on what you mean in the context of AI and how we should think about it.
-
23:46
Paul Hlivko:
Well invention for me is just purely technical progress, right? So patent is a patent is an invention. As a simple example, breakthroughs on how technology is built possibly is an invention. Like innovation necessitates that you're actually thinking about the end user. Innovation necessitates that you're making sure it's economically viable. Innovation necessitates that it works in a market. Market has competition, market has customers, customers have purchasing power. So you actually need to think through like all of the factors to get it to production in a sustainable way for the long haul when you're dealing with innovation, like I think invention is incredibly important and especially as technology professionals and almost anyone in a business environment, you need to pay attention to the invention cycles because that's where all the future is going to be created from. But then you have to process it separately to understand how is that going to diffuse into your business and into your markets.
-
24:53
Venky Ananth:
So you would say Open AI is a technical invention, whereas let's say Marriott if I'm doing a hotel booking, how they designed their product where I can make they can make my life easier to book a hotel room or cancel a room or upgrade a room would be an innovation.
-
25:14
Paul Hlivko:
Yes, so that's one way of looking at it,
-
25:16
Venky Ananth:
Sure.
-
25:19
Paul Hlivko:
But let me let me even break up Open AI. So the foundational model progress invention, OK, the chat interface on top innovation. Now the question on that is how is it going to fuse into business? So there's a recent study, the Department of Economic Research spent some time understanding adoption cycles of like chat oriented solutions, whether it's open AI solution, whether it's cloud from Anthropic or even Copilot. So when you look at that type of general purpose Gen AI and how it's actually showed up in an adoption cycle, daily active users, weekly active users, monthly active users, there are a lot of users. No one's debating that. I mean, ChatGPT one is the fastest growing consumer technology today. What this research paper from the Department of Economic Research suggests though is the intensity is incredibly low. So you can have a daily active user account that's high. You can have a monthly active user account that's high. But if I use it for one minute or if I use it for 5 minutes out of my day, the intensity is incredibly low. And their research showed that. So I think one thing that we're all working through now on the adoption cycle is if the intensity, which is what their research suggested is 1.6% on average of a work day across X period of time, we're actually not probably going to extract a lot of value because it's barely being used. It's like we need to think about how that turns into something that is more ingrained in business processes so that it is understood by people as to when to adopt and when can I get these benefits by using these solutions. And that is a form of innovation. But like even open AI, you got to separate the model development from the actual consumer or enterprise product that's being built because the enterprise product goes through all sorts of purchasing cycles, adoption cycles, user feedback cycles. Like it doesn't exist absent people, it exists in context of people in business. Whereas like just making progress on a foundational model can be scientific progress alone, which is more invention.
-
27:33
Venky Ananth:
So it's just 1.6% in terms of intensity, correct?
-
27:37
Paul Hlivko:
For intensity, yes. So like we don't talk about that though.
-
27:40
Venky Ananth:
Yeah.
-
27:41
Paul Hlivko:
Like we're all like we love and I equally love these products. I have a subscription to every single product. I use them as much as I possibly can. And I'm frankly being technically inclined. I'm probably on the higher end, I would assume, right? Like if you look across a diverse set of roles and a diverse set of industries and companies, which is what this research was conducted on, it's a very low intensity right now. So back to your mindsets, like we need to work on mindsets and change management.
-
28:10
Venky Ananth:
Fascinating.
