Good Disruption: Digital Twins

The Disruption, a lively discussion between UVA Darden School of Business professors Yael Grushka Cocaine and Mike Lennox on cutting edge technologies and practices that are challenging the status quo.

Hello Yao.

Hey Mike, how are you?

I'm doing well today.

Good, good.

I'm excited to talk to you about a topic that I personally see a lot of excitement and future for.

Well, tell me, tell me more.

What are we talking about?

Have you heard the term digital twin?

I have, in fact.

I have in fact.

Have you, have you ever dreamed of having a twin?

Good question.

Uh, no, and not actually, maybe yes, but not a digital one for sure.

But, uh, yeah, that's a curious.

My wife is actually a twin, but not an identical twin.

She has a brother, you know, twin.

But yeah, I don't know.

I mean, I had 6 years between myself and my sister, so the idea of having a twin.

That could be kind of cool, be a different experience than I had growing up.

Yeah.

What do, what would you want to do?

Why would you think that having a twin would be cool, like to play tricks on your parents?

I don't know.

I mean, maybe it's like the movies, uh, Parent Trap and things like that.

Which one confuse them, which one twin they're, uh, dealing with there, um, well, so, so what's a digital?

Have you, have you heard it?

I mean, it's everywhere these days.

I think the term has become, uh, popular in, in kind of, uh, common use, and you see it everywhere.

Like, in what context have you heard about digital twin?

I think where I've seen it, um, most often for two things.

One is for industrial processes, so the idea of creating a digital version of an industrial process.

And then some of these interesting pilots for cities where they've tried to create a digital twin of, you know, city operations.

I think Singapore has one of the first out there.

Yeah, Bologna, I've heard, um, even here in this area in Virginia, you'll hear of cities that have a digital twin.

And so what is the idea?

The idea here is that we can actually mimic the reality, all the systems, all the different interactions between People, between systems, if you think about complex uh traffic patterns and the like, uh, but in the digital space, and why would that be beneficial?

Because we can experiment with, um, decision making, we can make changes, we can kind of study phenomenon without having to worry about really making any harm or damage in reality and even anticipate maybe steps ahead in order to put in place corrective actions.

And the like, yeah, and when I think about digital twins, I also think about the notion of kind of something else we've talked about in the past, IOT, so internet of things.

Most of these systems are relying on remote sensoring to provide real-time data to craft the digital twin, which might be a little different than, you know, simulations chipping around for a long time, create a simulation of some physical, uh, environment.

But this is now real-time data constantly updating, quote unquote, the simulation.

Yeah, and so, uh, you and I, we've talked about this on our show before.

We share a background related to operations research, system dynamics, system engineering, uh, where in the past maybe we've built, I remember my assignment when I was in my master's studies, putting together a hospital simulation with the events of patients coming.

Systems was the best, of course, but the shortcoming there was typically that you could only model a subset of maybe the interactions.

There was no real-time data to your point, and it was very limited in the type of problems that you might want to solve.

And so now you can run in parallel to reality.

You can really run a system that is in the, in the, in the, uh, you know, the digital ecosystem.

It's not.

Interfering with anything on the ground, but it can help you study and understand what's going on.

So just to give you a sense of uh the size of this market, this is a real thing, OK?

We are talking about um the digital twin market currently is estimated around $20 billion and it's anticipated in the next like 5 years or so to get up to $150 billion and The players here and the companies that are investing in this are the companies that you might imagine, maybe others that you wouldn't, you know, everybody from General Electric, Siemens, Microsoft, uh, uh, IBM, yeah, Cisco and the like, and probably a lot of other players like you mentioned, municipalities and more, um, uh, folks that are doing it in a local level, um, uh, in order to improve the way that they think about their environment, improve the decision making, and to allow them really to.

Test and to and to and to be thinking more creatively.

So let me be the cynic just for a second, you know, so again, the idea of like smart cities and this idea of using these digital twins to optimize processes sounds great in theory, but like, does it, does it really materialize in in enough gains that it's worthwhile to do these things?

Well, I think like anything else, um, I think there are pockets of activity that we see a lot of gains and a lot of, um.

Uh, real decision making and training and, and studying and developing new products that we could have never anticipated before.

And so it might not be across the entire segment of the ecosystem, but you do need to model all of that in order to get even pockets of improvements and pockets of benefits.

So for instance, if you can imagine, and we'll start something small, if you can imagine um a large hospital system and you can design a digital twin for the hospital system.

Are there efficiency and uh uh um improvements that you can kind of derive at a hospital level potentially but the real benefits might be even at a much smaller and a more myopic level in a specific operating room with a specific physician allowing those physicians, allowing those residents to train and to practice and to simulate the environment and to be to have a much richer environment to learn and to, you know, run through a dry run.

Or digital run of a specific operation and then check where there's going to be some issues or maybe where they run into some pitfalls and then maybe correct course and improve.

Yes, we're going to see huge improvements in the, in, in the future.

Now, how would we separate out this concept from something we've talked about in the past, the metaverse?

I was about to say, is this, is this relate at all to the metaverse?

Could this be put into the metaverse?

Of course it could be related and like, you know, there's, it's not totally.

Dis, uh, disconnected to stuff that we've talked about in terms of AI and generative AI, these things, these technologies bleed into each other and help improve, um, all of them across the board.

The digital trend really from, uh, my understanding and the implement, you know, cases that I've seen and the implementations that I've seen is really relying on this notion of real-time data.

So the real benefit is that you're mimicking reality.

You're not creating a parallel reality.

You're mimicking reality with all of its complexities, with all of the different factors that intermingle, and that you're really trying to capture the pacing and the nature of the world that you live in, in order to make here in our world, better decisions, but you're, you're kind of learning from, from the digital twin.

All right, does my avatar in the digital twin world know that it's in a simulated world, or does it think it's in the real world?

Yeah, well, here you're more like simulation.

You know that.

In your in simulated world, you have agents that behave like you, that think like you, that mimic you, but it's allowing us to understand really broader, broader patterns here.

Can you, like, let's think of other environments.

So for instance, if you're a sailing or if you're an Olympic sport or any kind of sports team and you're trying to get new strategies or you're trying to play out new, new play, uh, uh, moves, um, a digital twin could be an environment that you could test things out.

Oh, I like that idea.

So you could take like, uh, the, the simulations we have for football or basketball, but then run a digital twin based on like the biometric data, of course, and imagine what's going on.

Like imagine something as high stakes as the Super Bowl.

Imagine you have a digital twin running in the background and that you can try out various moves there with the same players and the and the fatigue that they're recording and the conditions on the ground and then you can kind of take.

That as an opportunity to learn from that and be inspired for changes that you want to make on the field.

I like it.

I like it, yeah.

Uh, what would be some downsides or some concerns?

You, you, you said you were going to be the, you know, the skeptic.

I know, well, I, I, it's hard to see in this particular case, but I guess again, you know, there's always a Black Mirror episode, uh, for this, right?

Um, and, and again, I guess this is the, the blurring of reality with, uh, the, the simulated environment and does that create, you know, downsides of, of how people.

View the world or how they engage in the world.

And like anything else, I mean, some of the, the challenges that we can anticipate if you're thinking about digital twins from anything from um as you mentioned, a manufacturing plan for let's say, um, you know, uh, in Ford's case, if they're thinking about the products that they manufacture there to something like a supermarket where you have customers walking around and interacting with each other and with the staff, there could be some.

Downsides if we um have certain biases in these systems or if the simulation helps reinforce some negativities that then come out and play out, um, we want to make sure that these, the digital twin is also used responsibly with the same ethical frameworks and you're not totally, um, you know, uh, it's not as if all of that automatically falls out.

You have to ensure that the system is set up in a way that. ethical decision making, responsible, uh, responsible decision making, and the like.

Yeah, it's interesting to think about like, OK, in the, in the world without digital twins, how would we deal with some of these things?

Well, some of it would be like feet on the ground, walking the actual, you know, manufacturing facility and seeing if there's issues or like on the floor, you know, do we become overly reliant on the simulation.

In that world, I think about something high stakes like a nuclear power plant, um, and where again, making a digital twin makes a lot of sense, but do we then become overly reliant on, oh, everything seems fine because my digital twin is telling me everything's fine, but then I don't know.

The pipe, the water pipe is leaking over here and we're gonna have overheating, uh, and we just missed it because it isn't censored by the, um, the information flowing in.

And like anything, if you are a per a people's person and you believe that you need to look somebody in the eye to see that they're feeling OK, behaving OK, able to perform OK, as opposed to watching just, you know, the monitor and seeing what it, what technically the numbers look like, they might, the numbers might look like, oh, they're fine.

Fine, but if you're actually interacting with a real person, you might come to a different conclusion.

Yeah.

Are we at this point where we need some help?

Yeah, maybe we can talk to an expert.

So it so happens that one of our colleagues is actually an expert.

We are so fortunate to have, uh, Sam, uh, Sam Levy, who's gonna join us today.

He's, uh, an assistant professor here at the Darden School, and he is, uh, focused in the areas of marketing.

So his digital twin expertise comes from the area of marketing.

His research in Investigates um customer analytics in different contexts such as privacy, data fusion, and prosocial behavior, and he holds a PhD in marketing from Carnegie Mellon.

His dissertation introduced the novel concept of a digital marketing twin.

OK, so that's a specific application where he finds path of least resistance to individual level customer satisfaction.

Sam, welcome, welcome to our podcast.

Hi.

Hi Mike.

It's a pleasure to be great to have you, Sam.

So Sam, maybe you could start off by just telling us what is a digital marketing twin.

So sure, a digital marketing twin in my context is a virtual representation of a customer.

Right, so the idea is to create what if scenarios at the individual level.

So say you have a customer and you're, say a telecom company and you are running a bunch of surveys and those surveys are very expensive, you can only run them once per customer because those surveys are very long to fill and you have to pay customers and they they just don't come back.

So and.

The idea would be how do you create that virtual twin of customer where you could basically simulate different scenarios, say if you're the telecom company, change the pricing level or the network speed.

Would that customer switch to a different carrier, for example?

That would be the idea of a customer twin here.

Beautiful.

So it's like a scenario, a more sophisticated scenario analysis where we can experiment with potential futures and see how it plays out in our digital twin environment.

Is that the idea?

Something like that, yes, uh, digital twins, um.

Of customers are a very particular type of twins, but we could think of the twins of, for example, stores, right, in the context of in the retail, right?

And the idea would be, oh, how do you create the twin of a store here, um, you know, would you start with the shelves, would you start with the trajectories of customers?

There's a lot of questions here going on.

I can imagine, you know, when it comes to physical assets, it's easy to think about how we might simulate them in one of these environments.

And again, use, you know, uh, internet of things type of sensing to provide us information, maybe even like what's on the shelf and the like.

But key to what you were just describing is you have to have a model of the consumer, of the individual.

Like, how do you do that?

How do you simulate that individual and how They might behave in different scenarios.

So it's quite difficult actually.

So in the context of the store, you would have to predict customers moving in the stores and so people would quickly go out of patterns if you're using traditional machine learning techniques.

So instead you would have to use more advanced algorithms, more advanced algorithms such as reinforcement learning to be able to accommodate how people.

Go out of pattern where for example, you change the store layout, you would need to have a very robust model of people, how they move in the store, so that's quite complex.

The way I treat it in my research, I use a generative model, so I use a type of generative AI algorithm to be able to predict how people behave without the physical context at all.

And so this is where it departs from maybe what you and I studied, where it was more simulation where we'd.

Put in a distribution.

It was kind of lockstep, you know, off the shelf, and all we would do is kind of walk in and go to the front of the line exactly following up croissant distribution or the like.

It immediately came to my mind is I, I am not a great shopper.

My wife will confirm this, and, uh, she calls me the lighthouse because we will go to like a, a clothing store, and I will stand there and she will shop and I will like look.

Back and forth like a lighthouse there.

So, Sam, if you don't have the lighthouse in your model, like I, I can provide you the behavioral attributes of the, the lighthouse consumer.

But it makes me think, Sam, so are we seeing, um, so for instance, are we gonna see schools like, let's say Darden or UVA?

Should we have a digital twin thinking about the students of the future and how we should interact with them?

Well, it could be a good idea to understand.

How, how students would react in this virtual reality, I, I would see a, a couple of challenges.

Um, I would say the first thing to do would be to start small.

Maybe we shouldn't start with the school, maybe we should start at the classroom level, for example, right, because, you know, creating twins of cities.

So creating twins of complete virtual assembly lines are extremely difficult.

I, I, there are lots of costs involved with building the twins, gathering high quality data, for example.

So where would we gather the data?

Do we have consent to get that data as well, right?

It's an important question.

And finally, how to use different teams, right, different teams have different type of data, right?

Maybe um you would have to talk to a different team who has a data set and so the, the, the problems with many organizations is they work in silos and so you cannot really access that data in an, in an easy fashion, right?

So there's actually a lot of, a lot of problems here associated with building the twins, it's a, it's a lot of challenges to build them.

But I can see why we're talking about this disruption today in this day and age where more data, more higher quality data is being collected, so we're in a new era on that frontier.

As you mentioned, Sam, the algorithms have evolved.

We've talked about that in our, in our show, and so we're seeing that the data is meeting the algorithmic development that allows us and the sensoring that you mentioned, it's all coming together to enable these digital twins while.

Still costly and complex, they're becoming more possible than ever before because all of the combination of these kind of technologies, yeah.

I'm, I'm curious, Sam, are we seeing at all the kind of self-generation of these digital twins?

And what I have in mind here is, again, yeah, and I come from a generation where if you were gonna program this, you would, you'd be hard coding basically the behavior of the agents and the like, um.

Could we basically teach the simulation to teach itself how to create these environments or to create these, these uh different agents that are operating in these environments?

That's a good question.

I would say it's probably difficult right now to make these agents teach themselves.

I don't think we're there yet.

Uh, there are a lot of companies right now that take care of the heavy lifting, right?

So if you want to build a digital twin that even like teaches itself, you would, for example, go to Nvidia, you would use their omniverse platform or you would go to AWS, Amazon Web Services, and they have an Internet of Things twin maker system, but you wouldn't build it from scratch, so you would rely on third parties.

Techniques and products so that you don't start from zero and then To the question of being able to teach itself, I, I don't think we're there yet.

So reinforcement learning is very data hungry and so we would need tons and tons of data to reach that level of capability.

Let me ask you, we, we, you brought up briefly before, um, this ethical question around, uh, with consent, so collecting this data and capturing it with consent.

And, you know, we heard of applications as an examples in hospitals or training physicians where these digital twins are being used.

Are, are there ethical issues that are emerging?

Are there concerns that there are some uh use cases of digital twins which are not totally legit or that folks feeling could, could go in the wrong direction?

It could go both ways, I would say, because you could also say, you could also use flip the argument and say, look, the twin is built to actually prevent ethical.

Misconduct, right?

You could create the twin of a customer, for example, to be able to, uh, to run an experiment in an ethical way without actually harming, right?

The idea of a twin is also to live in a risk-free environment, right?

So now the question of collecting data with consent.

Um, it's a different question.

There's now a lot of things going on with, uh, privacy laws changing and the, and the, the end of third party cookies in, in, in many settings.

So companies are more and more careful now in, in giving data to others, but I think within the company.

I think there are still processes that could, you know, you can still create twins within the company, you just have to be able, I think the the the challenges are more organizational rather than ethical right now because you just need to work with so many different teams.

You know, one thing that comes to mind is that some companies are using, in essence, digital twins to train algorithms for the real world.

And what comes to mind is, um, uh, Waimo of Google fame who has been using virtual simulations of some of the cities they operate in to actually Train their autonomous vehicles that they're running on the road.

And I, I'd be curious, Sam or maybe AL, like what, what do we think about that idea that it could actually work the other way where the digital twin is now in essence training real world actors in this case robots, but Training them, I mean, I think to Sam's point that it's encouraging to hear if it can prevent any casualties or catastrophic outcomes in the real world, then why not use the digital environment to practice, right, like that, to the degree that the digital environment is close enough to reality.

I don't know, but again, this is taking it and bringing it into reality.

That's the, that's, I think the interesting thing.

Yeah.

Another thing is you have to keep these sensors and, and you know, you have to.

Have you have a lot of upkeeping here, you need to constantly monitor if the sensors are delivering the right data.

So you're right that there could be ethical issues related to the quality of the sensors.

Are they actually built, are they actually faithful to um to what uh what we would expect from them.

I can imagine.

So I spoke to a few folks who um are dealing with digital twins at a government or municipality level, and we mentioned these earlier like cities that are using digital twins locally.

Um, and it's interesting because I can see a, a desire to be more efficient and to be forward-looking.

So, you know, if you monitor uh births in the NICU or in the, in the hospital, and then you're already play that out several years and you think about the distribution. of your classes in the schools and you know, the different, uh, you know, number of students that you can facilitate in each neighborhood like could we get ahead of the game and optimize some of our planning, our urban planning around these using these digital twins, but then you have to go back and adjust it based on the decisions that you make.

So Sam, are we seeing these digital twins play out, so they're used, decisions are made, and then we go back to the digital twin and update it and continue to refine it so it continues to reflect reality.

Yeah, yeah, exactly, the, the, the digital twins should be continuously updated based on the decisions.

We see a a rise of also more probabilistic twins where the idea is that you don't just didn't generate one outcome, you generate a range of outcome and so you can make a better decision based on the worst case scenario versus the best case scenario.

So that, you know, that aspect of decision making is I, I, I think it's very important if you take the example of healthcare where you are now, we are now building digital twins of hearts, for example, right?

Hearts is a, is not a physical process.

There's a lot of noise.

You know, uncertainty is associated with the human body.

So that's why we actually need more to incorporate uncertainty in the digital twin models.

So I would say these probabilistic twins would help make better decisions, especially in healthcare.

See, there, there's always a Black Mirror episode, and it reminds me there is one about dating.

I don't know if you've ever seen this one where you don't know this.

I'm sorry for those who may not seen this episode before, but, um, what, what comes to light is that when people are dating through an app, they create two digital twins of the two, the couple, and they basically run a simulation multiple times to see what the outcomes are.

To then make a better prediction of whether you are compatible with the person that you just met, you know, at, at a bar or whatever to see your future kids in the future, don't you get along with your mother-in-laws, cut out the whole simulation, look at the, you know, probabilistically what the life outcomes could be, and then kind of work back to say, yeah, this is a good match.

You're gonna have a good life or, or not.

Yeah, um, that's interesting.

I can see that that would be a use case with, uh, with the dating apps being what they are these days.

Sam, we often ask this question as Mike and I try to get our head around, uh, whether this is a good disruption or a bad disruption or no disruption at all.

We're not gonna put you on the spot because you are an expert on this, but what, what does keep you You up at night?

Like, what concerns do you have as the use of digital twins continues to grow in various contexts?

What, what do you worry about?

I would say I worry about the fact that we might become over reliant on it if the technology keeps growing the way it is.

Um, we might completely overlook uncertainty associated with the outcomes because it's so easy to overlook, overlook at them, so.

Yeah, that's, that's what keeps me up at night.

Also the idea of digital twins of customers and stores at the same time, so full, the full thing is actually very hard to do, just modeling trajectories of customers and at the same time having.

Modeling the store can be extremely difficult just because you know you have to mix the laws of psychology and economics to actually the physical laws of a store, which probably very hard to do right now, so I would say, you know, starting small.

Is always a good idea with digital twins for first modeling the the the the the shelf and then going more, uh, you know, scale it up.

So I think the idea of scaling keeps me up at night is probably the most difficult part.

Starting small is easy and I I would say many companies and researchers can do that, but getting, getting it big and getting it reliable is very difficult.

Yeah, and in such a way that we don't over rely on it to the point where we forget that it's just a digital twin and not reality, right?

Um, yeah, fascinating.

Yeah, yeah.

And you see what is the use case that you're most excited about, Mike?

I don't know why my mind does go to like the, the smart city model like that idea of like we could, we could maybe it's a systems engineerer maybe, but like could we really optimize traffic and all these other processes to make cities more efficient that that has some just intuitive appeal to me.

Well, I know that there's a lot of interest around manufacturing and digital twins of manufacturing, um, lines in order to anticipate, of course, efficiency, but also anticipate new use cases.

To experiment with uh new products and new capabilities that maybe have not been done before and so the digital twins can provide some, some clarity on that, um, and they're much, much, much cheaper than actually setting up a prototype, uh, in reality.

Do we, do we think, and maybe it's a question for you, Sam, do we think that digital twins can push innovation?

Like, do, are they able not only to help us monitor and, and create efficiencies, but could it actually help us innovate in ways that we might otherwise?

I would say so.

It's it's mixing so many different aspects of technology nowadays, so AR VR accelerated computing through, you know, graphical processing units, accelerating innovation between teams, right?

Innovation coming up from from teams, human innovation.

Um, I would say yes, it, it's definitely robotics, right?

It's a, it's a big one too.

So for sure it will increase innovation over the years just because of the large array of fields that it's touching.

Including the metaverse, right, this is where the metaverse comes in.

Maybe, maybe.

Well, help us out here as we wrap up to, so that our, our listeners can get, um, a, a, kind of close their eyes and imagine it.

When we say a digital twin, are we looking at a screen and just seeing numbers and vectors of kind of results, or they're actual, right?

That's what I'm, right, that's one scenario or when we.

Think about a digital twin, are we actually looking at like a, a virtual reality and almost like, yeah, like you said, the metaverse, like a picture, an image that is realistic.

Like when you talk about your marketing, uh, uh, digital marketing twin, is there, are there actual customers that you see and they're like drawn on the screen, or are you talking about like just numbers that is an output on the screen?

So in my research, I visualize them with perceptual maps.

So it's, it's more abstract, but I've, I'm, I've seen research where it's trajectories of customers and you see them moving in the screen, so it gets more physical.

And of course, visual, so in context of like the, of the heart, it's very visual, right?

Um, in the context of, um, assembly lines and factories, it is also very visual.

You would see the assembly line in the screen, of course, you would see the sensors delivering the data in real time.

So there is a very visual aspect if you want, if you want it, if it's relevant for your content.

Fantastic and I've been in operations rooms and in airports and we all know with recent um events and tragedies that we've seen around us that this is, you know, all the more needed, right?

Like to the degree that it will prevent even one accident in the future then we should, you know, try to embrace it and make it as good as we can make it.

Um, well, Sam, thank you so much for being here.

I'm sure that folks will, uh, uh, be excited and also maybe curious as to how their data and behavior gets, uh, modeled in a digital, uh, marketing twin, um, but, um, thank you for sharing with us your knowledge and your expertise.

Thank you, Sam.

Thank you so much.

So Mike, is this a disruption?

Who are we disrupting?

So it's funny you went there because that's exactly where my mind's been going.

I, I, you know, we, we've had so many where we go to the, the negative and the, the downsides.

I, I think the downsides seem fairly limited to me here.

So I'm, I'm gonna rule out bad disruption, OK, but is it a disruption at all?

Is it a disruption at all?

And clearly a positive addition.

Like I've no doubt in my mind that this is going to help us in innumerable ways advance.

You know, society, human, humankind, if you, um, but if we take disruption, you know, in the technical sense that we, we tend to use it here, where is this going to restructure markets?

Is this going to lead to competitive reordering because of the advent of this technology?

Is it gonna change labor forces dramatically?

Yeah, exactly, all of those things.

It strikes me that maybe the technology itself will not.

Now, it has these secondary impacts that we've talked to from these other related technologies that could have those impacts, but I'm not sure just the nature of digital twin itself is, it's a useful tool to help us improve process flows and the like.

But I'm not sure it's gonna lead to the disruption in that, in that technical sense that we use it.

Um, let me give you one or two, contexts where it might, um, so can you maybe, uh, just to push on that disrupt how we teach again thinking about medical students, you know, they have to work 16 hours days in the hospitals and do the rotations.

Will it disrupts that kind of industry around instruction and education.

Uh, will it disrupt how we practice for various military operations?

Could there be something there?

Well, again, I think it would definitely be a learning tool that would maybe improve the outcomes in any of these applications, but it seems to me it needs to be married with something else on the back end to have that type of disruption.

So, Sam mentioned robotics that I could start to see that are we using the digital twins again to train real-world artifacts like robots and the like to do work.

Alright, then I would say that could be quite disruptive, but I'm not sure the digital twin in and of itself.

Go back to my Waymo example, right?

Using a simulated environment to train the AV, that's really interesting and value add, but It's the actual AV that's that's causing the disruption here, not necessarily the digital twin aspect.

And maybe I'm splitting hairs here.

No, I, I get it.

I get it.

So in and of itself, digital twins maybe are your ruling or our ruling in this show is no disruption.

I'm gonna go with that.

I, I rarely go.

No disruption.

You know, I'm all about disruption, but unless it's married with, for instance, a very exceptional advancement related to, uh, virtual reality, the metaverse, or something that takes us to the next level where suddenly people are living their lives and making their decisions and, and.

And really investing in the digital world or the virtual world in and of itself.

OK.

Well, I've learned something like always from you and from your perspective.

And of course from Sam uh today.

So, um, well, again, you know, always a fascinating set of topics here.

Uh, thank you as always to our producer, Gary and to our head researcher, Becky.

This is good Disruption.

Good Disruption is a podcast from the University of Virginia Darden School of Business.