Duration 30:28
16+
Play
Video

Computer vision, Data Analytics, Third Party data By Jaap - de Vries, PhD, FM Global

Jaap De Vries
Phd at FM Global
  • Video
  • Table of contents
  • Video
Request Q&A
Video
Computer vision, Data Analytics, Third Party data By Jaap - de Vries, PhD, FM Global
Available
In cart
Free
Free
Free
Free
Free
Free
Add to favorites
475
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About speaker

Jaap De Vries
Phd at FM Global

Dr. Jaap de Vries is a technical expert in FM Global’s Innovation team, based in Providence, Rhode Island, USA. In this role, he discovers and develops new opportunities to guide the strategic planning efforts of FM Global and its mutual members. He is responsible for building a network of sources to identify, research and assess emerging signals of change in the market, new technologies and scientific advances. Prior to his current role, Dr. de Vries served as lead research scientist at FM Global’s offices in Norwood, Massachusetts, USA. In this position, he established the Advanced Technology Laboratory and studied new, potentially disruptive technologies, such as machine learning/AI, augmented/virtual reality, drones and rapid prototyping. He also led large-scale fire testing programs in relation to radiation-activated sprinklers, lithium-ion batteries and SMART sprinklers—the findings of which resulted in new, patented sprinkler technology. He received his doctorate in mechanical engineering from Texas A&M University (Texas, USA); a master of science in aerospace engineering from the University of Central Florida (Florida, USA); and a bachelor of science in aeronautical engineering from the College of Amsterdam (Amsterdam, the Netherlands). He has authored more than 35 journal and conference papers, and is a member of the National Fire Protection Association, the Society of Fire Protection Engineers and The Combustion Institute.

View the profile

About the talk


03:36 AI expectations

06:50 Business as a part of an AI-driven world

09:45 Combined ratio

14:40 The majority of losses are preventable

20:31 Competing in the age of AI

22:20 The importance of loss prevention

24:45 Fire test

Share

And I'm going to be fairly modest here in the sense that I'm going to talk about the lessons that I've learned. So I'm not going to come up with rules or what other people should do. I'm just going to really talk and high-level overview of the lessons that I've learned on. Let me just start off. But what a year is Ben. I'm not sure if anybody remembers their New Year's resolution at the beginning of 2020 and how little of that it may come to fruition. I had a big trip booked you on to a belay that didn't come through with the family with one good part though. I was able to watch on Netflix

series. Excellent one on leadership princess the last dance with Michael Jordan and talking about his experience when he was leading the Bulls to six championships. I know it's so inspired by me to buy myself a pair of Air Jordans. I always wanted those kids. I finally have the money to buy them and I got them thinking that I would look super-cool. Impact of looking super cool. I was walking the dog the other day. My my wife is telling me that you look so clunky with these shoes on you actually look more like like Frankenstein and I'm was thinking upset by that

nothing much because I was cold Frankenstein with Frankenstein is not the monster Victor Frankenstein was a scientist that was putting disjointed systems together to try to create artificial intelligence and I thought well want an apt description for someone that works for a large Legacy organization. And of course, you know in engineerin we're not trying to create living on creatures, but the decisions that we make with regards to artificial intelligence do affect lives. And what I want to talk to you about today is that they can actually impact

provide power over life and death turn into Show me about my background anybody can reach out to me at LinkedIn if they want to message me as where I'm available and just what during introduction. That was made clear. My background is not in an underwriting or Insurance actually nor is my background in a I so here. I am giving a talk about AI in insurance. I wear my background is actually in aerospace engineering and particularly on combustion. My most of my academic a part of

my research. I spend on making things burn faster. I was adding aluminum nanoparticles to solid rocket propellants to see if we can have those burn faster and a stable conditions and later. I was adding seiling to on the gaseous fuel to see if we can accomplish supersonic combustion in Instagram Jets. And in my professional research career as a research scientist rapping Global, I went the other way and I have to look and see if I can make things blow burn slower and right here. You see a little video of one of my claims to fame which was Define

protection fire protection for distilled Spirits or Whiskey In this case and and what a better & Noble cost and making sure it at the single malt whiskey is well protected within the next 15 to 20 years. But in any case we want to talk to you about today is about AI right and I want to start with what Cai expectations are and then through the remainder of the talk. I'm going to talk about what that means for commercial property Risk insurance. So where would like to start is with the hype cycle? Everybody's probably familiar with the hype cycle.

It's been reported back by Gardner. It really starts with an innovation trigger some kind of Technology people get very excited about that until it reaches some peak of inflated expectations. I'm part of an innovation team now, so I see all this inflated Technologies come towards me and I'm people figure out that it can't fly do what they were hoping it to do and then it slides down into the trough of disillusionment and only to then eventually gained some popularity back and end up on the flip side of productivity and I love this hype cycle because it's really a good mental

model to think about All these technologies that go and because anything that's happening in AI is moving so fast, you can just put your hype Cycles year-to-year next to each other and see each of these technology moves on its way by look at Gardner's 2020 presentation of the hype cycle regarding aai and I just selected few items there. You can really see that the gpus are well within the a plateau of off of productivity and we all know this they're being used quite ubiquitously interesting ly according to Gardner deep learning machine learning chatbots are kind of on their

way down and autonomous cars and I got to assume that this level 4 autonomous cars are really deep in the trough of disillusionment right now. I think we're just realizing that were a little bit further away from fully autonomous door to door cars, then maybe we were thinking 5 6 years ago. But I'm very interested on the left side of the graph where we see things on the rise. That's smart robots decision intelligence and augmented intelligence in small data that are very important for us as a large commercial property Risk insurance Prince of augmented intelligence is an

alternative conceptualization of artificial intelligence that focuses on AI assistant role in this really emphasizing the fact that cognitive technology is designed to enhance human intelligence rather than replace it. And if so important when you were in the complex business to buy environment and related to that is deficient intelligence right where we not just looking at data science. We're combining it with decision Theory and managerial science into a framework that allows us to make the best business decision and also small day that we don't have A million pictures of dogs

and cats right and most problems that we see in our business since they don't present themselves at the dock at kind of problem. So we have to come up with ways that with less data. We can still create and provide a diligent decision making turn Ellie's me to one interesting book that I read by a Korean Connie and Marco Island City is competing in the age of a Ai and it's a great book. I highly recommended but the main takeaway for me from reading this book was ever competing and Essa title said we're competing in the age of a I sold. My first lesson was that

businesses are becoming increasingly Pard openai driven world. So it's not so much you asking yourself all the time. What can a I do for me and my organization and can I replace a t nutty Avenue in with a chatbot in and how much I meant hours does it save me? It's really thinking about it in a much broader sense write the questions. Like how does AI impact your supplier in the supply chain how to say I am packed the competitors or how does AI impact the clients in the consumer behavior in general? All these things need to be taken into account because we're living

in the highly connected world. And it's not just about us and what happens within our four walls. At the time when we talkin about the insurance business model, I'm going to talk about AI in Insurance a little bit at a very high level but what I really want to start with this is presenting the insurance business model and translating that end in type of objective functions that we we can work with when we dealing with a i and I'm going to keep it very simple in and use very little equations during this talk show. Don't worry. I want to just do a thought experiment. Just

imagine that we are working for an insurance company. Right and let's just pick something that we very intuitively familiar with something like car insurance, which is great because brass is an insurance company, which we imagine we are at this point car insurance mandatory so much better in this business model than to sell something that's mandatory for anybody to drive the car. So right now we have oldest premium coming in from all these car insurance policies. So right now it looks pretty good. Right. We got money going in. Unfortunately, some of the drivers are not that

grades and some of them are great. But just hit a patch of black ice right to can't always predict predict who the good and the bad drivers on but in any case some of these cars are going to record losses. So that's money out of our company. And there's another way that money is being spent and then there's the expenses of running the operations, right? So we're going to have to adjust these losses. You going to have to have the financial management of Distributing who gets what you going to have to spend money in marketing and all these running the the businesses and renting the office

space and so forth. So here you have it. This is this is the business model for any insurance company and this is about the thing that insurance companies have in common because as you will notice later on saying that you work for an insurance company doesn't mean a whole lot. It's like saying that you work for a transportation company that could mean you work for SpaceX where they keep me in your Uber driver or anything in between right with a general business model for any insurance companies based on this. So I would you turn this into an objective function or one metrics that you

can measure how successful you are as an insurance company, very simple. You just take your expenses and your losses you put that in the numerator and you take your premium and you put that in the denominator. And since this is a business talk. This will be the only equation I'll be using this is what we refer to as the combined ratio is an insurance company and it's a really easy way to evaluate the performance because of the numbers less than one that means you're running a profit and you do well. If the number gets above one, that means that you're actually having more losses and more

expenses than you bringing in in premium, and that smell so good. What impressionist we have three levers here that we can play around with if we want to do well as a company, right? We can reduce the extensive and you can see this happening with some insurance companies like lemonade where they try to fully automate some of the processes and takes as many people out of the loop as possible and really keep those expenses as low as possible. You can raise your premium. That's one way to bring in more money in and bring the combined ratio down. The problem is that you very hand tied through

Market condition is a very inelastic feature here. In fact that if you raise your prices you going to price yourself out of the market very quickly. The third and final ever that we can pull on is to reduce losses, right so we can reduce lost for a client so we can just take on clients that are less prone to losses or weekend at once. We have the client we can make sure through engineering and what not and I'll talk about that in a second how to reduce those losses. A good way to think about it is another book predicting machines is really

an economically way of looking at this so Avi Goldfarb Ajay agrawal and Josh you again solo working at the University of Toronto really take an economist view of a i and then one quote from AJ in particular. I like is that the rights of a I can be seen as a drop in the cost of prediction. So if you look at those three levers and I just talked about where the AI comes in from an insurance company, that's really thinking about how can it help us make better predictions. How can a male cat make us better predictions are valid whether a driver for to have an accent. How is he

going to make us better prediction about the future Market condition at Saturn? Is really an economic rule so we can really apply that to our basic mental model and the combined ratio here and just translate everything into how a I can help. So how can I help is reduce expenses. How can AI help us get more premium, how can AI reduce claims or pay them out faster? He's a very interesting questions from the insurance industry and I don't think that since I just mentioned the industry the insurance industry is so diverse. There's no single answer. But of course Mackenzie in 2018 did

make an attempt to stay that what they thought by 2030 would be the main influences of a eye on the insurance vertical and they basically put it in three buckets one is the distribution to how is Insurance old and how is the interaction with a client taking place? The next one is pricing. So can you you buy used as a I very accurately price to risk some cold at underwriting with a scalpel. Obviously the consequences of the actors that the margins are going to get razor thin at some point at some point you going to squeeze all the efficiency out of

that which modern smart of algorithm and finally it's a I used in claims. We're in the case of laminate. They they they pay out their claims in 3 seconds, which is great. This is all great. If you insure a renter's insurance, if you insure a car or somebody's bicycle, but that's not what FM Global does FM Global is the monoline large commercial property Risk insurance. So how does everything that I just said apply to one of the most prestigious technical institutes like a mighty why did you end an insurer MIT? How does any of this apply right or one of you

insured a place where they don't have Engineers but imaginary Ryan and they're building are not like anything else you see in the world. How do you price Terrace? Right how you evaluate that how you going to apply a I do that or how are we going to deal with the fact that warehouses are now full of robots and they're bigger and taller than ever seen. How about how we going to apply to protection that we've developed for the past few decades and Central Fire Protection. Okay, so just doesn't really is something where a I can easily be applied. Listen to

FM Global the company. I work for a large commercial property insurance. And we basically live on one motto for nearly two hundred years. And that is the majority of losses are preventable. The majority of losses preventable was really means that we applied engineering services to our clients. Right? We don't use any Actuarial signs showing this in of course, I want to brag about the company I work for but I wanted to have this as a full understanding of white applications of a I would be harder for us and it would be for some minor insurance companies that do

car insurance princes, but we make our engineering recommendation based on hard science and proprietary data and another next to Ariel Foreman. That's why we have the the scientist over two and a scientist working for a company and over 1,900 Engineers all over the world is fasting risk. I just need to find partnership where Mutual company so we don't have third-party shareholders were owned by our policyholders that are part of the decision-making and we really have that partnership where we do business risk Consulting. We have a global delivery so we can follow up flying all

over the world because we're monoline. We only do property Risk insurance. We really keep that specialty focus and because we're owned by our policyholders, we have a very fair claims. In fact the claims payout procedure, that's about fnglobal. How are unique the main thing you want to remember from this is that we only do large commercial property risk. We insure about one out of three of the fortune 1000 and for each of these clients, we have an individual risk assessment that we do at all with this was a place in mind that

the majority of losses is preventable. And what that really means for our clients that resilience is a choice and then we can talk later about where II comes in, but it really starts with the fact that resilience is a choice. Relation to Troy's used to be defined by how fast you can overcome obstacles that the world would throw at you but why would I do that now the complexities in the distance as you run and industries you operate in the world, you're living are redefining resilience. They making the job more in certain and more demanding. So being resilient is going to take much more

effort now that they're going to pay off in RPM because it provides a competitive Advantage if you're the only one that's standing after major flat flood or a major storm. Then you going to have a major competitive advantage and you can gain Market. If you are a chemical plant you do not want to make the news because there's some green clouds passing over school that comes because your factory is on fire, right? That is your brand that's on the line. So FM Global will replace that factory, but if we didn't apply Sound Engineering you would still lose be at risk of losing their brands and

that reputation. I want to bring it back a little bit to worst AI again and the difference between business-to-consumer and business-to-business and really looking at it from a data perspective. Okay. So this is to Consumer just imagine the car insurance that I talked to Dad and a business-to-business imagine FM Global ensuring MIT or Disney. What is the difference in data? Well, you can look at the complexity of the day if you think about insuring a car how much about a car do you need to know maybe 20 features 25 features were so how much data do you have

Ubiquiti up-to-date? Oh, well, if your car insurance not uncommon to ensure 10, 10 million or so and have 10 million policies just in the United States alone. So funny you bake with your data plan available Global. We only have a few thousand clients in each client's like a research campus or a large University can only be described as thousands and thousands of features nights are very complex data and small day that we can actually come out. Who died in a ratio in which I call Dai opportunity ratio, which I have the ubiquity of the data / the complexity of the data and

just to put that into a thought experiment. Let's get back to our car insurance example, and if you have a large car insurance, maybe with 10,000 a 10000000 policies and each policy is pretty well-described a 25 features or so. If I won by the hour, you got me going to happen AI opportunity ratio on the order of 10 to the 5th for FM Global. We have a few thousand clients. Each of these clients can literally be described a thousand feet are AI opportunity ratio from a data and consumer incline perspective is more around Unity, right? And since we do some Dimension a dimension reduction

here, we can put it in a quadrant and we can put it in plot that against other features like digital-native furnace and this is an interesting a quadrant year because When you work for a complex, well stablished organization and you listen to a lot of the the Consultants Business Consultants in your beat articles that constantly pouting about people that are the leaders in AI in the championship Ai and similar names come up and I think we're all familiar with them right to recognize. We're in this quadrant these companies sit, right? So is it really fair? If a business

business consultant comes to me and say why why aren't you more like Amazon right or why aren't you more like and financial or Uber? We're running a completely different business and I think he needs to be completely different. That's well. Lemonade is interesting here super digitally Native Pride only found it a few years ago, really everything is automated and now that they're building more and more customers are actually on their way up in the AI opportunity because now they have more clients or policyholders in more data to apply there a island.

So similarly and again getting back to competing in the age of my eye by Karim lakhani in Marco Island City Day date to come up with the NCD. They came up with a similar metric and that's really the value that you can generate versus. The number of users super traditional operating model that you can generate good value. Even if you have limited number of users, but they need to have a digital operating model which in this case is another way of saying of being able to apply AI for your business. It really goes up exponentially after a certain point of the number of users. So if you work

for a business-to-business company, or you have a limited set of plans, you may really want to ask you shall where am I on this graph? And where can I really leverage it? Really about becoming knowledgeable. That's where my lesson 2 comes from that I learned so become knowledgeable about what a i Solutions exist today how to evaluate them and which ones are expected to exist tomorrow. So don't just rely on with other people tell you how people become an expert in your own company because most of the field is only three for five years old and the experts that existed 10 years ago are

no longer experts unless they keep learning and every organization should have people in my opinion that keep learning about these things. It's another thing focus on the IMEI. What is the intelligence? What makes your business unique? Right and I am in our case in FM Global is his engineering and science based loss prevention and don't lose that. You don't want to replace that with AI you want to keep that but if you use a iu1 enhance it and I just want to spend a few minutes with a thought experiment just as an example of the importance of just intuitive understanding

expertise and having knowledge about about loss prevention. This is actually two pictures from inside FM Global's large Fire Technology labs. This was a job that I that I have prior to joining The Innovation team and forgive me for using fire examples, but that is my background but similar thinking is going to apply two natural hazards in flight and wins and so forth. But let's just assume that we were tasked as a fire protection engineer to tell the client what type of sprinklers they need to install two on the left here. There is

a 50 ft High storage of expanded plastic. So these are the type of things that you see in me trae's cool boxers in in the kind of foam in expanded plastic on the right. You see hard plastic pallets on this is unexpected an expanded. Vulcan Materials are the same right? So you got me Tracey are solid plastic. Use both polyethylene to on a molecular level is very much the same it's just a physical configuration that's difference between the phone form here. And it says in the solid form here as you can see right here in a close up.

Don't want you to imagine doing the following. I want you to go screen and grab this phone call and then grab the solid plastic any chance. We got the phone ball in your lap dance all the plants in your right hand and I just want you to think about which hand is going to feel warmer. Is it going to be the last pants with the styrofoam or is it going to be to ride hand with it with a solid plastic? I think we can all agree that is going to be the left hand is going to feel warm. Right and it's not because the styrofoam generates heat is to keep from your hands not being able to penetrate into

it because it works so well as an insulator ride, you can just imagine if you hold us a metal bar all the heat from your hand gets on diffused into the material so effectively that it feels very cold. So what can you do with that information? Well, if the surface of the styrofoam heats up very easily because it's very isolated if there's a flame impinging on it. It means that the service is going to heat up. That's what is the service is going to heat up very fast. If it heats up Beyond a certain point is going off gas flammable gas in the process is cool pyrolysis. So in the Star

phone case, it's going to happen very very fast because the service is going to heat up very very fast. So what you would expect from the styrofoam from the expanded plastic The Very Fast Fire Grill, and in fact, if we should jack both of these set up to a fire testing, this is one minute, then you can see that the styrofoam and it's expand a plastic the fires already reached closer 30 ft. Plastic fire hasn't grown at all. Dana's purely derived from the knowledge. That one hand is feeling warmer than the other. So what it what kind of sprinkler do we need? We want to

know what kind of sprinkler we need rights. We got very fast by a growth. So we need to get a sprinkler that responds very very quickly. So that means that we need a sprinkler that has a very tiny link that heats up very quickly. So that activates Road quickly. Also, we know we need to Spring her that is a low temperature rating. So don't reactivate to the low temperature should really gets water to that fire quick because you don't want this fire to grow any bigger because it is really going to grow fast. So what else do we know just from the fact that that hand felt warm, right? This is

just there's no math here. Nothing. I just arrived from the fact that one material feels warmer in the hands in the other. Well, if the fire grows very fast is going to have a very violent Flume the fires going to have a very violent Flume is going to is going to go up very very fast. So what does that mean about the sprinkler? It needs to have a very big droplets because it'll drop us are very small. It's like I'll be able to penetrate the flu Manresa fire. Just going to blow away. How do we get big drop? It's coming out of sprinkler, but we need to make sure that the pressure doesn't

get too high. Cuz if you got the pressure too, high you going to Adam eyes. Are you going to really get this fine Mist which we don't need because we're going to run and get that water onto the onto the fire. The other part is this is going to be a big fire. So how do we get a lot of water on to this fire? Well, we're going to need a sprinkler with a large floor beside a large hole because we want to have low pressure but we still want to get all the water. We better have his finger with a large orifice. So there you have it just from feeling the material and feeling that it feels warm in

your hand. I was able to derive that we're probably going to need a large sprinkler that's quick response with a 165 low temperature rating at moderate to low pressures low and behold if you go to the NFPA standard for if you go to the FM Global same as you going to find the recommended for this Saturday is a sprinkler off NAT type. So I didn't use any AI I didn't use any math. I need use any equations. This is purely understanding the problem. Just imagine if you wanted to do that with a i e want to make a prediction with AI how would you formalized that knowledge? Right? How many

tests would you have to need? Bring it to a dogs and cats and now imagine fire to grow fast in Pfizer. Don't grow fast how much training day. What you need to say anything sensible like that. That's my whole point. Don't forget about the IMEI. Don't forget about the things that you do that and you're intuitive thinking. You can apply the same thing to roll paper fire paper feels kind of warm too, especially if it's later and this is really from the point of ignition. This is 40 feet high and her 66 foot ceiling in the fire reaches the ceiling only within 20 seconds,

right? This is the kind of test that we can do. How do we use AI hair? Right how we do how can I use a high here to tell us anything that makes sense about how to protect these buildings? Focus on the IMEI how can artificial intelligence enhance your strings and me to another book by Thomas Malone super Minds the super minds are groups or individuals acting together in a way that seems intelligent and soda Milan Squad here as we are thinking too much about people or computers and not enough about people and computers and that's going to be my

my point here you have this really domain expertise as an organization. You're not going to replace that domain expertise with AI you got to think about how can you interact? What problems can you start with a i in between and connect that to the people that have that domain expertise? We have your if you have a classical organizational structure that structure might stay in place. That's fine. But you're going to have to find a way how to seamlessly integrate that with the machines in with the computers. Right? So everyday de and every piece of information is now for the

accessible cuz if you can do that now you can log in outer pieces, you can have cloud-based data where you can have people cell phone data, and if you can interact with that, you can really create one big super mind and just my lesson for is a literally examples. Ariana time, I'm almost done. I'm too tired to think about people and computers can be a lot more intelligent than a person in a group and then finally Bill quotes about this information at your fingertips. What do you need to do

if those fingertips are no longer human, but if it's a machine fingertip against me to lesson 5, which I died when you implement a i and a data strategy think like a machine rather than forcing a machine to think like a human because it's a much more effective way to make use of that data and your AI strategy. So that's really it and sorry. I think I use for my time for a reminder everybody that not every transformation needs to be digital That's My Son shows here. And I with his Halloween costume that he's going to use next week. And without I would just like to say

Cackle comments for the website

Buy this talk

Access to the talk “Computer vision, Data Analytics, Third Party data By Jaap - de Vries, PhD, FM Global”
Available
In cart
Free
Free
Free
Free
Free
Free

Ticket

Get access to all videos “Global Artificial Intelligence Virtual Conference”
Available
In cart
Free
Free
Free
Free
Free
Free
Ticket

Similar talks

Hudson Mahboubi
Sr. Manager Of Data Science at Workplace Safety & Insurance Board
Available
In cart
Free
Free
Free
Free
Free
Free

Buy this video

Video

Access to the talk “Computer vision, Data Analytics, Third Party data By Jaap - de Vries, PhD, FM Global”
Available
In cart
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
566 conferences
22974 speakers
8597 hours of content