Alex Fleiss  0:00  

Learning from its failures is a very big part, by the way. And in fact, the past performance is a very useful data set, for a machine learning system on top of data, you feed it itself.  Its own performance, that’s great helps, and helps it learn. But if you don’t moderate the system, it will try and learn quicker than it should. And then it’s failures will rise more than it should. So we try to keep our machine learning you know, kind of moderate, you know, so that it doesn’t, you know, moderately positive Of course. You know, think of it as you know, singles National League style baseball, I suppose the home runs, will never want it to trade on its highest prediction because its highest prediction will have so much inherent risk in it, that the risk reward is too high to be buying for clients. As a business really.

Intro  0:53  

 Welcome to CREPN Radio for influential commercial real estate professionals who work with investors buyers and sellers of commercial real estate coast to coast whether you’re an investor, broker, lender, property manager, attorney or accountant We are here to learn from the experts.

J Darrin Gross  1:13  

Welcome to Commercial Real Estate Pro Networks CREPN Radio, Episode Number 257. Thanks for joining us. My name is J Darrin Gross. This is the podcast focused on commercial real estate investment and risk management strategies. Weekly we have conversations with commercial real estate investors and professionals who provide their experience and insight to help you grow your real estate portfolio. Let’s get into the show. Today. My guest is Alex Fleiss. Alex is the CEO of RebellionResearch.com. They’re an artificial intelligence company. He’s also an instructor at Cornell Engineering. And in just a minute we’re going to speak with him about AI and the potential to disrupt real estate. And before we do though, I’d like to remind all of you that if you like what we’re doing here with CREPN Radio, there are a couple things you can do to help us out, you can like you can share, you can subscribe. And as always, we’d love to hear from you if you want to leave a comment. Also, if you’d like to see how handsome our guests are, be sure to check out our YouTube channel. And you can find this at Commercial Real Estate Pro Network. With that, I want to welcome my guest, Alex, welcome to CREPN Radio.

Alex Fleiss  2:40  

Thank you for having me on. I really appreciate it.

J Darrin Gross  2:43  

I’m looking forward to our chance to talk today. Before we get started, if you could take just a minute and share with the listeners a little bit about your background. 

Alex Fleiss  2:53  

Great.  I started programming video games in the mid 90s and then in the late 90s. I got very preoccupied with machine learning, which is the branch of artificial intelligence that deals with the teaching itself. So AI can be any intelligent system, dishwasher essentially one of the most common forms of AI, this planet.  But machine learning can adapt its code to a changing environment. And so that really caught my eye in the late 90s. I’ve been dealing with that ever since and RebellionResearch.com has clients in 40 plus countries.  We use a base who’s a well known mathematician  we use his brand of machine learning and we cover the global economy. So you know, we really at the end of the day are just economic forecasters. You know, machine learning is my passion. and applying quantitative metrics to problem solving is also my passion. But you know, we also do a lot of research Rebellion Research often see it’s in our name, but no, we publish up to 20 pieces. As we know, our contributors will include the head of data science from Accenture, or a well known machine learning author like Peter Cook. Now or he’ll sit down General David Petraeus, or astronaut Scott Kelly is a perspective on technology. So, you know, our primary focus is, you know, really just staying on the cutting edge, whether it’s through our technology research, or through our development of AI, and real estate’s something that you know, really fascinates me. You know, I, I was lucky enough to marry into the Silverstein family who control built the World Trade Center. And so, you know, I’ve been slowly learning about the industry over the last five years. And it’s really a very fascinating industry that is highly quantitative as well.

J Darrin Gross  5:00  

Yeah, no. The, you know, you mentioned the machine learning then in the dishwasher had thought of dishwasher with AI and that. Can you expand on that? Just say I mean that example where you how you apply or how you quantify machine learning there.

Alex Fleiss  5:21  

Yeah. So as a career, yeah. So basic AI is just any machine that can follow an intelligent set of instructions. And so with a dishwasher, release cold water.  Release soap.  Release hot water spin, you know, release hot water, again, dry whatnot. It’s an intelligence cycle. And you can open the machine so that you can mix up the cycle depending on the ingredients. And so that’s, that’s all you need for AI. Ray Kurzweil, has spent a whole career making these big grandiose speeches and, you know, I feel like half the time of the speeches. He’s already just explaining to people how simple AI really is.  Machine learning is where things get exciting. And machine learning is where, you know, the potential really becomes limitless. Machine learning is very simply the idea of, you know, a, you know, for in layman’s terms, think of it as a robotic brain, one where the system makes its own decision on how to proceed. And you know, it can be if you’ve seen the Dustin Hoffman, movie Rain Man, that gives you a good idea kind of in terms of a vantage point of how this intelligence operates.  In that it’s very stupid in most ways, but it has certain applications like counting toothpicks or counting cards where it’s fantastic. Of course, if you move a few degrees, the applications abilities are zero. And so, think of it like an outboard motor for a boat, you know, that might work well in a small lake in Maine, but take that thing to a rough Bay or an ocean and it’s really worth worthless. And the application beyond that is, that’s probably not the best example because, you know, a, a small outboard motor will have some effectiveness in you know, in a bay, but in a combination but most machine learning, it really can’t function outside of its specific application. So you build a machine learning for facial recognition. It will take a lot of work to alter that so that it can work on something else. Be it smart document search, natural language processing whatnot. So you need you need to really work with the technology and treat it like a like a Dustin Hoffman’s characters type and that it’s very fragile. And you really have to spoon feed it as best as possible otherwise your results will fail very quickly very soon.

J Darrin Gross  7:53  

Got it.  So question for you the the the system In the data that these systems are utilizing, is that is that constantly changing? Are there are there? I mean, is there a, is the machine able to learn, based on the data that it’s receiving isn’t able to decide it needs additional data points? Or is that more of where the programming comes in?

Alex Fleiss  8:26  

Well, it’s funny you mentioned that actually, because when we first started in, 2007, I think like a year later, we were giving it not nearly as much data as we’d give it today. We weren’t giving it data on currencies, but our machine learning was able to find data on currencies and use it.  And then 

J Darrin Gross  8:47  

Oh really it was it was it was seeking it out and found it and used it all by itself.  

Alex Fleiss  8:51  

Yes, yeah. And so we so we started giving it you know, we saw that was using the Euro. And so we started giving it more and more currencies.  So that was very interesting. And so, you know, we’re trying to give more data, but the data has to be concrete and consistent. And so like Twitter data, you know, we really can incorporate that. We like an industrial output is something that’s more consistent retail sales. You know, our our technology views the global universe, as you know, very, very, very interconnected when it comes to the economic universe that is rather. And so, you know, when it looks like Chile, for instance, it looks at all of the assets that make up Chile. And, you know, as those assets have greater and less effect over time, you’ll see that like, what’s Australia, from 1990 to 2010, Australia went from like 5% correlation to China to like 50%. So really over 20 years, Australia went from very little sales to China to basically becoming China’s leading trading You know, it’s so that that makes Australia’s you know, data inputs for what is going to predict the economy’s movement very different because you have a different economy 20 years later.  The way Apple Computer 1994 is a very different entity, than Apple Computer 2014 sells different goods, different masses at different margins, remember different people. So the idea that it’s a constant asset is almost it’s not it’s extremely changing asset. And it might still be in technology, but a lot of the correlations ends there.

J Darrin Gross  10:41  

So the machine is actually able to learn. I’m curious, is that based on the preset programming that was originally the real code that was written or is this machine currently writing its own code?

Alex Fleiss  10:54  

No we can also no we that’s how it’s designed to work but you know, we can Also moderate the rate of learning so that it doesn’t learn too fast, because the systems are naturally aggressive. And so they will try to learn as fast as possible, and then become aggressive with what they think will happen. And so, you know, we have to moderate every part of it, we, you know, have to, you know, really kind of treated very, very carefully with, you know, kit gloves as well. And so, you know, it, you know, it’s a fantastic beast, but it really, it must be very monitored and very massaged.  And so it’ll overexpose itself to industries, it’ll get to, it’ll decide that it can’t pick natural gas, it should never buy natural gas again, or, you know, it’ll look at its failures and try to learn, you know, learning from failures is a very big part, by the way. And in fact, the past performance is a very useful data set, or a machine learning system on top of data, you feed it itself.  Its own performance, that’s great helps, and helps it learn. But if you don’t moderate the system, it will try and learn quicker than it should, and then its failures will rise more than it should. So we try to keep our machine learning, you know, kind of moderate, you know, so that it doesn’t, you know, moderately positive, of course, you know, think of it as you know, singles, National League style baseball as opposed to homeruns. And we’ll never want it to trade on its highest prediction, because high vision will have so much inherent risk in it, that the risk reward is too high to be buying for clients. As a business, really, it’s just, you know, whereas if you scale down the risk, and you scale up confidence ratio, confidence ratio is very important. So there might be a 60% chance of this asset. will go up. But how high is your confidence ratio that that will actually occur. And so the confidence ratio is very low. And we take that very much into account. So that’s a case where, you know, maybe the system likes global foods, but it’ll end up buying Nestle because all the other global food companies are too expensive or have other issues that make it too risky and the competence ratio is too low in terms of moderating the risk. Risk is very, very important.

J Darrin Gross  13:30  

Yeah, no, on the different ratio. You mentioned, the confidence versus the I apologize I can’t remember the other ratio. You’re coming down there that are you constantly developing new I guess measures for the words the machine able to pick up and identify other measures that then correlate help to correlate or

Alex Fleiss  13:58  

Yes. At the End of the day, we’re pattern seekers, you know, like, from the movie, The Imitation Game with some touring mathematician who helped solve the Nazi Enigma code, you know, so we’re really just trying all the time to find those patterns, those patterns that predict positive or negative performance. And, you know, so you know, the problem with this black swan event of COVID-19 is that we get a strong economy. And there was no reason to go defensive on the US economy except for panic that quarantining would kill everything. And so things change very quickly and it was a it was definitely a it was as Black Swan event, as will come. So it’s very difficult. If you look at all of the a lot of the best computerized firms, some of the results haven’t been pretty there. You know that that’s about the markets not always purely rational. This was this reminds me of 2011.  In 2011 the market crashed maybe 35% people thought it was all over for like a month or two, and then the market just ran back up. 25 30% was nine years ago now. But, you know, and of course COVID-19 is, I think, a good more significant issue than the debt downgrade. But once COVID-19 is finished, you know, washing the economy not be what it was, I mean, of course, we’ll lose restaurants and whatnot. But no, it’s, it will see it for us, it’s hard to make predictions like this.

J Darrin Gross  15:33  

Right, right. So when you’re designing a system or training, how many data points what’s what’s a minimum kind of a number of data points that you look for, in order to establish some rules?

Alex Fleiss  15:52  

Well, for for trade to occur, we need to have you know, 50 to 150, positive factors and each factor relates to a cheap PE Ratio, strong balance sheet strong industry, maybe coppers, you know, flying and this is a copper company, maybe France’s economy is doing great. This is a French company. And so, you know, whether it’s the geography, or the industry or the specific country, you know, we’re looking for, you know, these positive patterns and we want as many positive correlations as possible. So this company correlates to a strong industry, a strong country, strong financials, you know, its pricing has been working, that’s something else, this system tends to sell stocks that don’t work, you know, and so if the markets humming along and one stock goes down for no reason and stays down for months, the system will try and figure out why it’s wrong. It will just move on, which makes for a very tax efficient index. So at the end of the day, you know, we had this AI index on 120 stocks. It’s been running for now 14 years in SP performance, but if you Generally creates a tax loss because it sells the losers within three to five months and holds on to the winners for multiple years. So we you know, I, I always say I challenge anyone to find a more tax efficient index than our index because of how not just how we design but the the nature of the system and that it sells its losers quickly and holds on to its winners for a long time. So that’s a nice benefit of it. .

J Darrin Gross  17:25  

Got it thinking of real estate and potential applications? I’m wondering would, would it would it be more useful in commercial real estate or what do you think it would apply well, in residential real estate given the emotional you know, the intangible with a kind of the it’s going to be your home as opposed to a commercial real estate where it’s more about the the numbers.

Alex Fleiss  17:54  

I think both are very tricky, to be honest. I don’t think actually AI works perfectly for real estate. I’ve been examining this now for quite a few years. So I actually I think, you know, there’s some trouble, I think it could work more in smaller localized situations, because AI performs best in a vacuum, if you will, when you when you can lessen the factors, it’s best. So maybe like in Manhattan, you could have a machine learning system that’s looking at all the different blocks, looking at the values looking at the sales, and it can spot like an apartment in one complex that’s trading 30% below its comparables, something like that, that’s where it could, you know, help or maybe for you know, like a localized basis. So you know, maybe a area of Phoenix. So you know, there are those potentials. And I think it will be applied. Like I said, just because it’s trippy doesn’t mean it won’t be done. It doesn’t mean someone won’t do a fantastic fantastically well. But I think both commercial and residential real estate, have significant avenues for people to develop them using machine learning for investing in real estate and maybe in time even finding the best broker, what broker will sell this type of partner the most. You know, when people pick brokers, it’s almost like you’re flying blind into the wind. I like this guy. My buddy likes this guy. My college roommates, wife plays poker with her sister, you know? And so okay, but how about what broker sells, you know, three 4000 square foot houses in this zip code for the highest price on average over the last two years? I’d be curious to know who that is. And that would be the broker I would want to hire. Even if it’s only by chance, well, maybe that broker by chance and got lots of rich people overpay for houses. And so there are these kind of these niche areas of real estate, where there’s quite a bit of a market, I think you can make money, whether it’s a broker search, and that’s something that compass is doing and actually, we put on our board. The lead machine learning engineer from Compass.  You know Compass is a big dislocator in the real estate world, and we’re excited to have him join.  You know, machine learning is still still very at the early stages when it comes to real estate. And so it’s very fascinating watching how it’s applied. And that is,

J Darrin Gross  20:19  

What about beyond the the actual kind of the trading aspect where you’re determining value or, you know, one’s better than the other kind of thing, but, but actually in the operations, is there an opportunity for machine learning or AI? With respect? Whether it be reservations or

Alex Fleiss  20:41  

Yes, yes, yes, you know, throughout the entire system. Machine learning is a really fantastic way to automate with slight intelligence, a lot of menial tasks. So from reservations, to insurance, And in the actual operations of real estate themselves in real estate measure companies, you know, they are going to be able to automate. I mean, you look at something like energy, and that’s one of the kind of most old school industries out there. Original Cowboys, you know, I’m sure we all saw There Will Be Blood. And you know, we get, you know, you got some, some heavy personalities and energy industry, definitely not a techie place, but energy because of the oil crash. The last six years has really embraced AI because there are a lot of simple parts to the business where you can apply AI save a lot of money. And so I think that you know, if energy and drilling can embrace AI, without a doubt, you know, real estate, the same, whether it’s drones for inspecting houses, you know, or investing in houses or for managing your properties, instead of, you know, having, let’s say a New York management company wants to buy a property in Philadelphia, but their main property guy that you know, he doesn’t want to be driving back and forth for hours. You know, in time, maybe they can have the outside of the property inspected by a drone, as an invention in the inside will be inspected by another drone. And so the idea to virtually inspect your property without having to be there, that could really change the game in terms of cost you don’t know. It’s all about creating intelligent ways to automate a lot of these menial tasks and replace, you know, people and then with that cost savings, you can expand or grow more and overall, more people will be hired. So I always I always caution people don’t think this is like a zero sum game where you lose a job here and you don’t get to there because that’s not the way that automation and innovation I’ve worked the last few thousand years.  Over mankind’s history, innovation has consistently made jobs higher, paying safer, and more comfortable. And so jobs that were riskier, 100 200 years ago are less risky. Of course, there’s still risky jobs out there. But the amount of risk has decreased significantly and will continue to decrease significantly. That’s that’s one thing about, you know, technology, but not only does it make these apps actually retreats, more jobs and more jobs. When you look at something like a zoom video, which we’re using today, created by a Chinese immigrant who didn’t speak English, flunk thousands of people created billions of dollars of value. You know, I zoom has enough of a reason to be a capitalist, the whole socialist movement really sick and to me, frankly, because, you know, 40 50 years ago, we serve a country where whether you’re liberal or conservative, realize socialism was a losing cause. John F. Kennedy was a very, very, very, you know, hawkishly anti socialist communist person. So yeah, that, you know, that definitely very much upset me. But yes, as we’ve seen with technology and innovation over time, there are more jobs, more jobs that pay more, that are safer and more comfortable. So it’s really consistently been a win win win for the worker as technology unfolds.

J Darrin Gross  24:17  

So speaking to that kind of the history and how it’s moved. How, at what speed Do you feel like AI and machine learning is upon us we’ll be moving as we go forward. I mean, there there is it, I forget, there was some sort of like Moore’s law about technology about how it you know, the the pace 

Alex Fleiss  24:41  

Yeah, I don’t I don’t.  Yeah, I wouldn’t, quote any specific law with the progression of AI. I will say that we are very much at the dawn of the dawn. If this were the New World. I’d say it’s more like Cortez has burned his sails and he’s marching towards  Tenochtitlan, but you know, we’re not at Jamestown yet Plymouth Rock.  We’re still very, very, very early on singularity, we’re still years from significant cognitive, you know, intuition years from these are still very stupid machines.  You know, they they need to be applied in very, very, very direct manners and you need to spoon feed them the data as much as possible, otherwise they will fail. So, yeah, we’re still very early on.

J Darrin Gross  25:28  

One of the things it’s, it’s always been kind of the, I think the challenge for for myself anyway with technology is that I get all in with a particular brand or model or style and then thinking that that that’ll it’ll be built upon that or future versions will be built upon that only to be you know, only find out later that that one’s totally scrapped or in a you know, a new one and you’ve got to constantly be learning. Is there is there any kind of like a base level for This, this, I mean is or platform or I mean, I don’t know if I mean giving you the right question or if I’m speaking in a way that makes any sense. But is it does it continue to build upon itself? Or is it something can they can be totally dis unattached, that becomes the new focal point of of how this works that make any sense?

Alex Fleiss  26:23  

No, no, it makes sense. Yeah, no, very much academic research is building upon itself. And, you know, as we move forward, moving forward faster, faster. And our progression is, is definitely increasing.  And and unlike the previous  AI bubbles of the 50s and 70s, this bubble has created a lot of profit. And so there’s a ton of AI companies that make a lot of money. And you have biggest tech companies completely embracing machine learning. The largest companies on the planet, Microsoft, Facebook, Amazon, Google, all They are the ones who are the leaders of machine learning. Machine learning has gone from bubbles to now leading the economy, they are buying the best engineers. And they are getting most of the data.  

Data is the, data is the blood that makes the body move. And without blood flow. You have absolutely no sick no human being you have no animal. And so the more the better the bulk blood flow, the stronger you are, you know, I remember Steve Prefontaine, the famous middle distance runner from University of Oregon. Yeah, he was known for having one of the most efficient blood flows in the history of the planet, I think was the you know, the co founder of Nike who was his coach, I had partnered with Phil Knight said that, you know, he never ever wants

J Darrin Gross  27:32  

Bowerman

Alex Fleiss  27:36  

 to If it’s Yeah, exactly, Yes, exactly. Nobody’s, you know, blood was as efficiently pumped as Prefontaine, which allowed him to be not just so good, but also he was a chunky guy, he was a large guy, you don’t typically have guys of that size, able to run that fast. And so with Google, Apple, Microsoft and Facebook, because they have so much more data, their machine learnings, you know, will will be the best on average because they will have the most data to work with. And the more data the system has more intelligent isn’t has to work with working make intelligent decisions. So it’s a race now it’s totally an AI race and AI engineers around very expensive.

J Darrin Gross  28:48  

So you mentioned data in the collection of it. Is is the data that’s being collected Is it is it recognized up front is is a data point they want for a specific person purpose? Or do you find that a lot of times the data is being collected, so that it’s available for future analysis.

Alex Fleiss  29:10  

I mean, Netflix is trying to collect as much data as they can on their customers and their customers actions, and they’re working on it and every different way they possibly can to better their systems.  You know, they have teams, ie these companies and teams and machine learning engineers that are playing with their data, trying to make their companies more efficient. And so, you know, in terms of data, they want as much data as they can, wherever they can get it, your data is it is the new oil and you know, every every little bit of data has become valuable now and so because the you know, the Fang the Facebook, Apple, Netflix, Google have the most data. That’s why they’re gonna naturally be leaders in this industry.

J Darrin Gross  29:52  

Yeah, no, it’s interesting as an insurance broker. If you like for years, we’ve been plugging data into the machine. But it’s only like recently that it’s really started to be used. And if that makes sense that you know it, I don’t know if it, I don’t know if that was understood going in or if that’s just kind of reality of the fact that you have now data scientists that are able to, you know, analyze the data and all all different views and come up with different correlations, and then the machine learning etc. That’s been able to utilize it and, and draw some correlations that can then be translated into economic success. 

Anyway, hey, Alex, if we could like to shift gears here for a second. As I mentioned you before we started by day, I’m an insurance broker. And I work with clients to assess risk and try and manage it. And there’s a couple different strategies we consider when managing risk. The first is called can we avoid the risk for not able to avoid the risk we look to see if there’s a way to minimize the risks and then minimize or avoid it we look to transfer the risk and insurance is a transfer risk Transfer Tool. And as of late have been asking my guests and and if they can take a look at the world as they see it and their world and identify what they consider to be the BIGGEST RISK. And just for clarification, not necessarily looking for an insurance related answer. But if you’re willing, I’d like to ask you, Alex Fleiss what is the BIGGEST RISK?

Alex Fleiss  31:49  

Well, you won’t forget my last name because my cousin is Heidi Fleiss, of 1990s fame and Dennis Hof’s widow, but I’d say the BIGGEST RISK is the unforeseen risks without a doubt, which obviously is manifest in the COVID-19 Black Swan event or the S&P debt downgrade of the US government in 2011. The unforeseen risks is always always the worst risk, but it’s very, it’s very hard to worry about risk, you know, because there’s, it’s almost a it’s really endless. I think about my, you know, my, my, my Papi and 911 some people told him Don’t take any insurance out for terrorist events, and then other people telling him to take more. And so but Terrorist are gonna blow up the building. That sounds ridiculous, but then of course, what six weeks later happened. And so, you know, we’re living in, you know, in times where the unexpected is expectedly happening now. Whether it’s America’s Cup, the best comeback of all time, the idea of winning seven Sailing races in a row. So unlikely I’m a sailor. And so I got to play the odds that are just almost impossible. But, you know, we’re getting to these kind of new. We have such a globally connected world, which obviously has nothing to do wiht sailing. But, you know, it presents more and more and more potential risks against the globally connected world. If a risk arises in India or China, it will travel faster than ever did before because we didn’t have that connection in the world that we have now.

J Darrin Gross  33:30  

No, absolutely. We’re a little beyond bird flu. As far as a means for transmitting that kind of stuff, and like so just information in general, just the way we’re so well connected and stuff. It is kind of, it’s a fascinating and almost a scary reality of just how, you know vast is how, how easy, it all moves.

Alex Fleiss  33:53  

So I like to think about Carl Sagan, the famous astronomer astrologist, who says, Don’t forget, you know, we’re just A bunch of little creatures on a forgotten rock in a nothing solar system and a ho hum galaxy. There’s nothing, there’s nothing special about it. Now ours our star alignments aren’t special planets aren’t special, nothing about anything about us is special. The fact that our one planet allows our, you know, specimen to live here and other specimens. Look here doesn’t make anything about us. So, and when you start realizing that there’s, you know, another hundred or 200 billion galaxies out there, you know, it’s very small, and the potential for risk gets at most.

J Darrin Gross  34:39  

Right, right. No, that’s good. Good to remember. Alex, where can the listeners go if they’d like to learn more connect with you. 

Alex Fleiss  34:49  

Thank you.  So if you want to learn more about rebellion research, go to our website Rebellionresearch.com  you can open a brokerage account.  Or you can go to our research and read about AI machine learning flying cars, covered hundreds of topics and hope you’ll become a reader and a subscriber.

J Darrin Gross  35:32  

Alex, this has been great. I appreciate you taking the time to talk. And I’ve learned a lot and I hope we can do it again soon.

Alex Fleiss  35:42  

If it’s not a pleasure is all mine. You have a great day.

J Darrin Gross  35:44  

All right. For our listeners. If you liked this episode, don’t forget to like, share and subscribe. Remember, the more you know the more you grow. That’s all we’ve got this week. Until next time, thanks for listening to Commercial Real Estate Pro Networks. CREPN Radio.

Intro  36:02  

You’re listening to CREPN Radio for influential commercial real estate professionals. For more information on this or any of our guests like us on Facebook CREPN Radio