René Morkos: And long story short, ended up developing ALICE, first construction, you know, generative construction simulator, a tool that can generate millions of different ways of building a given project. Really to solve the problems that I wanted solved when I was in the construction site.

Ginette: I’m Ginette,

Curtis: and I’m Curtis,

Ginette: and you are listening to Data Crunch,

Curtis: a podcast about how applied data science, machine learning, and artificial intelligence are changing the world.

Ginette: Data Crunch is produced by the Data Crunch Corporation, an analytics, training, and consulting company.

Curtis: Welcome, Everybody, to Data Crunch again. Today we have René Morkos on the show.

Um, he’ll talk about some interesting things in machine learning/AI specifically in the construction industry, but expands to a lot of different areas as well. He’s a professor, adjunct professor at Stanford. CEO of a company called ALICE Technologies. But René, maybe I’ll, I’ll let you introduce yourself and tell us where you’ve been, what you’re doing, and we can go.

René: Yeah, absolutely. Really great to be here. Thanks. Thanks for inviting me, Curtis. Uh, my background is I’m a, I’m a construction guy through and through. My dad gave me a good piece of advice when I graduated high school. He said, “Son, study anything you want, just don’t do civil engineering.” Like, “got it. I know what I’m going to dedicate my life to now.

And so, yeah, I, uh, I like building stuff. You name it, I built it. I’ve done under water pipelines, $350 million gas refinery in Abu Dhabi, on middle of nowhere, I was in Afghanistan as a private contractor, worked for a German company down there. Designed, built, procured my own jobs.

I worked my way up the ranks. I started as an assistant site foreman, became a site engineer, became a project manager, then was running several projects, then ended up doing a PhD at Stanford. Kind of was doing six months on/six months off. So I kind of fund the program that way, really consulting you’re in my six months off. I started getting interested in how do you run companies?

I was doing a lot of work in like cost coding, daily reports, monthly reports. BAM was a big deal, you know, back in, must have been, 2010. So yeah. And long story short, ended up developing ALICE, first construction, you know, generative construction simulator, a tool that can generate millions of different ways of building a given project. Really to solve the problems that I wanted solved when I was in the construction site.

Our field has, does not have a solution that answers the most fundamental question, which is how do you build the darn thing in the first place? Or what do I do if there’s a delay? Right? Um, and my thinking was always that, which is, “I want to come to the office in the morning, do my site walk, right? I want to plug in what happened yesterday in the computer, and let it do all the thinking for me. Right? I want to finish my coffee. And then it spits out, “Hey, here’s what you need to do today. And if you don’t do this, here’s, here’s how late you’re going to be.” Right? So that’s kind of, um, the so that’s the background.

Curtis: It sounds like you started in construction, civil engineering. And then how did you get into the AI field from there to figure out how to solve this problem?

René: Yeah, that’s a great question. I think that my environment was definitely, you know, a part of it. I mean I studied at Stanford, which is, you know, Silicone Valley. And so everything’s a nail to hammer right, so to Silicon Valley, everything, everything’s a software problem that needs to be solved with software. There’s definitely a limitation to applying that to everything in the world, I think, on in various discussions have come up recently in the press about that.

But, uh, for me, uh, my background was construction. I took a course back in high school for Pascal programming. Um, did, ah, another class in basic programming the language, and then did a class university C++, and then the class in Java, right?

But, uh, so I’ve taken like a few classes and basic programming languages. I think that the one class I took, which was an introduction to AI at Stanford, so it was like a big eye-opener to me. It really, it’s an introductory course, but it really I think helped shape how I think about problems, how to apply machines to solving them, you know?

And then if anybody’s listening to the podcast, it’s all, it’s all available. Stanford’s CS 106A and 106B is on YouTube, right? Um, you can literally take it nowadays with Coursera and Udacity, all this stuff, like definitely worth three months of a couple hours a week of seeing how do computers work. So that’s kinda how I got to it, I think.

Curtis: Got it. And this, this was born of this, this need to solve the problems you saw in the construction, because you were so deep in construction, right? And there’s gotta be a better way to solve these issues.

René: You know, Curtis, like a lot of people don’t realize that your rear end is on the line when you’re building these things, in a big way. Like I’ve been on jobs that we’re looking at 50,000 euros of liquid damages per day, 50 million euro job. You wake up today, you’re $50,000 short. Tomorrow, that’s a hundred; day after, 150; day after, 200. Another week you’re out a quarter million dollars.

You can imagine like that the mood, right. And on the construction site, when this starts happening. Yeah. ‘Cause you’re, you know, you quickly burn through your profit margin, now you’re losing money. Now and so-so, to me, like, what I always wanted to do was, was to help that person, that girl or that guy or whoever it is it’s sitting there dealing with that situation. That was the objective.

You know, to get like . . . the way I kind of view it, right? Um, is people today in construction are trying to fight a war with muskets. You know, it’s, I’m sorry. It’s kind of ridiculous at what, you know, a lot of the tech that they’re being given, right? You know, these Gantt charts, right. You know, and it takes 16 lines of code to write up a CPM algorithm.

One six, we’ve been stuck in that technology since 1961. Like, there’s gotta be a better way. Right. And every other, like I’ve hired software developers that have come to the company and said, “René, um, everything in the world uses AI.” I remember this one guy specifically, so “everything in the world uses AI, like satellites, computer games, cars, like everything. How are you telling me that an industry the size of construction isn’t using artificial intelligence?” Like he wouldn’t believe me. And I’m like, “I’m not kidding you. I swear to God, like, here’s what, here’s how this field works.” And so, but it’s happening since then or 2017, right? All this money’s being for construction tech is now the sexy thing and sexist thing in the world, right, which is great.

I’m so happy to see all these startups, you know, sprouting up and being successful, right? And Fieldwire just sold to Hilti for three them. God bless those guys. You know, you’ve gotten, starting to have, you know, multi hundred million dollar exits, right.

Curtis: So, so tell us about this approach then it’s AI we’ve seen applied to other areas, like you said, construction hasn’t seen it in a long time. They’s starting to be more, uh, interest in the space. What, what is your approach? ‘Cause you’ve been there, right? You’ve seen the deep problems, like how does this get solved?

René: So there’s kind of like the, the north star is always that. The north star was, was we set out to solve the problem. We didn’t want to build a photo sharing app we didn’t want to build some file sharing thing.

We didn’t want to build some forms that you fill out. Like, we didn’t want to take an existing process and digitize it. That’s, you know, and we realize that that was going to be a lot harder, a lot more difficult. There’s a lot more unknowns. Like we spent three and a half years with core R and D look like there was literally, you know, three guys in a basement kicking this thing over and over and over and over and over again.

And it caught fire more times than I can remember, you know. Um, so the, the, the approach was, was for you to apply AI to our field, that you first have to convert it into an AI formulation. And that’s kind of what I did in my PhD or our field actually doesn’t have a developed conceptual model. So what is the conceptual model?

It’s like, people would come to me from the computer science department, say, “Hey, I’m interested in working on construction. I’d be “like, oh, that’s great. Really? Like, thanks. Thanks for being excited about our field.” And they’d say, “Hey, can you point me to a book I can read so I know like, what are the pieces of the puzzle like I have to put them in a computer?” And I’d always say, “oh, we don’t really have one.”

Well, what do you mean? Like I’m looking for a book. Th th that will tell me, like, I have to think about labor and equipment and materials and production rates, you know, here, here are the piece of the puzzle and I’d say, well, you know, we don’t have something that like specifies it in that way. You’d have to read a whole bunch of books.

And so one of the things I did in the, in the PhD was, was really developed this, this conceptual mode, right? Like, here’s a question I like to ask, “is the production rate an attribute of the crew or the task or the element that’s being worked on, right? ‘Cause carpenters can install formwork or they can remove formwork and they can do it for walls or Slouch right?

So these are the sorts of questions that you get to think about as a PhD. Now you’re earning $24,000 a year. You know, your budget’s pretty simple then pizza, right. But you have to spend 16 hours a day thinking about that, you know? And so for us, the way we did it was we basically figured out how to formulate the construction problem as an AI problem, right, or as an algorithmic problem, that is super powerful.

That is the, what the, what, what I did in my PhD. And I remember going to my advisor and saying like, “Hey, look, this thing’s a key.” And he was like, “yeah, whatever.” And I was like, “no, like, I swear to God. This thing is key.”, yeah. And we proved it. You know, you can, you can use all these techniques to solve the darn thing.

And so, yeah, we, we basically, you know, applied a whole bunch of stuff to it, and figured out the best way to do it.

Curtis: Got it. And three and three and a half years later. So three and a half years was sort of the R&D phase, right with this. Um, and tell us about when, when you started the use it, it’s in production now, right, people are using it. What is the mental shift they have to make to think about this problem in a different way. Right? ‘Cause I, I, I imagine in the industry, people aren’t used to using AI, right? So they have to kind of think about things in a new way.

René: So the, the, the, the mind shift is an interesting question. Here’s what we’ve realized about the software we’ve developed. Um, it’s a new way to think about construction. I had a bunch of, you know, very high powered consultants use it, you know, and then come back to us two weeks later and say, Hey, “this thing’s not just a software. It’s a new way to think about construction.”

And I kind of smiled. And I was like, “that’s exactly what it is.” Um, so the, the, the, the point you’re making is an excellent one, which is, “you can learn the software in a day and a half. There’s nothing to it. I’m not kidding you. Like you can it’s what tasks do you need? What resources do you need? Right? You want to build a slab? Here are the five tasks. I’m going to need some carpenter crews. I’m going to need some steel crews. Here’s some durations or production rates. Which way do we want to set it up? So the, the, the setup or the software is simple.

The, the mindshift, that takes on the order of weeks. You know, um, because you’re going from a universe where there’s one option that took you a long time to create. That’s full of errors, and that’s really hard to redo. To suddenly go into a universe where there’s lots of options. Right. And, um, there’s lots of options, and, um, you now have to think about not just comparing a solution to solution, but groups of solutions to groups of solutions, if that makes sense.

You’re also like, in some ways it’s been interesting to see, like from a cultural perspective, you know, a lot less vested in that one solution because it doesn’t cost you anything.

So it’s been interesting to see that sort of mind shift mindset. Like the thing that we see with AI over and over and over again at Stanford is that what it, what it does is it becomes the thing that does the crunching, and the human becomes the thing that sets up what needs to be crunched and interprets the result. A little while later, somebody develops the next version of the technology.

And so what used to be the setup and the analysis now becomes part of the AI and the human takes a further step back and sets up again what’s going to be crunched, and then interprets the results, if that makes sense. One on the head innovation guys at Exxon told us something really interesting, which was a guy called Vinit, Vinit Verma.

Um, and he said, “what technology does is it makes the art of yesterday into the science of today and unlocks the art of tomorrow.” And I was like, bam, bam. I was like, that’s exactly it. When people hear about ALICE, they’re like, “oh, it’s just going to take my job.” Whereas. It’s like, in some ways I almost wish, right?

Like, we’d have a life, but like, if you look at, since the industrial revolution to now, it’s not like we’re working less hours, right? Yeah. That’s what ALICE will do is ALICE will be able to crunch, build your project 6 million different ways. One crane, two cranes overtime, no overtime, very, like very in production.

And so you can tweak a variable out of crane and ALICE rebuilds it for you. So that’s kind of the, you can say, “Hey, the material’s delayed, you know, and ALICE will go through six months sequences. So that’s the, um, that’s the power of the technology.

Curtis: How do you, how did you go about training ah algorithm like that? Right? Is it, is it looking through a bunch of records of past construction projects and, or like, how do you approach that?

René: There’s nothing, there’s nothing in my opinion to learn from currently. You know, that’s the thing that I think because of like our field doesn’t even have a construction, conceptual process model.

It’s a bit of a mouthful, but we don’t have like a worked out version of what are the pieces, how do they interact? Right. And so, like, what I’m trying to say is that we don’t even have the conceptual model, right? Let alone a bunch of software that are running on a common conceptual model that are learning from each other or learning from the results of other performance, et cetera.

Like we just don’t have that. And, and what, I’m what I, like, I had someone about four years ago, come to me and say, “Hey, you know,” very sort of, um, how should I put it? Self-confident gentleman. And he said, “you know, I have 40,000 permit various schedules with their, you know, percent complete. And if you’re lucky, I’ll sell one to you.” I said, “okay, well, let me go think about that night.”

You know, I spent a couple of days sort of mulling it over and I got back to him, and I said, you know, “I really appreciate the offer, but I, I don’t, I don’t know how I’m going to extract value out of it.” He said, “what do you mean?” I said, “well, your schedules, are they tied to a vin model? No. Are they resource focused? No. So the only thing I’m going to be able to glean from it is the description of the task and that’s not standardized.”

So, you know, but there’s this, like, there’s not enough encoding in the data for you to be able to glean that out of there. At least I, I am not aware of it. I’m not. And I’m sure there’s somebody, you know, listening to this somewhere in a basement in Silicon Valley, or, you know, other innovation centers around the world thinking, “oh no, I’m going to prove that guy wrong. I know exactly how to do it. Right.”

But, um, yeah, in my opinion, currently our field doesn’t have that data dataset. And so ALICE learns from herself, which is how we got around that problem.

Curtis: Got it. So now are you trying to build a data set like that with, with, um, with ALICE currently?

René: Yeah. Absolutely. Right. I think it would be great. Like, you know, what I would like to do if I had some time, because I have kind of two hats, right? One is the CEO at ALICE and the founder, and the other is this adjunct professor at Stanford. And so, and I, and I really try to differentiate the two and work on both, but what I would really love to see is a conglomerate of construction organizations, you know, general contractors, owners, you know, schedulers, you know, so-and-so forth, uh, estimators that get together and say, “okay, you know what? Let’s, um, let’s kind of put some, put some resources at this. Let’s build out a conceptual model that everybody can use so that we all agree on. Let’s share some data sets. Let’s put, start putting stuff up in the cloud.” Right. But, um, yeah, it would be great to do that. I’d love to see that.

Curtis: Got it. Yeah. Yeah. That’s interesting. And I want to get, I mean, you mentioned how this can change the construction industry, but I’m really curious your, this R&D phase that you were talking about, we haven’t talked to a lot of people on the show that have necessarily done something like that. What were, what were some of the biggest challenges or things that you hit in this phase? Like what would you recommend if someone was going to do this in their industry, for their own stuff and come up with some conceptual model and try to do what you did in construction. What, what would you say to them?

René: So we’ve gone through a very, you know, almost unique journey, and here’s why: we, we went from conceptual research, which is, “Hey, what are the pieces? Like, do I need to think about labor or crews or individual people on crews or, you know. What are the pieces and what does the information that is attached to each piece that’s conceptually research.

We then moved into theoretical research. To me, theoretical research is as . . . well, how did those pieces interact? ‘Cause if a crew goes works on a component and an element, what’s the output of that? Is it considered a task or is it an operation or is it an activity? Is there any difference in that? So that’s kind of theoretical research. How do those interact? Theoretical research.

Then we do prototypes, right? Because without a prototype, you know, who’s to say that the theory and the concepts are correct. Prototypes we did the commercial product and the commercialization. So that’s the journey that we’ve been on, or I’ve been on for about 13 years now. So, um, the advice I would give is as follows.

Um . . . it is remarkable to what extent things are not solved yet. Like if somebody’s sitting there and you’re like, oh, I, I, because I have literally been blown away by how many times I’ve thought to myself, “oh, yeah, I’m sure that’s how it worked out. Like somebody somewhere has worked this out. Somebody somewhere has figured this piece of the puzzle out. Somebody somewhere.

Like, I don’t need to go think about it because it’s all been done. And I think that’s kind of maybe the biggest piece of advice, which is you will be surprised with how much of it has not been done. You know, if you start, you know, slowly and you, you, you have a white board and you start iterating on the idea, it’s writing on the idea and it’s, um, you know, I would rec go read how to fly a horse.

Best book on, on innovation I’ve read. It was recommended to me by, by one of our investors. I read it, and I was like, “yeah, that’s what my life is like.” What that book will tell you is that, you know, people think of innovation as like, oh, some really smart person going, “oh my God, I have this great idea bow, right?

Whereas it, it tends to be quite the opposite. You go through iteration after iteration, after you solve a little problem by trying six different ways to do it and you solve it. And then all you’ll unlock three different ones. So you try it, you know, six different ways to solve the first of the three and you figure it out and you try four different ways of solving the next and you figure it out and so on and so forth.

And that’s, if you can stick to that and you can stick to that for say two, three years of your life, then you’ll start to get something that’s valuable, and we’ll start to kind of have some traction, but that’s basically. It it’s not some, you know, hidden realm for, for some, you know, privileged few. It, you know, it’s not like you need a lot of money or a lot of intelligence, or a lot of whatever it is.

It’s just perseverance and sticking to the darn thing and sticking it to it for two, three years and keep iterating on it. After that amount of time, you will start to get stuff that’s useful.

Curtis: Interesting. How much of that would you say was deep domain expertise that you had versus like cutting- edge AI algorithms, that kind of thing. When you’re going through this process?

René: Ninety-five percent with domain expertise, 98% of domain expertise. You know, I’m good with algorithms. I’ve been good at math my whole life. Right. I figured out solutions that would work that were simple. Right. But, um, once we had it up and running, we brought in some heavy, heavier duty kind of folk to go lift that piece.

Um, but yeah, I mean, I think that what’s interesting is that I feel it’s surprisingly one of the most important things, if not the most important things I did was tell everybody what not to do.

Curtis: In what way?

René: So, you know, I’ve been on construction sites, so here’s, here’s an example. Do you need to, if you have assimilation, like we do right first in the world runs all sorts of different ways to build something.

Do you need to model movement? So I’m working on column A, and now I’ve moved to column B. To somebody outside the industry, you’d think like, “oh, probably right?” With ALICE we knew that knew that, or I knew that you don’t need to model that. When you’re on a construction site, you don’t think about crews moving from A to B or B to C.

Right. And so that was kind of, um, I think an important, important piece of the puzzle that the domain knowledge brought, which is, what do you not need to focus on it? Alternately even like, what do you need to focus on?

Curtis: Right. Really interesting. And what do you think this is going to do for the construction industry? I mean, this is. What would you say, order magnitude or more improvement in efficiency or what’s going to happen?

René: You’re reducing costs by 13% for labor and equipment and your construction duration by 17% of average. That’s what you’re doing. And the reason you can do that is because your effort to generate a schedule or assimilation as we call it, uh, is 800 times.

It is, it is 800 times less effort to schedule, reschedule, etc. That’s the production, and the way it works is that, the way to think about it is people are trying to fight this war muskets. We’ve built an F16 jet. It still needs a pilot. That’s kind of interesting. It doesn’t fly itself. And I don’t know how to do it. Even with all this time and effort and all these things I’ve read on it.

And you know, all the top AI in the world. The trick is not removing the person with the construction experience. The trick is leveraging that construction experience. Interesting, a lot to see. How do you take personal construction experience and let the machine leverage it, use it? The person now has a way to guide that machine with their experience to generate lots of solutions.

It’s not AI, artificial intelligence, but IA, intelligence augmentation. The software is being used in some of the largest projects in the world. Um, with some of the largest construction companies, uh, we’re involved in, um, You know, uh, the Edmonton rail, 2.6 billion, we’re involved in many, many large infrastructure jobs. We’re doing, I think we’ve done about $30 billion worth of construction managed through the system in the last six months alone. And it’s awesome. Yeah, it’s definitely an incredible piece of technology.

Curtis: Great, well, René, we appreciate you telling us about it and about your journey and hopefully there’s much more to come.

Ginette: As always, head to data crunch Corp com for our transcript and attribution’s.



“Loopster” Kevin MacLeod (

Licensed under Creative Commons: By Attribution 3.0 License