00:01So I'm with SOM in the Chicago office, and in many respects I think we're kindred spirits with Chris' firm.

00:10Do a lot of super tall architectural work, and the main reason I mention this is because these types of buildings…

00:19…require a tremendous amount of innovation that has spurred a lot of computational work in order to address…

00:25…especially from a structural engineering perspective, how do you actually execute these kinds of projects.

00:30So, in the architectural world in the office there's a pretty strong tradition of using computation from…

00:35…that goes a long way back, and leveraging those kinds of tools as part of the design process.

00:42We also have urban design, we have a significant urban design practice, and it's remarkable to me as sort of an…

00:47…outsider - I'm trained as an architect, not an urban designer - to see the way that the urban designers were working…

00:54…and are working still, is nowhere near as sophisticated in terms of the use of the computational tools…

01:00…I would say, compared to the architectural world.

01:03And so one of the missions that our group has is to see if we can infiltrate that and begin to introduce the designers…

01:12…to the power of these kinds of tools and this way of thinking.

01:16So I'm going to talk primarily about five different sort of domains of this sort of influence.

01:24These are those domains, and I'll just go ahead and jump right into it, starting with parametric geometry.

01:29And the reason that parametric geometry is important, in my opinion, is simply that it enables much more efficient iteration.

01:41And iteration is a fundamental part of a design process.

01:45And if you have parametric geometry, if you have the ability to command it at a parametric level…

01:50…you have the ability to iterate much more quickly through different design iterations.

01:54And, there are tools in the architectural world that have become quite common now, that architects use…

02:01…that have a parametric framework to them.

02:04The urban design group is still working primarily in tools like AutoCAD, which is not…

02:07…fundamentally a parametric environment.

02:10And so, what that means is that every time there's a change being proposed to a design that wants to be evaluated…

02:15…there's a very laborious process of having to redevelop that model from scratch, or at least the pieces…

02:20…that are going to be changing.

02:23So, we looked, and I'm not going to go very far on this because this is really sort of a, an amateur…

02:27…version of CityEngine, I would say.

02:29It was developed - this was a few years ago when we were working in this.

02:33For the purpose of even just showing to urban designers what it means to work in a parametric environment…

02:37…what sort of efficiencies are potentially gained by being able to do that.

02:41So in this case we were using a piece of software that our group uses quite a bit called Digital Project…

02:45…which is built on a product called CATIA, which is a very strong parametric environment.

02:50And what we did here simply was show them what happens when you create a rules-based model environment…

02:57…and begin to then come back and move things around and have things update on the fly and repair themselves…

03:03…be able to keep track of all the quantities and have instant feedback on all that information.

03:08The kinds of things that I believe from appearances - I haven't taken a close look at CityEngine yet…

03:15…but we definitely will be.

03:17It seems like that kind of power is definitely built into that, so I'm really hopeful that Esri is able to…

03:23…move at a very rapid pace to get that ready for this kind of activity.

03:29Hand in hand with that, then, when you get into you know this, the use of computation is the analytic…

03:35…portion of it, so once you've generated, you have the ability to generate all these parametric variations very rapidly.

03:41You need some way of being able to evaluate them, and so we are very, in Black Box, interested in...

03:47…looking at analytical tools, developing them there where we need to, and using commercial tools…

03:52…where they are available.

03:54And this is, so, in this particular case, one of the things that we're interested in is the analysis of view considerations…

04:00….which is a very sort of nebulous kind of topic for measurement.

04:05And in some aspects it's easier than others but, for example, view quality.

04:10What's a good view versus what's a bad view?

04:11That becomes a very difficult thing, or a very interesting thing, to begin to try to quantify.

04:17Working with ArcGIS - and this is really the only GIS, true GIS piece of my presentation, and it was not…

04:23…done by us, it was done by Esri, started to look at this particular project that we were involved with for a…

04:31…redevelopment project in China, which involved some sort of midscale development activity…

04:38…and then a single sort of iconic high-rise piece, and…

04:43…we were interested in beginning to try to understand what happens when you start moving the forms of…

04:49…the buildings around, reshaping the sort of overall skyline, the placement of the midrise buildings, and so forth.

04:55In terms of their ability to see the iconic tower, in terms of all of the buildings' ability to see other landmarks…

05:01…that are in the vicinity, and for the tall tower, in particular to be able to see different things in the region.

05:10So, there's a process of discretizing the geometry into a series of points that become the evaluation nodes, essentially.

05:17And, and then a process of basically rate tracing to connect point to point and determine if there's interference or not…

05:26…and then to somehow summarize that, and add it up, and turn that into kind of a heat map that gives you some…

05:31…information about where you, a more intuitive way of where you have issues related to views.

05:37So the graphics are interesting.

05:38Where we didn't get to, I'd say, where we still need to get to, is to the idea of how do you actually quantify this?

05:44So, if this represents one design scheme for this particular project and we have another design idea…

05:49…that's a completely different idea, how do you compare the two?

05:52How do you determine that overall this one is better than that one with respect to this idea of view?

05:57And one of the areas that we were interested in exploring, that I think would still be interesting to do that way…

06:02…is to look at these sort of midrise towers with respect to the view of this iconic tower, that this will become…

06:13…a very important tower in that there will be an economic impact in terms of real estate with the ability…

06:20…to see that tower or not.

06:22So spaces, you know, condominiums, office space, whatever, that has those views will get priced differently…

06:29…in the real estate market than those that might not.

06:32And does that then lead to some sort of an interesting optimization exercise where you can…

06:36…begin to move these towers around or reshape that…the contextual towers in order to maximize the real estate return…

06:46…you know, that that's the potential for that entire development.

06:49So as I said, you know, we weren't able to get down the road as far, but it's pretty clear to me…

06:53…there's a path to some pretty interesting exercises, even just with that little problem that we identified.

07:01However, again, and I'm sort of mainly addressing the Esri developers that are in the crowd today…

07:07…with this particular next thing.

07:10The view thing is something that we've been interested in Black Box for a quite a while.

07:13It didn't start just with this exercise where we engaged with Esri.

07:16And in fact we developed our own View Analysis tool a few years ago that's a plug-in to a program that Chris mentioned…

07:23…that's really common in our industry called Rhino...

07:26And this was more of a pixel-based idea, where again, and it's certainly with respect to a similar sort of idea…

07:32…where here's the subject building and there's maybe some landmark elements in the skyline as well.

07:40And the series of points are placed on the subject building and then basically we use the internal rate-tracing…

07:46…capabilities of Rhino to do the evaluation and what that, and then creating a user interface that allows us…

07:53…to control the view cone and even wait, you know, get into the idea of waiting, you know, a view that you have look…

08:01…you know, 90 degrees to one way or the other is…

08:04…maybe not as valuable as one that you're looking straight out your window, and sort of the nebulous aspects…

08:11…of view that come along with that.

08:13And then also the landmark recognition, to quickly understand on any point on the building where…

08:17…you can see a particular landmark or not.

08:20But what the output is, is this kind of qualitative information that allows you to sort of click on a point on…

08:25…any location on the first face on a building and instantly get some feedback in terms of what it looks like qualitatively.

08:32But also quantitatively to a certain extent, because it's color coded according to distance.

08:36So the buildings that are closer are darker, the buildings that are farther away are lighter…

08:40…and then the landmark buildings themselves are actually colored, in this case, a unique color, white.

08:45So it gives you some qualitative feedback as well.

08:48And…But, again going back to the idea of comparing one scheme versus another, the sort of relative…

08:54…you know, design merits of one scheme versus another, we weren't able to, we didn't take this as far yet…

08:59…to sort of quantify that at that sort of macro level.

09:03But I think View Analysis is an interesting problem.

09:06There's certainly some economic implications related to it, and I think it's something…

09:12…I think this particular approach to view analysis is actually much more effective for this type of problem…

09:16…than what we were working on currently with the Esri folks.

09:22So once you have parametric geometry that you can begin to update, you know, model ideas…

09:26…and different variations fairly quickly and you have analytical tools that you can begin to then sort of…

09:31…get numeric measurement information on…

09:34…the next sort of step I see in sort of the process is then beginning to use automated search…

09:40…optimization processes to find better solutions.

09:43So, when you have the parametric geometry and the analytical feedback you can sort of in a manual way…

09:48…sort of start guessing, well, I think if I made these adjustments, I think this would be better.

09:52Then you run it again, and see if it works or not.

09:54And what we're saying is, interested in is the idea that we can use computational power to help that process…

10:00…of searching for better solutions.

10:03And so this goes back to, again, a few years ago, we developed our own genetic algorithm.

10:10We got interested in this and wanted, frankly weren't aware of any commercial tools that were available…

10:15…to sort of serve as optimization engines, and so we did our, some research and developed…

10:21…our own genetic algorithm in order to address this problem.

10:25And for those who are not familiar, I'm going to spend just a little bit talking about what a genetic algorithm is…

10:33…in the context of how we used it, anyway.

10:37So this was the project we were asked to look at, something fairly discrete here, which was the design of the windows…

10:43…for this military academy in Kuwait.

10:46And with an objective of minimizing the direct solar gains through the windows, to minimize heat gains…

10:53…and to maximize the indirect lighting so we could minimize the indirect lighting, or the artificial lighting.

11:00So those were sort of the objective functions of the optimization exercise.

11:07And, the way that we set it up was basically, we had a basic shape, this sort of rounded rectangle shape…

11:15…was the window opening itself, and then we had this sort of diamond-shaped shroud that, well…

11:20…came out from that surface to provide shading.

11:24And, so the first thing with any sort of automated optimization exercise is just to have…

11:29…to identify a parametric control mechanism.

11:33And so in this case, it's these numbers, which are essentially the genome, we would call it.

11:39And so as you change these numbers, it automatically generates new geometry.

11:44And so in our case what the numbers refer to, just to remove the mystery, the first two digits refer to the x,y location…

11:52…of the corner points relative to the center of the window opening, so as you adjust those x,y coordinates, basically…

11:58…you're moving the diamond with respect to the center of the window opening.

12:02The second control, Bezier curves, that shape two of the edges.

12:06And then the last para control, the other two edges.

12:08So, you know, by simply modifying those numbers you can quickly generate new geometries.

12:15In a genetic algorithm, at least our sort of version of it, I call it, what you do, what happens is, it starts by…

12:22…generating a random population of window geometries.

12:25A random number of genomes.

12:28And in our case I think there was about 100 that constituted a population.

12:33And so each one of these genomes, as you can imagine, is a different window shape…

12:36…and you automate a process of sending each one of those into a piece of evaluation software.

12:42In our case it was a lighting and radiation analysis program, and when it runs a calculation…

12:48…it basically kicks out a score.

12:51How well that that particular design performed.

12:55You go through a process of sorting them, the better performers move on to the next generation.

13:03And then some of the poorer ones are basically killed off, some number of some percentage of the ones that remain…

13:10…are mutated by taking a random selection of one of their gene values and changing them to a new random value.

13:17Those move on to this next population and then the populous second generation is completed through a process…

13:23…of cross-breeding, where some of the genes of one genome are combined with some of the genes of another genome.

13:30So it, you know, matches, you know, to some extent the natural, the process of natural evolution, right?

13:36Survival of the fittest and so forth.

13:38And, so all of these new genomes now that have been created have to be tested and compared to those…

13:44…that survive from the first generation.

13:46Some perform better; some perform worse.

13:48But you get a new mix, a new sorting, and then you go through the whole process again.

13:51And over and over and over and again.

13:53And we went through I think about 250 generations of this.

13:57And what you get in the end is a series of, is a collection of very well-performing solutions.

14:04And so in this case, and we did this for the east orientation, the west, the north, and the south…

14:09…so it was a different solution for each orientation.

14:12And you end up, in this particular case, with a geometry that looks sort of like this.

14:18So, the thing about optimization though, at least most forms that I'm familiar with, is that it…

14:25…becomes difficult to evaluate, simultaneously evaluate, more than two, maybe three, different objective functions.

14:33And so in this case we're really looking at just one idea which had to do with lighting and radiation.

14:39But there's all kinds of other criteria that feed into the process, right?

14:42Like how costly is it to build this, how constructible is it, really? How does it look aesthetically?

14:47All these other things.

14:49So when people talk about optimization processes in general, my sense is that only so much of it can be sort of…

14:56…automated at a time, with these types of evolutional optimization algorithms, anyway, so…

15:04…and then we also have used the same concept, basically, to move up in scale to the building.

15:10We haven't used it at the urban scale yet, but it's clear to me that there's no reason why you can't…

15:16…as long as you can define the problem parametrically and you have an evaluation method…

15:23…that doesn't take too much time, then you can probably do that.

15:27In this case we looked at a building and the problem, and this is purely hypothetical…

15:31…we said, What happens on this site which has other buildings around it that cast shadows on it…

15:38…if we want to maximize the shape - if we want to shape the building in order to maximize the amount of…

15:42…incident solar radiation, say if you want to harvest solar energy, what would that shape of that building…

15:48…want to be under certain constraints, and under the constraints that we laid out, basically, this is what it produced.

15:55And the thing about these types of programs, what I really like is that what they can really do…

16:01…is help accelerate the process of building designer intuition, because you can't help but ask…

16:07…when you see that image, well, why did it shape it that way?

16:09What's going on there that sort of led to that, that outcome, and you begin to sort of piece it together.

16:14I mean, it's fairly obvious here, it's just trying to lean into that gap between the two buildings…

16:19…so that it can get some early afternoon sun.

16:21And the geometry at the top gets pulled in so that you get surfaces that have…

16:27…an orientation that's more normally oriented towards the sun.

16:32So, you know, without having gone through that, you might not have thought of that intuitively, and that's…

16:37…I think one of the good things that comes out of these kinds of approaches to searching for design solutions.

16:44So the next thing then is that an area that we're interested in, we're just now getting started in…

16:51…is semantic modeling at the urban scale.

16:55And so we're aware of, you know, Esri's tools and so forth that enable the construction of these kinds…

17:04…of things, and we're very interested in using them, but just like with the genetic algorithm, where we went…

17:11…through the pain of developing our own GA and we learned a ton by doing that…

17:16…a lot of very good knowledge came from that.

17:19I think that we're looking at this exercise in the same way, that we could use things out of the box, we could use…

17:24…databases that are already set up to enable sort of semantic intelligence.

17:32But, we're going to go through the process of trying to do that ourselves from scratch, just to try to get…

17:36…a better fundamental understanding of what that, what does that really mean, to have a semantic model?

17:40How, in a database, how does a computer know that a building is a building?

17:46And what does that really mean?

17:48And what is it, how do you build the relationships between all those other things within the environment of a database?

17:55So, so, I mean we started with sort of the easy part, which is sort of constructing the geometry.

18:00But I sort of liken it to what I understand, anyway, about the world of neuroscience that has happened…

18:08…in the last decade or so where, with the advent of so many new tools that allow these neuroscientists to image…

18:15…and to see patterns of movement and so forth within the brain, how that has accelerated our…

18:21…understanding of how the brain works, by being able to sort of get these windows on the brain.

18:26And my sense is that a similar thing could happen if we had similar capabilities with cities.

18:31They're so enormously complex that if we had the ability to develop some really useful windows…

18:38…and to the way things are operating and simulate that, I think there's probably potentially a lot that could be learned…

18:45…about the behavior of cities that we sort of take for granted or think we understand, that maybe we don't.

18:52So the idea is to build this model, to integrate the buildings as objects within a much larger system…

19:00…and to connect all the systems of systems together.

19:05I mean it's a hugely…I mean we probably won't get very far, I don't know, but, I mean, it's an enormously ambitious project…

19:10…to just get a fundamental understanding of what all of this means, not being software developers ourselves.

19:17So this is sort of the schematic of the idea of what we're hoping to achieve.

19:26This is where it gets down and dirty, basically, where you get into understanding the schema of the database…

19:32…and trying to piece everything together in order to create the intelligence that you're looking for.

19:41And then finally moving into what are called procedural and generative design, so.

19:46In a way the optimization stuff I think is a form of generative design, that the computer is automating processes…

19:53…of putting choices in front of you.

19:56A couple other examples.

19:58The first one is very abstract.

20:00This was done, again, quite a while ago.

20:04But it's looking at the idea of converting information into a bitmap background, so we took a pic…

20:14…have a picture of some flowers here; it doesn't matter what the picture is.

20:17And this was done definitely well before we were thinking about applications to urban design…

20:21…but it occurred to me recently that actually this could potentially have some really interesting applications.

20:27So this is something that was generated in a platform called Processing.

20:32Some of you may be familiar with that environment.

20:36And what it's doing is generating a vine here.

20:39And the way that it's generating the vine is it's reading the RGB values from the pixels in the background image.

20:46And then there are rules embedded that tell the vine what to do under various circumstances with respect…

20:51…to the values of those pixels with respect to, you know, when to turn, when to split, when to grow leaves…

20:59…and basically you know obviously just, you know, can substitute any kind of image.

21:06And the implication, obviously, it's probably pretty apparent, I mean it seems like for urban design, is thinking…

21:12…about the way that Esri is using background imagery for some of the analytical tools that they've developed.

21:19You know, is it possible to create a system of rules that might begin to sort of lay out some logic for the…

21:26…development of the design of an entire city.

21:29So, you know, I think the, as I said, this is very sort of speculative - and obviously, it looks nothing like a city…

21:36…but I think there's some interesting ideas that could be leveraged with further research there.

21:43So this is just running on another, another image.

21:47And then finally, last thing I want to show, is this project, which relates to Carl's, one of the things that…

21:59…Carl's introduced yesterday, with the idea of agent-based modeling at the urban scale.

22:03And this is something, again, it's a couple years old that we started looking…

22:08…the reason so many of these things are a couple of years old is because something happened around…

22:13…2009, 2010, which was the economic downturn.

22:17When Black Box got started we were purely an overhead research group, and so there was a lot of this…

22:23…type of research was what was conducted in that time frame.

22:26You know, when the downturn hit we had to become much more billable, frankly.

22:29And, so everybody sort of became embedded in teams doing work on real projects.

22:34And so there was a double-edged sword there.

22:36On the one hand, we became much more intimately integrated with the actual project work going on…

22:40…and that has been good; there's no doubt about it.

22:42On the flip side, we've had to sort of postpone further development of a lot of this type of research work, but…

22:48…we're actually, with the semantic model, for example, I think we're getting, you know…

22:52…moving back into some of that more speculative research work.

22:55So anyway, this project is basically a game.

23:00Like what Carl said with some of the workshop stuff, it doesn't prove anything.

23:05It was something that was of interest to us to see what would happen if we did it.

23:09And frankly, I'm not entirely sure what we even have on our hands now, once we've finished it…

23:14….but it is, I think, you know, quite interesting.

23:17So, it's a game that works sort of like this, each building is a - each building color represents a different building type…

23:24…typology, and there were five different typologies.

23:26We didn't discretely say this one's commercial, this one's residential, this one's retail, and so forth…

23:31…but basically that's what it represents.

23:33And then for each of the five typologies, each one had a desired mix of all the other typologies that it wanted to have…

23:41…in its immediate vicinity.

23:43So a Type A building wanted so much other type Type As, a certain amount of Type B, Type C, Type D, and so forth.

23:49And each building type had a very different desired mix.

23:53So if you had, for example, two buildings that were right next to each other of different, you know, desires…

23:58…there's, you're setting up a friction there, because this one wants a certain mix immediately around it…

24:02…and the other wants maybe a very different mix around it.

24:04…until it hits some threshold, and then it starts to age, its happiness starts to decline just because of age.

24:05So, you know, so it sets up that competition, and the game play allows it to sort of play out.

24:14So, have a sort of time lapse here of this.

24:21So, the way it works is, at each iteration, one new building is pushed into the system.

24:26There is a sort of a process of probabilities that's going on that decides where it's going to locate itself…

24:32…and what typology it's going to assume, and then once it gets placed it immediately surveys its…

24:39…immediate surroundings to find out who else is around it, and what type of building are they, how big are they…

24:44…how happy are they, because ultimately what we're - I forgot to mention, the whole thing is…

24:48…based on an idea of scoring happiness.

24:50So the happier the building is with its context, the more it grows.

24:54The less that it likes its context around itself, it begins to shrink.

24:58We also had an idea of age in there, where a building is not affected by age in terms of its happiness…

25:10And if it, you know, as I said, if it's unhappy long enough then it'll actually free up the site.

25:14And what you'll see in some cases is a building show up and there's not enough other stuff to make it happy…

25:18…around it and it'll actually disappear relatively quickly.

25:22And then as you get further down the line, in its history, you start to see what I think is…

25:31…beginning to look like balance.

25:34It's finding a sense of balance, and I think that's one of the interesting things between optimization exercises…

25:42…as we conduct them now, and sort of these kinds of agent-based interrogations where an agent-based environment…

25:49…is looking for stability, and I think an optimization exercise is looking for a super freak; it's looking for something…

25:55…that just stands out from the crowd completely and is not really worried about balance.