Computational designers in architecture (and grasshopper dilettantes such as myself) love to (over)use genetic algorithms in everyday work. Genetic algorithms (or GAs, as the cool kids call them) are a particularly fancy method for optimization that work as a kind of analogy to the genetic process in real life. The parameters you’re optimizing for get put into a kind of simulated chromosome and then a series of generated genotypes slowly evolve into something that more closely fits the solution you’re looking for, with simulated crossover and mutation to help make sure you’re getting closer to a global optimum than a local one.

For those that don’t regularly optimize (I know I should more often, but it’s so much easier to just sit on the couch and vegetate), the imagery that gets used is of a “fitness landscape” where you’re looking for the highest peak or the lowest valley, which represents the best solution to a problem:

Often, however, you’re talking about a landscape with many dimensions, with lots of local optima – the lower peaks in the image above – that can seem like the best solution when they are not. Searching the entire problem domain takes far too long, and so this is when your optimization algorithm takes over – it uses a series of rules to try and search intelligently for the best (or at least close to the best) solution.

Genetic algorithms tend to get used a lot by architecture students, despite being fairly controversial in mathematics. A lot of this controversy has to do with how good GAs are at finding the actual optimal solution in a certain amount of time, versus selecting or writing another algorithm that works more well on that specific kind of problem. I think they are popular for the following reasons:

A. Design problems often involve dozens of variables, and GAs are well-suited to complex fitness landscapes.

B. Architects are often more interested in a very good solution rather than the best, and GAs can find a very good solution with very quick setup.

C. Architects are less interested in the runtime of the algorithm, so “faster” methods that aren’t quite as robust or optimal are not necessary.

D. GAs work well on a wide range of problems, which architects interpret to mean “work great on all problems” for the reasons above.

E. “Genetic algorithm” sounds really cool, particularly in a presentation or lecture.

F. Grasshopper has a built in GA solver called Galapagos that works really well and looks cool.

As it goes, I don’t really have problem with any of this. Architects are not mathematicians, so it’s great that we have a fairly robust, flexible method for optimization that can get us better solutions than we can think up by staring at the screen. I’m not going to complain.

Where I was trying to get with all of this is that when I read the post today on BLDGBLOG one of the things that popped into my head eventually is that a lot of what design does is similar to a genetic algorithm. Bear with me here…

The post linked above shows two projects by Lik San Chan that, loosely defined, are architectural narrative pieces about food and space. They are really interesting pieces of speculative urbanism, and you should check them out. However, in Geoff’s write up he mentions

…it’s one thing to create, analyze, or even editorially promote architectural projects as narrative ideas—that is, as scenario plans for future landscapes—but it’s another thing to look at whether or not such proposals do, in fact, operate successfully as solutions to the problems they highlight.

To which I responded in the comments,

Geoff, I love your blog, and I really really enjoy diving into projects such as these, but I have to say that I’m skeptical that using the harsh light of quantitative analysis on either of these proposals would really be a good idea. The tempelhof project has some obvious sociopolitical issues (as well as some probable agricultural ones), but even the smaller-scale fish market seems to have more to do with the Bartlett aesthetic and a romantic fictionalization of urbanism and light industry than it does with a real proposal for social change. You’ve argued very eloquently in the past for a place for architectural fiction. Suggesting that such fiction should be held up to rational standards would, I fear, not be kind to the work presented.

Don’t get me wrong: I would like to see both of the above projects re-imagined as real proposals. However, such an exercise would most likely involve an interdisciplinary team, a lot of effort, and the result would be much less likely to lend itself to viral blog dissemination.

As is usual in blog commenting, immediately after clicking “post” I thought of another thing I wanted to say to clarify what I had just written. I’m not going to double-post on someone else’s blog, however, as that would identify me correctly as a long-winded blowhard. However, I have my own forum for long-winded blowhardery here, so here it goes:

I can’t help but feel that a lot of design that takes a fictional or narrative slant is valuable precisely because it doesn’t attempt to quantitatively define the solution from the outset. To use an analogy to optimization and genetic algorithms (that I have thoughtfully provided exposition for above), the kind of freewheeling, speculative work that you see on BLDGBLOG provides a kind of free-associative mutation that help find globally optimal solutions to global problems (see what I did there). By definition, a lot of these projects are going to be completely nonrational and impossible, because if rationality was the prime directive we’d simply end up with a solution that was almost identical to the status quo.

It is absolutely vital to design that we simultaneously promote both speculative/narrative and optimal/performative approaches to architecture, as they are both vitally necessary to exploring the incredibly complex domain of reality.

…also known as math doodling.

I would like to point out that the majority of computationally minded architects (myself included) really don’t move that far beyond the surface presented in the videos linked above, or as Robert Woodbury would put it, “visual construction trumps mathematical inference for design.” Many thesis projects are really just (slightly) more complex and better-drawn versions of the infinite elephants.

But more seriously, this kind of “math doodling” (which is, if you watch the videos, obviously self-deprecating) should be seen as a valid way analogue for the way mathematics are explored in architectural design. I’m not saying that the deeper implications of the form are non-important, but to a designer they are only important in that they may point the way to a better or different implementation of the concept at play. What I find most energizing about Vi Hart’s videos is that they suggest that not only is this kind of visual understanding more fun, but that it is an end unto itself – you don’t need to sketch snakes just as a hook to understanding graph and knot theory (and how they interrelate) but also because it’s a valid use of the math itself. It’s non-abstract and non-arbitrary, which is pretty damn cool for math.

Wow, six months away from BIM and Autodesk drops something awesome while I’m not looking. In case you’re as clueless as I am, the labs released a free demo of their new project Vasari for preview and experimentation. Vasari is basically an enhanced, lightweight version of the conceptual massing environment in Revit, with some energy and carbon analysis tools built in. Think of it as Revit’s version of Sketchup. What this description belies, however, is the fact that all of the CME tools are there for you to play with, including reporting parameters and adaptive components. In addition, models created in Vasari play well with the full version of Revit. Which means one hell of a fun toy.

Someone at Autodesk must be whispering to the right people. A design process, whether digital or physical, thrives in an environment that is responsive, quick, flexible, and robust. Feature depth and breadth are fantastic, but when sketching a long feature list can make things complex and laggy, which can hurt more than help. Not to mention that in a larger office many of the people that will be doing conceptual massing won’t even care about half the categories or tools in Revit. Part of the reason that Rhino + pluginofchoice is a preferred tool in architecture schools across the country is that it is omnivorous, fast, and simple out of the box, with an enormous library of tools you can add on demand. It is very good at getting out of the way when you just need to get something done. Now, Revit will never be Rhino (and Rhino will never be Revit) as BIM has demands for object and model rigor that most surface modelers short of CATIA can’t really handle. But having tools like Vasari, with a more directed (and playful) environment go a long way towards making BIM more accessible to a design mindset and methodology.

Below is a demo video from labs themsel(ves?) The audio sounds a bit like the Home Shopping Network to me, but you’ll get a pretty good idea of what they’re getting at.