Skip to content

Breeding Architecture

July 12, 2011

A few weeks ago I posted about a white paper I’d read from SOM on their experiments with genetic algorithms and design.  It lead me to some further investigations into optimization and architecture.  Frankly, I have been reluctant to buy into the idea of computer optimization of architecture.  There are simply too many competing objectives to be optimized and I like to believe that there is an aesthetic component to the built environment that is important.  I say “believe” because I run into plenty of folk who feel the exact opposite: that justifying anything based on the aesthetics of form is a waste of the taxpayer’s money.  Too bad for the taxpayer.  Others have tried to quantify beauty by identifying specific characteristics that we are drawn to and arguing that the more of these found in a particular object the more beautiful it will seem to be.  Good luck with basing your art career on that, much less deriving much satisfaction from the process.

Long Arm by Opah at PicBreeder

Typically, computer optimization uses something called “directed search” and involves defining a space of problem solutions bounded by acceptable solutions and a cost metric that measures a particular solutions “goodness”.  Your goal is to find one or more solutions with a minimal cost.  Once you have a legal solution your cost function gives you a direction to go in so you just keep moving that way until you run out of legal solutions and that must be your best one.  Simple, right?  Wrong.  Even if you can model the problem accurately and define the right cost function, complex solution spaces are full of “local minima” or solutions that look good from where you are at but if you just went over that hump it turns out there is a better solution

over there.  Exploring the solution space efficiently can be very difficult to do.  Even so, designing architecture is all about adjusting the objectives or changing the constraints as the problem becomes better understood.

Butterfly by Adeleinr at PicBreeder

In steps a new area of computer optimization and search loosely termed non-directed or human-directed optimization.  Ken Stanley, of the University of Central Florida, is championing this technique with some fascinating work in creating imagery and music that is generated in part by a search algorithm and in part by human choices as to what intermediate solutions are interesting or not.  The human in the loop “quantifies” the unquantifiable value of aesthetic judgment.  At some level the algorithm has an understanding of value but that valuation is based on human input and changes as the search evolves.

One such project Stanley has produced is called “PicBreeder” in which users interact with an evolutionary algorithm to create two dimensional imagery.

Egg Wearing Hat by Robert at PicBreeder

Some rather beautifulexamples of two dimensional imagery in fact.  The human is not making the choices of how images in this program should change, they simply reward images created along the way that they find interesting.  Setting out with an initial objective or intention as to what you want to produce invariable fails.  Stanley claims that this system produces “art” while removing the “artist” from the equation.  All I can say is that statement that leads to a whole host of philosophical arguments that my inner artist strongly objects to.  But I will say that the images are great, the site is fun to explore, and there appears to be something here that has real potential to enable search in new and complex problem spaces.  Maybe even architecture.  Check it out!

2 Comments leave one →
  1. Peter Stubbs permalink
    July 12, 2011 11:13 pm

    The contrarian weighs in again!

    Which is to say that while I’m willing to allow beauty being in the eye of the beholder, the computer generated “art” displayed here looks suspiciously similar to what I used to see in High School when a few of the stoners forced to choose an elective *didn’t* go with pottery and ended up in art class with me. Which is to say, like the product of someone who doesn’t want to spend an hour working in pastels but is even less willing to repeat a grade because they failed art. Well-realized versions of it, certainly, since my pals couldn’t hold their pastels steady enough to render lines so clean as these, but of about the same level of complexity and conceptual & formal depth.

    That out of the way, I agree with your basic skepticism regarding algorithmic design. The problem I have with it is the same one that arose with the original systems-based architectures championed by academia in the ’50s and ’60s, namely that the relationship between issues and solutions isn’t of a binary enough nature to allow those systems that existed to be effective at anything but relieving designers of the burdens of decision-making.

    I’m put in mind of both Picasso & Aalto. When Aalto had a building to do he (to hear him tell it) would deeply familiarize himself with the entire program, every aspect of the building to be addressed. Then he’d put it out of his head and begin drawing abstractly, usually a few lines. Picasso was accused of “drawing like a child” by more than one critic, but his early work showed a sure and skilled hand at realistic portraits. The reason his later, more childlike work was possible was that early foundation. What do these have in common? There is far more information, a lifetime of it, going into the process than coming out, or even being consciously used. I think that the user interaction you describe may get to the surface of that, but to have a computer really participate in the process we’re going to have to reach a level of computation that I don’t see on any near horizon. (And of course, I may be very wrong, especially given your deeper engagement in the CS field.)

    • October 20, 2011 6:59 am

      I think you are right on, Peter, in that drawing aids the design process in a way that won’t happen here. I just returned from the ACADIA 2011 conference in Calgary & Banff and although it was a great conference, I was disappointed to see very little “architecture” there. Lots of digital fabrication… some beautiful, some not so much. But I think that the fact that little of the digitally fabricated forms and planarized freeform shapes are getting built isn’t because they don’t scale to full size (whoops, double negative there) but because they aren’t beautiful and elegant enough to arouse the interest and financial support required.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: