Architecture Gets an Upgrade: Generative Design Expands Design Possibilities

This piece is basically a white paper that appeared as a magazine article. It was written for Bentley Systems, the world’s largest privately held software firm, and maker of—in my (widely shared) opinion—the world’s best engineering design software. Bentley used this article as part of a campaign to establish thought leadership in generative design, an exciting new architectural movement. As of Wednesday; June 30, 2010, this article (which appeared in Cadalyst magazine) appeared near the top of the 1st page of results for the a Google search on “generative design”.

Generative Design ArticleWriter Jim Paul has remarked that only with the invention of the catapult did engineers begin to attain preeminence in world affairs, for it was the use of the catapult that viscerally demonstrated, for the first time, humans controlling mechanical forces many times more powerful than the mere human body. It feels as if some similar shift is now happening in architecture, as a new class of software tools—collectively termed ‘generative design’—allow humans to control intellectual forces many times more powerful than human minds to design and construct buildings that could not otherwise exist. This is not the first time, of course, that humans have harnessed vast computational powers for practical ends, but something about walking around inside the results brings home the magnitude of what is happening.

Humans are fond of impressive buildings. China’s 2008 Olympics, for example, will probably be remembered as much for two spectacular buildings—the Beijing National Stadium and the Beijing National Aquatic Center—as for the many worlds records set. Better known, respectively, as the Bird’s Nest and the Water Cube, the design, engineering and construction of both buildings depended heavily on generative design. By seizing the world stage so dramatically and successfully, the two buildings raised the profile of generative design (also termed ‘computational’ or ‘parametric’ design) and, to some extent, architecture will never go back.

“Generative design is not about designing a building,” says Lars Hesselgren, “it’s about designing the system that designs a building.” Hesselgren is Director of Research at Kohn Pedersen Fox Associates (KPF) and also a cofounder of the SmartGeometry Group, a collective that is taking the lead in bringing generative design concepts to actual architectural projects, mainly via a series of groundbreaking conferences. By bringing together working architects, academics, and Bentley Systems, a commercial software developer, SmartGeometry has raised the bar for generative design in actual practice, influenced some of the best minds now entering architecture, and fostered an unusually close collaboration with industry to produce GenerativeComponents, software from Bentley that is, at present, the lingua franca of the generative design movement.

Generative Design: a necessarily inadequate overview

Generative design exponents (mostly) agree: it’s not that architects are now using computers to do things that couldn’t be done previously. Rather, computers are enabling architects to take on design tasks that would otherwise be inconceivably tedious. “It’s like fractals,” says Hesselgren, “Ada Lovelace knew something of them in the early 1800s, but to work them out by hand and obtain significant results would have taken her several lifetimes.” Similarly, computational techniques exist to work out, say, the interwoven façade of the Beijing National Stadium, but to resolve a particular iteration with conventional CAD techniques would take so much time that, at best, just one or two versions would have been considered. “But using parametrics,” says J Parrish, “I was able to investigate far more alternatives. We built version 34, because it was better… but version one would have worked fine. Generative design allowed us to get better results in a fraction of the time.”

Parrish, also a cofounder of SmartGeometry, is the director of ArupSport, the division of Arup at the leading edge of large venue design—ArupSport provided engineering and design services for both of Beijing’s major Olympic venues, the National Stadium and the National Aquatic Center. He’s a proponent of the ‘rapid prototyping’ face of generative design, and began working with Microsoft’s Visual Basic and Excel—”a useful interface”—and Bentley’s MicroStation in the 1970s to quickly design stadiums. By tweaking any of 3-4,000 parameters, Parrish was able to quickly compare, for example, 50,000-seat positioning schemes, optimizing them for views, distance from the field, and other factors. “I could do in a morning what used to take me a month,” he says. Combined with other new technologies, such as real-time rendering and 3D printing, parametrically-enabled rapid prototyping amounts to a new way of doing architectural design—arriving at optimum solutions via side-by-side comparisons of multiple, slightly different versions, rather than depending on the ‘flashlight in the dark’ of one designer’s genius.

A complementary approach is also possible: rather than using multiple versions to decide what’s best based on comparison, architects can instead use computing power to find structural solutions that are self-organizing, that is, not decided on by an individual but arrived at by ‘genetic algorithms’ that iteratively apply relatively simple rules. The Water Cube’s soap bubble-like structure is an example of this approach. Per the book Spacecraft: Developments in Architectural Computing, “Denis Weare and Robert Phelan, Irish physicists, were able to calculate that the most efficient way to divide a space into cells of equal volume while minimizing the surface area between them was to use a stacked arrangement composed of 75 percent 14-sided shapes and 25 percent 12-sided shapes.” So far so good, but since the resulting structure would have 22,000 steel members connected at 12,000 nodes, generating an actual model based on the idea exceeded the reach of conventional design. Instead,

“In order to manipulate this complex geometric system dynamically, Arup wrote parametric software that automated the drawing and analysis process. Based on specified design constraints and under 190 different loading scenarios, the algorithm iteratively checked the distribution of forces through the entire structure, based on specific member sizes, allowing the team to test different design configurations and receive feedback within 25 minutes. The result is a spectacular building with a sophisticated structure that is optimized in terms material weight to strength, achieved with relative ease.” [emphasis added]

In addition to the structural advantages, Arup estimates that $10 million was saved on design costs alone, compared to traditional methods.

Jenny Sabin, a co-director of Sabin+Jones LabStudio, a lecturer at University of Pennsylvania, and a longtime tutor at SmartGeometry conferences, offers her take on generative design. “I think architecture is inherently about relationships; I’m interested in the way small local relationships and simple parameters give rise to complex scenarios—bottom up design, as opposed to top down. Generative design gives rise to emergent behaviors and forms.” By ‘bottom up’ and ‘emergent’, Sabin is alluding to the natural processes that create complex structure such as termite mounds: without benefit of a designer,  simple, rule-applying units like termites interact with basic materials—in this case, soil—to create complexly ordered structure.

Sabin uses generative design in fields only tangentially related to architecture. For example, LabStudio is investigating the way surrounding cellular structure affects breast cancer growth, and finds that modeling cell growth patterns with computational techniques is yielding real insight into cancer formation. “Generative design allows us to navigate and generate complexity,” she says, “in a way that technology hasn’t achieved before. Algorithmic design is not necessarily new, but the tools we now have for approaching bottom up design are far more sophisticated… it’s almost as if the software itself has become a new material to work with.”

Generative design is hard to label—even Hesselgren says, “I’m not entirely happy with any of the terms currently being used.” Generative design is even harder to do. There are few guides available, and the software tools are currently the special domain of what Parrish calls, affectionately, “the geeky sort who can get them to work.” But this is changing rapidly, due mainly to the work of the SmartGeometry Group.

What’s SmartGeometry?

In an article written for the online magazine Core.Form-ula, Hesselgren had this to say about the origins of the SmartGeometry Group:

“I got together with some friends who all had an interest in complex geometry in architecture and we together created the SmartGeometry Group. Hugh Whitehead was and is the head of Foster+Partners specialist geometry group; he had just finished organizing the London ‘Gherkin’ geometry. J Parrish, head of ArupSport, had developed a vast spreadsheet which he alleges allowed him to design a 100,000 seat stadium in a morning. Robert Aish, a computer scientist, was working for Bentley Systems researching object-oriented programming. I myself at KPF was just starting to contribute to the ‘Pinnacle’ tower in the City of London. So each of us had tools and techniques and a gut feeling that ‘programming architecture’ was going to be a powerful source of new inspiration for architects.”

This was in 2001, and the group’s first conference was held in Cambridge, UK, in 2003 and every year since except for 2005. The 2009 conference was in San Francisco, California. “We just picked the right time,” says Parrish, “it’s actually difficult to start this kind of organization at the wrong time. But with the passion of the founders, and the mutually beneficial relationship with Bentley, the rest is history.”

The ‘object-oriented programming’ that Aish was developing eventually became the commercial software GenerativeComponents, but at the time it was the equivalent of a ‘concept car’: powerful and fast—and flashy—but also unstable and changing daily. At the SmartGeometry conferences, Bentley provided computers and software to work with, tutors, and other resources (such as a room full of 3D printers for prototyping) and the SmartGeometry directors used their contacts in academia and architecture to provide smart people pursuing interesting projects. In other words, a match made in heaven.

“Cross pollinating has been good for us,” says Huw Roberts, Bentley’s director of global marketing, “as our trainers and programmers get feedback on the beta versions, they can change the tool as it’s being used and sometimes the equivalent of a couple of versions will emerge during a conference.” Moreover, the tool is changing in response to user needs, and when you give smart people tools tailored to their needs, fascinating work emerges. The SmartGeometry conferences are an interesting scene, full of the exhausted exuberance that comes when mental limits are pushed and success comes at three o’clock in the morning. “I think I gained as much as anybody,” says Parrish, “As we worked with these very bright people, parametric design started to hit the mainstream, and stopped being a tool just for geeks—people could do things without fighting the software.”

Parrish says that the maturing of the software is changing SmartGeometry and the practice of generative design: “We have gone from not really having the tools, to a stage where a capable design system has parametric capabilities that more people can use. The conferences have to reflect that. In the first years they were places for like-minded people to get over hurdles and make things work. Now we can focus more on the actual problems, and attracting the best designers, not just the people who are best at working with software. So next year we can focus more on the end product, and not judge on the simple fact of getting something to work.”

In other words, just as the bright people at conferences forced the software tools to evolve quickly, the maturing tools and new wealth of knowledge is forcing SmartGeometry and the larger community to evolve. And all of this is happening in ‘real-time’; changes are occurring so fast that even those who think they know the field are continually amazed at what is now being done.

The future’s so bright…

Hesselgren makes an interesting point about the use of computers in architecture: “There is a big difference between designing architecture, as opposed to recording it properly. Things like BIM [Building Information Modeling] are important, but they’re concerned with recording design decisions. They don’t talk to you. SmartGeometry points out computers haven’t done much but record until recently, and that the ability to record is not the most important thing we do.” Parrish adds, “What we’re doing is not really new. In the ‘70s, I saw people doing parametric design with spreadsheets—they could work out lifts, 2D plans, etc. But with the SmartGeometry Group and now commercial software like GenerativeComponents, the work doesn’t die when the people creating it move on. Other people can build on it, and the returns are substantial.”

This iterative, emergent process—people designing better tools to take on interesting projects, and then conceiving more interesting projects to take advantage of the better tools—is at once new, and quintessentially human. By building and using better design tools, and creating a culture that gets the most out of these tools, the generative design movement is rapidly becoming humanity’s most potent tool for evolving our built environment.

# # #


Angus W. Stocking

Angus W. Stocking, L.S., is a licensed land surveyor who now prepares information marketing content for the infrastructure industry.


Angus W. Stocking, L.S.
P.O. Box 872
Paonia CO 81428
270.363.0033 office