The Bizarre, Bony-Looking Future of Algorithmic Design
Pick a building on the horizon—any building will do—and consider for a moment how it came to be. Long before construction began, there was an architect and a blueprint. The blueprint bore a design that came from an idea the architect had about how the building should look. The process by which an architect’s idea becomes a blueprint that becomes a building is an example of explicit design. It’s more or less how we’ve always built not just buildings, but the majority of the physical objects that surround us.
Jordan Brandt, Autodesk’s resident futurist, makes a clear distinction between explicit design and generative design. Explicit design is when “you have an idea in your head and you draw it,” he says. “Generative design is when you state the goals of your problem and have the computer create design iterations for you.”
With generative design, a designer begins with an objective or set of objectives—the desired energy consumption for a building, for example, or the amount of sunlight a room should receive—and then lets algorithms take the reins on drafting solutions. This might be a big ask for designers, because when we build something, whether it’s a skyscraper or a trash can, “we have a preconceived notion of how it looks,” says Brandt. But a machine—Autodesk’s software, in this instance—is an unbiased agent. “[It’s] simply looking to optimize the criteria we set forward,” says Brandt.
Unencumbered by preconceptions, a generative algorithm can run wild, spitting out thousands of possible solutions, the performance of which it can assess, digitally, after the fact.
Consider the prototype pictured here. Autodesk made it for Lightning, an electric motorcycle company out of San Carlos, California known for making the fastest production bike on the market. Lightning wants its vehicles outfitted with the lightest, more durable parts possible. That’s where Autodesk comes in: With the company’s Dreamcatcher software, it can craft a swingarm—the piece that hinges the rear wheel to the bike’s frame—by starting with weight and strength as the beginning design elements, rather than a predetermined shape of metal. Once the designer has provided this set of constraints to Dreamcatcher, the software starts cranking out potential solutions.
One of the most enthralling aspects of generative design is that it churns out objects that look incredibly organic—almost as though they sprouted from the earth, rather than a string of a code. Take that swingarm prototype, for instance. “It looks like bone,” Brandt says. Specifically, it looks like a cat pelvis—Autodesk’s CEO actually found a real cat pelvis at a bone shop in Berkeley, CA, and it looks an awful lot like the Lightning swingarm. Brandt finds this unsurprising: “If you think about the function of the two they’re actually quite similar,” he says. “Just like nature has optimized for weight and improved stiffness, so too have these algorithms.” Like biology, he says, generative design is about trying things out and seeing what works. Crucially, however, it does so over the course of a few hours, as opposed to a few million years. “We’re essentially running accelerated artificial evolution,” says Brandt.
Generative design principles have given rise to more than motorcycle parts. Back in 2006, Dutch designer Joris Laarman partnered with carmaker Opel on a limited-run series of chairs. Laarman used Opel’s proprietary software to digitally fabricate and optimize the metal chair’s design, resulting in what he named the Bone Chair. The webbed structure of tibia- and femur-like chair legs was an early example of what might happen if you let an algorithm do the designing: It’s strongest where it needs to bear weight, has open spaces where a support structure would be unnecessary, and doesn’t look anything like your standard dining room table chair. In 2009, a collaboration between MOS Architecture firm and designer George Michael Brower yielded the mosCells table. The MOS team came to Brower with an unusual request: to create a computer application that could spit out design solutions for a table made from a given building unit (a flat circle supported by its sides, folded underneath.) The resulting metal table looks like leaves, or lily pads on a pond.
Medical implants stand to improve from generative design solutions. With Autodesk’s Within Medical software, biomedical engineers are designing and 3-D printing surgical implants that are more organically structured and perform better in the body. These parts typically have tiny lattice-like structures that can attach and allow porous bone structures to remodel and heal. Common applications thus far include implants for skull and facial reconstruction and hip-joint and vertebrae replacements.
Lattice structures in medical implants aren’t new, points out Scott Hollister, a professor of biomedical engineering at University of Michigan. What’s novel is that a software company is pursuing ways to design even more sophisticated micro-lattice structures to provide better implant options for doctors and patients. “If you think about a basic analogy, where tissue growth is like flowing fluid,” Hollister says, “having a porous structure with optimized lattice structures makes it easier for the liquid to flow through than it would for a random structure.” The more an implant can be precisely tailored to a bone’s pore structure, the more easily the body can heal around it. This kind of optimization should be applied to materials, too, Hollister says. Titanium facial implants can erode the soft skin tissues in the face, so a polymer material, for example, is better suited to that type of surgery. “There’s no silver bullet in terms of materials,” he says, “but the ability to make these lattice structure in combination with these different materials will push medical implants ahead.”
That’s the medical industry. In the auto and aerospace industries, they can save fuel and amp up performance. Mark Davis, senior director of design research at Autodesk, says that’s the real value proposition. The software can detect minuscule areas in need of improvement that, spread over the manufacturing of thousands of cars, can affect change. “You keep the structural integrity of a part, and its functional design, and you take material away where it’s not needed,” he says. An easy example is a bracket. By whittling the weight of a bracket used in cars down by 20 percent, you decrease weight over the whole system, at scale.
Brandt says the next challenge for generative designers will be materials. Right now, software can determine what the optimum weight, size, or structure of a certain object should be. Autodesk wants to fine tune that even more, to architect materials on a micro-scale. “It used to be a designer would say, 60/60 aluminum, or a specific alloy, and then you design the shape around the properties of the material,” he says. Creating permutations of the material could raise broad questions such as metal or plastic, or it could get as granular as determining the exact density of the microns in a given design. And as for generative designed buildings? “I think in the next decade we will see it. It’s not going to be a 100-story high rise, but we’ll start seeing smaller structures,” Brandt says.
If this version of the future—one where algorithms are designers—comes with a dystopian ring, it’s worth considering all the ways these design tools free up time for creators. By just speeding up the process by which we create motorcycle parts or knee joint implants, designers can more carefully consider with design objectives are most important and which could potentially perform better. “With explicit building designs, you might go through one or two per week,” Brandt says. “With the cloud, you can go through hundreds per week. And what that shows is there’s an innate human nature to create.”
Read original article: