Creative AI: Algorithms and robot craftsmen open new possibilities in architecture
Computers have transformed architecture in remarkable ways. They've made it possible to visualize designs in fully-rendered 3D graphics and to automatically check designs against building codes and other standard specifications. And they've made designs possible that were unthinkable or unimaginable 50 years ago, as they can crunch the numbers on complex equations and even generate plans or models from high-level requirements. Architecture, like music, art, games, and written stories can be created algorithmically.
One architect exploring this idea is Michael Hansmeyer, who is currently a visiting professor at Southeast University in Nanjing, China. He sees architecture as being at an inflection point whereby the maturation of computation and fabrication technologies means that we're entering an era where the formerly impossible is now doable and the unimaginable is taking form.
Complexity is no longer an impediment but rather an opportunity, he writes on his website. And we see this embodied in his work, which is full of fine lines and curves and intricate patterns and shapes that look almost alien.
Most famous among his projects is Digital Grotesque, a 2013 collaboration he did with fellow computational architect Benjamin Dillenburger. To make their design, the pair wrote a program that uses a subdivision algorithm to divide the surface of each column into four smaller surfaces, each slightly different in texture (though not randomly so), and then divide those surfaces on and on to ever-smaller surfaces. In the process, the form of the structure morphed into something elaborate and alien yet also clearly rooted in geometry. You can see a video showcasing its production and the final result below.
The complexity of the result wowed people who saw it in person. Hansmeyer tells Gizmag that people insisted on touching the columns, despite signs asking them not to. He finds this exciting, as it shows how algorithms can so easily encode different scales and levels of information into architecture.
Asked whether a human could have designed Digital Grotesque via a traditional method, Hansmeyer jokes that "it's difficult to draw sections of a column with sixteen million facets using a traditional pen or a mouse," though he concedes that it's theoretically possible. More interesting, he notes, is that algorithmic design such as this is three-dimensional to begin with – it needs no intermediary 2D representation, as is the norm in architectural design.
The interior of Digital Grotesque's grotto is a sight to behold (Photo: Michael Hansmeyer/Benjamin Dillenburger)
That's not to say that algorithms can't be used in designing 2D representations of space. Take Stanford University researchers Paul Merrell, Eric Schkufza and Vladlen Koltun's program that generates detailed floor plans from a set of high-level requirements. But the strength of computer-generated and computer-aided design (CAD) is its power to cut through the cruft of abstraction to get straight to the most direct representation of the final product, and in the case of architecture, at least, that's 3D.
In design and architecture, most research into algorithms and computational creativity has focused on this kind of basic AI – tools that support the design process by generating endless streams of possibilities by mutating existing designs or that separate the artistic aspects from the functional ones.
The latter is what the Institute of Artificial Art Amsterdam (IAAA) specializes in. It automates the design process according to a "visual grammar" (or some other grammar, as the situation dictates). "In one concrete example," says the IAAA's Jos de Bruin, "I designed an AutoCAD extension that took the global ideas of an architect for the design of a penthouse and turned these into an algorithm for generating variants instantiating this style."
University of Sydney Design Computing lecturer Rob Saunders develops computational models of creativity. He's into curious agents that explore possibility spaces – intrinsically motivated by code that "rewards" them for learning new things – and creative systems that drive change in social or physical environments.
We previously wrote about one of his projects, Accomplice, which embedded robots into the wall of a gallery. Their curiosity led to them knocking holes in the wall so that they could observe the outside world, and this added to their mental models of how the world reacts to them – leading to further destructive transformations of its environment.
Saunders' work typically skews more into the conceptual realms of design, but he tells Gizmag that he's beginning to work with architects on marrying the technology behind these creative robotic systems with some of the fundamentals of architecture – namely those that relate directly to materials.
"Robots are great at repetitive tasks and working with materials that react reliably," Saunders explains. "So when you're working with high-grade timber or high-density foam or high-quality stone of one form or another, it basically will always react the same way. What we're interested in doing is trying to develop robots that are capable of learning how to work with materials that work in non-linear ways. Some simple things like working with hot wax or expanding foam or more practically with low-grade building materials like low-grade timber."
The benefit of this is two-fold: robots that can manipulate low-grade building materials in construction could be sent to remote areas to work with local material, which will be low grade, and they will be able to produce buildings and structures that are unique – because they must adapt some generic design to meet the constraints of what is at hand.
"It may seem a long way away from the composing of a sonnet or the painting of a picture," notes Saunders, "but the idea of having to make do in the environment – that you've got to change that environment in some ways – is a sort of practical form of creativity that I'm very excited to explore."
"I think this is something that has the real potential to make a difference," he continues. "And you could argue, 'well this is just [straight, non-creative] artificial intelligence,' but to me this gets to the core of creativity – to work within constraints or to challenge those constraints by developing a competence that's situated in the environment in which you're working, or in this case in which a robot would be working."
These machines would in essence be craftsmen, only applied on an industrial scale. Gramazio Fabio and Matthias Kohler at ETH Zürich's Department of Architecture are among the pioneers in this area. They are looking to develop fabrication processes that produce unique environments as cheaply as the Industrial Age model of recreating the same unit innumerable times (which we currently see applied most commonly in low-cost and community housing projects). They've already developed robots that can stack bricks and other reliable, uniform materials to form curved and unusual designs by a generative process.
Gramazio Kohler Research at ETH Zurich built a robotic system that can recognize its own position and respond autonomously to the surroundings and its components while building a wall (Photo: Gramazio Kohler Research, ETH Zurich)
Saunders is looking to bring his computational creativity background into this kind of work, to have the robots get greater control over the process such that they can, like in the Accomplice project – only now constructive rather than destructive – become masters of working with a material by exploring the possibility space, analyzing the results, and learning from their mistakes.
Right now this research is at a very early stage. The current task is to develop a robot that can stack irregular rocks to form a wall – whether straight or circular or some other desired shape. Further along, they'll move onto other materials and other industrial robots, and Saunders notes that there's a potential for knowledge to be transferred between these far more easily than with a traditional master and apprentice relationship. "We could literally download the program from one robot, put it into another, and have it continue on – possibly with a slightly different material – and build on the knowledge of the previous one," he says.
Where this gets exciting is in thinking about the resulting craft. Saunders refers to the e-David painting robot we wrote about previously, which mimics human painting techniques. He loves that project, but he believes that it would be even better to have a robot that learns its craft "in a way that's entirely dependent on the fact that it is a large industrial robot." Humans should play to their strengths, and robots to theirs, and in this way Saunders foresees robot stonemasons and other craftsbots that could use their greater talents for precision and strength to accomplish feats unique to them, perhaps working from basic skeleton plans designed by a human.
There are too many buildings for an architect to produce distinct designs, but algorithms can – and increasingly will – overcome the pitfalls of uniformity that result from the need for many buildings built very quickly at low cost. The question is how, and we've seen here two broad approaches that may both prove common: on the one hand there's the matter of algorithmic architecture, whereby computers generate designs from some kind of high-level brief, and on the other there are creative agents – our robotic craftsmen – that work directly with the building materials to meet some high-level goals.
In a quirky twist of fate, modern technology – with its IKEAs and factory-assembled buildings – may well bring a futuristic spin to the old ways. Soon every house, apartment block, office tower, chair, and table will be unique, whether by design or by craft.