Algorithmic design techniques have proliferated in recent years, invading architecture studios at institutions around the world like a swarm of bees taking up residence inside the walls of an old house. Specifically, a number of practitioners and their students have begun exploring the potential of agent-based models to derive a plethora of complex forms, patterns, and organizations that were previously inaccessible with the use of conventional modeling tools. In addition to their capacity for producing architectural effects, the ability to work with, understand, and control these systems has opened up a vast new territory of knowledge . . .
Algorithmic design techniques have proliferated in recent years, invading architecture studios at institutions around the world like a swarm of bees taking up residence inside the walls of an old house. Specifically, a number of practitioners and their students have begun exploring the potential of agent-based models to derive a plethora of complex forms, patterns, and organizations that were previously inaccessible with the use of conventional modeling tools. In addition to their capacity for producing architectural effects, the ability to work with, understand, and control these systems has opened up a vast new territory of knowledge that will become increasingly important to the field in the coming years. While such projects and practices may be framed by some as merely a subset of formalist exercises in complex geometry, the understanding gained in working with complex multiagent systems is in fact crucial for students of architecture in the 21st century. The question of how to comprehend and predict the dynamics of higher-level orders, produced by local interactions between large numbers of individual components, remains a fundamental one in the biological sciences as well as across a number of other fields of inquiry. These self-organizing processes have far-reaching implications for fields as diverse as physics, biology, the social sciences, and, as an increasing number of projects have demonstrated, for architecture as well.
The pedagogical value of investigating such systems lies not merely in their capacity for advanced form making, but in their potential for instilling an understanding of the ways in which complex systems operate in the real world, at multiple scales, and across disciplines. The process of defining a set of rules by which agents interact with one another, comprehending the emergent macrolevel effects of these behaviors, and tuning these rules to generate a continuum of desired architectural effects represents a fundamental shift in design methodology. This workflow opens up the potential for understanding the way systems behave from the bottom up, an invaluable skill set for dealing with a vast array of challenges in contemporary design practice both architectural and nonarchitectural. From the fluctuations of global financial systems to the complex interactions that define climate models, to social networks and the spread of information or disease through populations, a robust understanding of complex systems is crucial for future generations of architects, despite whether or not their design work ends up taking the form of generative protocols. In addition, these modes of production have raised a number of questions regarding the agency of the designer in this new regime. Today’s students engaging in algorithmic design are confronted with issues of authorship and intent that are fundamentally different from those of conventional design practices. When one is tasked with designing forms of agency and devising a set of behaviors, interaction rules, and relationships, one is no longer concerned with simply producing an artifact, but with cultivating desired traits within a population. A whole new set of questions arises regarding the agency of diverse elements of geometry such as vertices, lines, and surfaces, as well as the interactions between building systems and architectural elements now imbued with a set of internal desires and behaviors. Dealing with these artificial ecosystems requires a new set of skills: knowing when and how to intervene in a system, when to let things play out, and how to select for useful traits or evolve new behaviors.
While architects have always looked to the natural world and to the sciences for inspiration, a fundamental shift has occurred in recent years as the sophistication of computational tools has steadily increased. Rather than merely being influenced by natural forms or structures, generative practices are borrowing techniques and adapting protocols drawn from the sciences toward their own ends. Today, we find a situation where the tools being developed and deployed in the design studio, especially in the realm of agent-based modeling, have begun to approach the sophistication of simulation models being used at the highest levels of scientific inquiry. Thus architecture finds itself in the position of actually producing new knowledge about the behavior of systems, despite the fact that these systems are inherently artificial and operate according to an internal logic invented by their designers. A fascination with the aesthetics of natural systems has given way to a recasting of the role of the designer, not as the creator of some fixed entity known as a building, but as the author of protocols, rules, and scenarios—that is, as a manipulator of systems and behaviors. This is the designer as developer of new architectural species through the creation of artificial ecologies; this is the space of experimental architectural practice, a speculative act that is fundamentally different from the role of experiment in the scientific method, as a means to test a hypothesis. Such experiments are not concerned with answering a question, as in science, or solving a problem, as in engineering, but rather with provoking or causing problems, and with opening up new spaces for inquiry and production. Despite a recent embrace of concepts and techniques informed by science, there is no scientific method to architectural production. Or rather, there are a multitude of methods, each probing around some web of relationships, throwing tools scrounged from wherever they can be found at the task until something coalesces. It is, in fact, the design of these methods for confronting a particular project or set of concerns that has become a fundamental task for today’s students and practitioners.
As agent-based design approaches have infiltrated further into the design studio, the initial novelty of their dynamic self-organizing properties and aesthetic appeal has clearly worn off. Likewise, in the field of collective animal behavior, the desire to construct simplistic models to explain or predict the movements of animal groups has waned in recent years. Early efforts to find universal laws underlying swarming and flocking dynamics across species have given way to a more nuanced approach examining the particular contingencies of individual-level behaviors. At present, a fundamental problem in the study of collective behavior is ascertaining which of a number of competing models, all of which may visually reproduce group-level behaviors in a somewhat convincing way, may appropriately capture the local behavior of individuals of a given species in a specific ecological context.
The agent-based models used to investigate collective behavior are generally described as self-propelled particle (SPP) models. Well-known examples of these are the model proposed by T. Vicsek et al.1 and the earlier Boids model introduced by Craig Reynolds.2 These models incorporate a simple set of nearest-neighbor interaction rules and utilize a Lagrangian approach, which follows the position of an individual particle over time, as opposed to the Eulerian approach, which analyzes the overall flow of a system at a specific position and time. While the analysis of the dynamic properties of these simple models is of great interest to statistical physics, for biologists these basic SPP models typically do not accurately reproduce the dynamics of real-world systems. Therefore, a plethora of variations on such models have been created to understand the dynamics of numerous species and systems over the years, from fish schooling3 to bird flocking,4 to the movement of human crowds.((5. D. Helbing, I. Farkas, and T. Vicsek, “Simulating dynamical features of escape panic,” Nature 407, no. 6803 (2000): 487–90.)) In order to evaluate competing SPP models and propose extensions capable of better capturing real-world dynamics, further information is needed about local interactions between individuals, which typically must be obtained experimentally, adding biological realism to a given model.
Designers working with agent-based models are in a sense working in the opposite direction, typically starting with an abstract SPP model and inventing a set of rules for how agents interact with one another and with their external environment. In the context of design, questions regarding the spatial positioning of agents are not as interesting as what those agents do when interacting with one another and how these interactions are translated into form. In both the scientific and architectural cases, a consideration of behavioral ecology, or the evolutionary drivers of individual behavior, becomes important. While classical animal behavior represented an attempt to get inside the mind of the animal, as demonstrated by the work of Niko Tinbergen, Konrad Lorenz, and other early investigators, behavioral ecology seeks to understand the evolutionary mechanisms that drive specific behaviors. Mathematical and computational models have become invaluable for this task, and current approaches to the study of behavior typically utilize a back-and-forth workflow between model and experiment.
Likewise, design experiments are crucial to the fine-tuning of agent-based models deployed in the context of architecture. Much of the recent modeling work in collective behavior has utilized evolutionary algorithms as a means to test why certain behaviors may evolve given a set of ecological constraints. Theoretical work along these lines has demonstrated the spontaneous evolution of specialization and leadership in migratory populations5 and the evolution of cooperative signaling behavior.6 More recent work in collective behavior has begun to push the boundaries between modeling and experiment, the real and the virtual, even further. In an experimental study combining an evolutionary model of virtual prey behavior with a real world predator, C. C. Ioannou et al. demonstrated the spontaneous evolution of coordinated collective motion in the virtual prey population.7 As the culture of agent-based design continues to mature, the coming years will see an increase in the use of similar evolutionary algorithms to cultivate new and unexpected forms of architectural behavior, with designers inventing and revising definitions of fitness to determine the interaction rules not only between individual agents, but also among various species of entities. Concepts drawn from ecology like competition, mutualism, sexual selection, producer-scrounger dynamics, and game theoretic interactions will begin to inform our design processes still further as the artificial ecologies we create begin to take on increasingly complex lives of their own.
1. T. Vicsek et al., “Novel Type of Phase Transition in a System of Self-Driven Particles,” Physical Review Letters 75, no. 6 (1995): 1226–229. ↩
2. Craig W. Reynolds, “Flocks, herds and schools: A distributed behavioral model,” SIGGRAPH Comput. Graph. 21, no. 4 (1987): 25–34. ↩
3. I. D. Couzin et al., “Collective Memory and Spatial Sorting in Animal Groups,” Journal of Theoretical Biology 218, no. 1 (2002): 1–11. ↩
4. C. K. Hemelrijk and H. Hildenbrandt, “Some Causes of the Variable Shape of Flocks of Birds,” PLoS ONE 6, no. 8 (2011): e22479. ↩
6. V. Guttal and I. D. Couzin, “Social interactions, information use, and the evolution of collective migration,” Proceedings of the National Academy of Sciences 107, no. 37 (2010): 16172–6177. See also C. J. Torney, S. A. Levin, and I. D. Couzin, “Specialization and evolutionary branching within migratory populations,” Proceedings of the National Academy of Sciences 107, no. 47 (2010): 20394–0399. ↩
7. C. J. Torney, A. Berdahl, and I. D. Couzin, “Signalling and the Evolution of Cooperative Foraging in Dynamic Environments,” PLoS Computational Biology 7, no. 9 (2011): e1002194. ↩
8. C. C. Ioannou, V. Guttal, and I. D. Couzin, “Predatory Fish Select for Coordinated Collective Motion in Virtual Prey,” Science 337, no. 6099 (September 2012): 1212–215. ↩