As described by Christopher Alexander in “Systems Generating Systems,” we can only recognize ‘systems as a whole’ by virtue of some observable, holistic property that emerges from the unpredictable interactions between constituent elements of and behaviors within the system. By simulating systems, one can draw causal relationships between low-level behaviors and high-level patterns. In fact, simulations, as philosopher Manuel DeLanda argues, give conceptual legitimacy to the notion that observable complex phenomena can emerge from interactions between the elements of a system at all. In Philosophy and Simulation: The Emergence of Synthetic Reason, he writes, “Simulations can play the role of laboratory experiments in the study of emergence complementing the role of mathematics in deciphering the structure of possibility spaces.”1 In this view, simulating is a necessary partner to generative work, as it provides a method for designers to explore and draw conclusions about the relationship between the outputs of generating systems and their constitutive rules and elements.
In a canonical example from 1971,2 economist Thomas Schelling created a simplified model of cities as a 2x2 grid, with individual pieces representing citizens. In Schelling’s model, a citizen of a certain ‘type’ will move around on the grid until their neighbors are composed of a minimum percentage of their own ‘type.’ With two ‘types’ in play, black and white, the model demonstrated that a certain phenomenon would inevitably arise — the board (or city) tending toward starkly divided areas (or neighborhoods) defined by ‘type’ (or race). The parentheticals denote the findings of Schelling’s model: Systems-level effects such as racial segregation might arise from the actions and interactions of individual actors regardless of their intent. This model (which, astonishingly compared to capabilities today, actually was implemented hundreds of times using physical pieces on graph paper) helped to set the stage for a key computational technique in simulating emergence: Agent-based modeling (ABM). Computer scientist and ABM researcher Uri Wilensky defines it as “a form of computational modeling whereby a phenomenon is modeled in terms of agents and their interactions.”3 This dry definition belies the real potential of agent-based modeling. One of my interview subjects, Max, an artificial intelligence researcher, describes the values of ABM in contrast to other approaches:
“Most machine learning is oriented towards producing an answer of some kind... Agent-based modeling is more about understanding how all of the different components of a system interact. So it’s less about the final state that the simulation produces, but how it got there, and why it got there.”4
Agents are individual actors in a simulation, with behavioral rules that drive their actions within an environment. ABM has been used to model such diverse phenomena as urban form,5 the spread of infectious diseases,6 and labor markets.7 Agents do not necessarily represent individual humans, however. David, a PhD student in a technology-oriented architecture program, described a project he worked on to use ABM in an urban design context. The agents, in this case, are building footprints:
“I created a prototype to take maps of a city, recognize the buildings, and make the buildings behave as a physical entity in a physical simulation environment… The interesting part is that, even with such a simple behavior — they’re reflex agents — you start to find out really interesting clusters and patterns of how you can map the city. Even with no intelligence on the side of the agent.”8
David underscores one of the most compelling aspects of agent-based modeling — individual agents typically have very limited awareness of the system beyond their immediate ‘neighborhood’ (defined by the designer). However, the rules for how they behave in relation to each other and to their environment almost always result in emergent patterns at the scale of large groups of agents. Emergence is unpredictable, and opens a new, productive line of inquiry into the mechanics of and relationships between individual agent behavior and systemic patterns.
Another interview subject, Ken, a technologist and organizer, describes an experimental project to simulate house parties. In this case, agents represent individual attendees to a party, with behaviors that include socializing, eating, drinking, and going to the bathroom. In calibrating the agents and the simulation, Ken and his collaborators ran into some unexpected emergent patterns: “The design process wasn’t form-based, it was system-based, so it’s kind of like trying to tweak this system that’s going off the rails. We had problems like, ‘People are going to the bathroom constantly in this simulation! We can’t stop them!’”9 Although this is a humorous example, it’s also indicative of the capability of the emergent properties that arise to surprise the simulation designers. An awareness of emergence as a phenomenon can aid in using simulations as a research tool. Ken explains how, as a result of his experience with ABM, he works differently now: “Maybe one of the ways in which agent-based simulation has affected my thinking is that I see inside, mentally, in physical spaces, all the varied agents doing whatever they would want.” He encodes behaviors at the level of individual agents, but has gained the ability to imagine systemic behaviors that arise from the agents’ interactions in simulation space. Closely tied to systems thinking and socio-technical complexity, exploring emergence through ABM is a potent technique for designers, architects, and technologists.
1. DeLanda, Manuel. Philosophy and simulation: The emergence of synthetic reason. Bloomsbury Publishing, 2011. ↩
2. Schelling, Thomas C. "Dynamic models of segregation." Journal of mathematical sociology 1, no. 2 (1971): 143-186. ↩
3. Wilensky, Uri, and William Rand. An introduction to agent-based modeling: Modeling natural, social, and engineered complex systems with NetLogo. MIT Press, 2015. ↩
4. Phone interview with Max, August 2, 2017. ↩
5. Batty, Michael. Cities and complexity: Understanding cities with cellular automata, agent-based models, and fractals. The MIT press, 2007. ↩
6. Perez, Liliana, and Suzana Dragicevic. "An agent-based approach for modeling dynamics of contagious disease spread." International journal of health geographics 8, no. 1 (2009): 50. ↩
7. Neugart, Michael, and Matteo Richiardi. "Agent-based models of the labor market." LABORatorio R. Revelli working papers series 125, 2012. ↩
8. Skype interview with David, July 24, 2017. ↩
9. Skype interview with Ken, July 17, 2017. ↩