One may build models for many reasons: to predict, to explain, to guide policy, to reconstruct historical events, to challenge prevailing assumptions, and explore the consequences of contrasting ones, even to clarify what one’s assumptions truly are! In their scale, models can range from small “Toy” simulations meant to crudely capture core dynamics to high resolution models at the national, or even planetary, scale. CAM scholars have built models at many scales for many purposes. A selection, with movies of the models in action, is offered below, with brief descriptions.
Published in 1996 by Epstein and Robert Axtell, Sugarscape is among the best known, and most widely replicated, agent-based models. It posits a spatial landscape of “sugar,” an idealized resource that agents need to eat in order to survive and reproduce. Sites differ in the maximum sugar level. When eaten, sugar grows back at unit rate, until capacity is reached. Shown is an initial landscape with all sites at capacity: there are two high capacity sugar mountains (dark yellow) separated by sugar lowlands and deserts (light yellow and white). Agents come into the world with a vision range and a metabolism. Their basic rule is: Look at all sites within your vision. Pick the unoccupied site with the most sugar. Go there and eat it. Their sugar stock is increased by this amount, and then decreased by their metabolic rate. If the result is negative, they die; otherwise they go again. This rule, with reproduction and inheritance, proves sufficient to generate power law wealth distributions characteristic of human societies, and a surprising range of other social phenomena. A typical run is animated below. To explore the basic model interactively, the NetLogo version is recommended http://ccl.northwestern.edu/netlogo/ . In the course of their book, Epstein and Axtell explore a great variety of variations, including combat, trade, cultural evolution, demography, epidemiology, genetic evolution, and combinations of these, in the first elaborate “artificial society.”
The Artificial Anasazi
Sugarscape principles were employed by Epstein and a diverse team of archaeologists, climatologists, soil scientists, and, ethnographers, from the Santa Fe Institute, the Brookings Institution, and Tree Ring Laboratory at Arizona in the first computational reconstruction of an ancient civilization, the Kayenta Anasazi of Longhouse Valley. Using the true environmental history—maize potential, hydrology—and simple rules for selection of household locations and farming sites, the model successfully generated a millennium of social history, and has helped understand the environmental basis for the enigmatic disappearance of this civilization from Longhouse Valley in the mid-14rth Century.
The Smallpox Model
This model was designed to develop novel vaccination strategies in the event of a smallpox outbreak. We used a “toy” two-town model with commuting. Shown are Square Town and Circle Town, each with one hundred, two parent two child, households. Agents go back and forth to work, school, and the common hospital, the prime social units and transmission points evident in a large European data set. The Base Case movie shows no interventions.
We showed that pre-emptive (pre-outbreak) vaccination of Hospital workers, and reactive (upon diagnosis) vaccination of infecteds and (merely) their immediate households was powerful in suppressing spread. The strategy offered an alternative to mass vaccination (with the risk of adverse reactions to the vaccine) and vaccination of broad social contacts(trace vaccination), which is infeasible in modern urban settings.
The National Scale Agent Model
Unlike the Smallpox Model, the US Large-Scale Agent Model (LSAM), developed by CAM’s Senior Software Engineer, Jon Parker, is very high fidelity model. It includes 300 million distinct agents whose distributions match the US Census data on age, gender, household size, and other categories. It includes all workplaces, schools, and hospitals and all routine travel throughout the 30,000 zip codes of the US. It is used in the NIH MIDAS Project to evaluate containment strategies for pandemic flu, and to study health behaviors and their effects on epidemic dynamic. Below is a typical run, beginning with a swine flu case in Los Angeles. Agents colored black are healthy and susceptible; reds are infected, and blues have had the disease. The LSAM is novel in its computational architecture and unprecedented in its speed, attributes that earned it the NTSA’s Outstanding Achievement Award in 2009.
The Global-Scale Agent Model
Under Parker’s direction, CAM possesses the only planetary scale agent-based infectious disease model. The Global-Scale Agent Model (GSAM) runs 6.5 billion distinct individuals on a map of the planet, with all intercontinental travel included. The model has been featured in Nature, and published in TOMACS, the journal of record for advanced computing. The coloring scheme is as before, with a pandemic emerging in Asia.
As noted under our methods, CAM has pioneered the synthesis of important fields. One such exercise tracks a behaviorally rich agent population as it reacts to a toxic airborne contaminant in the 3D replica of Metropolitan Los Angeles. This hybrid of agent based modeling, computational fluid dynamics, and environmental health is an example of collaborations between the Center (School of Medicine), and other Schools of the University, in this case the Bloomberg School of Public health and the Whiting School of Engineering. Traffic is color-coded for velocity for yellow (highest) to purple (lowest). We use the model for evacuation design and catastrophic event preparedness.