• Each member has been created to react individually to his neighbor's actions and the ever-shifting dynamic. The environment, too, has been made as realistic as possible, from the way the fire burns to the types of materials the buildings are made of.

    Credit: Paul M. Torrens

    Each member has been created to react individually to his neighbor's actions and the ever-shifting dynamic. The environment, too, has been made as realistic as possible, from the way the fire burns to the types of materials the buildings are made of.

Materials can be tested, methods of fabrication can be tried, and even destructive natural forces can be simulated, but human crowds have remained largely unpredictable. Until recently.

A growing body of research is attempting to demystify the nature of crowds to predict what they may do in extreme or calamitous situations. Complicating this effort has been the fact that large groups cannot be tested empirically. No right-minded scientist would force an unsuspecting crowd to evacuate a sports arena or incite a riot simply to study people's behavior. And rehearsing such scenarios with willing participants would never yield truly accurate responses.

So researchers are turning to the computer to model such events. One of the emerging leaders is Paul M. Torrens, 33, an assistant professor at Arizona State University's School of Geographical Sciences. Although he has a Ph.D. in geography, Torrens' particular field blends that discipline with urban studies and computer science.

Crowds have been modeled before. But usually they have been conceived of as basic, generalized aggregations or with participants acting as identical entities—not unlike molecules of gas bouncing around an enclosure. 

Torrens' model is different: Individuals are individuals. Torrens calls the approach an “agent-based methodology,” in which the agents are capable of processing what's going on around them. Thus the virtual crowds—like the real ones on which he's basing his model—are made up of individuals endowed with human behavioral traits.

  • In a virtual scenario created by researcher Paul Torrens to learn how actual crowds work, a suited throng runs from an explosion in the street.

    Credit: Paul M. Torrens

    In a virtual scenario created by researcher Paul Torrens to learn how actual crowds work, a suited throng runs from an explosion in the street.

“We already know a lot from past research in behavioral geography and spatial cognition,” Torrens says. So he picks up where previous scientists left off by equipping subjects (including himself) with sensors that detect body movements and reactions during different scenarios—a crowd running down the street, say, or an angry mob. Then, starting with a basic behavioral template, this motion-capture data and other behavioral algorithms are used to add on higher-level behaviors to create individualized agents—as many as might be needed for a particular scenario—each with its own “flavor.”

“The agents will employ these behaviors very heterogeneously,” Torrens says, “as the surroundings for each agent are always going to be unique.” The modeled individuals, aggregated into a group, will then make the sorts of decisions that make crowds unpredictable. Rendered three-dimensionally, the model is immediately easy to understand.

While previous models provided only cross-sectional snapshots of particular moments, Torrens builds his with time in mind. Working on the scale of 1/60th of a second—the reaction time for human movement, Torrens notes—his models can run dynamic situations that reflect close-to-real-time scenarios.

The uses are manifold. Architects and engineers can watch buildings evacuate, city planners can observe congestion, and law enforcement officers can study how crowds turn into riots—all previously untestable events.

Convinced that architects, town planners, and city officials will benefit from his work, Torrens is hoping to model an entire city over the next five years. Citing Phoenix's explosive exurban growth, urban redevelopments, and unlikely environmental conditions, Torrens calls the city “a perfect lab to study urban dynamics.” A $400,000 grant from the National Science Foundation will help to get him going in that direction.

  • The green cloud hanging over this digital version of Salt Lake City isn't severe weather-it's the shape of the Wi-Fi environment. The cloud's coloring and height represent signal strength in decibels relative to 1 milliwatt (dBmW), ranging from -51 dBmW at the low end (dark green) to -14 dBmW at the high end (pale green).

    Credit: PAUL M. TORRENS

    The green cloud hanging over this digital version of Salt Lake City isn't severe weather-it's the shape of the Wi-Fi environment. The cloud's coloring and height represent signal strength in decibels relative to 1 milliwatt (dBmW), ranging from -51 dBmW at the low end (dark green) to -14 dBmW at the high end (pale green).

And Torrens' work also brings an intangible, yet significant, contribution to better understanding urban settings. “I once saw a 3-D model of London, with all of the city's buildings precisely represented,” he says. “But it was a ghost town. It wasn't really London.” If Torrens has anything to do with it, cities will one day be modeled the way they actually are: as incredibly dynamic sites of human interaction, not simply as a collection of buildings.

John Gendall is a freelance writer based in New York City.

WI-FI GEOGRAPHY

As if modeling crowds wasn't enough, Torrens is also setting out to map the signals generated by wireless networks, which over the past decade have flooded urban areas with a new kind of invisible activity. Torrens hopes the effort will help urbanists and geographers understand the effect of wireless use on cities.

“This is particularly interesting for geographers,” he says, “because [Wi-Fi is] changing the nature of interaction, but it's completely invisible.” As such networks increasingly become the mode of communication, and of community, Torrens aims to reveal the geography behind them. While this realm typically has been understood as difficult to characterize or categorize, he notes that it has become the site of real interaction. Torrens cites the architectural model of space vs. place, where “space” is understood broadly and generally and where “place” is space made personal. Following this logic, he believes “cyberspace” should more accurately be understood as “cyberplace” and should thus be studied geographically. To do this, Torrens has developed his own technology (patent pending) to survey Wi-Fi access points and detect such things as transmission density, network security, and public vs. pay-per-use coverage.