Anthony M. Townsend - Smart Cities - Big Data, Civic Hackers, and the Quest for a New Utopia

Like the psychohistorians, the urban modelers of the 1960s had a maddening habit of relying on approximations, a practice that had devastating consequences in New York. As Joe Flood describes in his 2010 book The Fires, facing a rising wave of blazes and union demands for more resources, in 1969 New York City fire chief John O’Hagan turned to the New York City-RAND Institute, a partnership with the think tank that Mayor John Lindsay had formed little more than a year earlier. It was a bold attempt to apply cybernetic thinking to the operations of local government—as Lindsay described it, the “introduction into city agencies of the kind of streamlined, modern management” that defense secretary “Robert McNamara applied in the Pentagon with such success in the past seven years.” Focusing on just a single measure of fire company performance, response time, RAND developed a computer model of the city’s firefighting system.61 Despite the RAND analysts’ own misgivings about the usefulness of response time, it was the easiest indicator to measure reliably, and was less variable and therefore simpler to model. As Flood explains, “RAND made a fateful choice: gather the response-time data, model it to the best of their abilities, and put their concerns about response time’s shortcomings to the side.”

The assumptions, and the distortions they created, compounded from there. RAND’s model also assumed that fire companies were always available to respond from their firehouse, which in actuality was “a rarity in places like the Bronx, where every company in a neighborhood, sometimes in the entire borough, could be out fighting fires at the same time,” Flood explains. Another half-witted shortcut left out the paralyzing impact of gridlock; “in the most congested city, traffic played no role in response times, rigs able to cruise through Midtown Manhattan at rush hour at the same speed as through Queens at midnight.”63 Politics distorted the model too, without much pushback from its designers. As RAND’s Rae Archibald told Flood, “If the models came back saying one thing” and fire commissioner John O’Hagan “didn’t like it, he would make you run it again and check, run it again and check.”64 During a wave of budget cuts in 1971, RAND’s model counterintuitively recommended shuttering several of the busiest fire companies in the city, based solely on its calculations of response times.63 The resulting closures were concentrated in poor areas of the city; the demands on remaining fire companies soared, and the Bronx (and several other neighborhoods) burned. Flood puts the tally of persons displaced by the fires at more than a half-million.66

By the mid-1970s, in every domain of urban planning and management to which computer modeling had been applied—generic system models like Forester’s, land use and transportation models like Pittsburgh’s, and even relatively narrowly focused operational models like the one built by RAND for the New York City Fire Department—serious doubts about its effectiveness had been raised. By the mid-1970s, planning scholars moved swiftly away from their earlier embrace of such all- encompassing, predictive city simulators. In 1973 Douglass Lee’s “Requiem for Large-Scale Urban Models” sounded their death knell in the pages of the Journal of the American Institute of Planners. Then a professor of urban planning at the University of California, Berkeley (today he still works on models for the US Department of Transportation’s Volpe National Transportation Systems Center), Lee had studied the Pittsburgh model up close while working there. The article was a scathing indictment, calling out “seven sins” of large-scale models—hypercomprehensiveness, grossness, hungriness, wrongheadedness, complicatedness, mechanicalness, and expensiveness. But Lee reserved his most searing commentary for Forrester, the MIT professor who “buries what is a simplistic conception of the housing market in a somewhat obtuse model... then claims that the problem cannot be understood without the irrelevant complexity.” While the Pittsburgh modelers had dumbed their model down to make it tractable, Forrester had embellished his to make it look more sophisticated.

City planners relegated cybernetics and system dynamics to the doghouse for the better part of thirty years. The Urban Systems Laboratory at MIT closed doors in 1974 for lack of funding. Louis Edward Alfeld, who directed Forresters urban research in the early 1970s, wrote in 1995, “The past twenty-five years have not treated urban dynamics kindly.... It has become a curiosity, a relic of the past that few have heard of and most dismiss.”69 The same year, in a retrospective on “Requiem,” Lee noted that “modeling is mostly a cottage industry, not much different from what it was ten or twenty years ago. Despite upheavals in planning and the massive changes in computing technology, the role of [large-scale urban models] remains unresolved. That [they] are alive and well may be fine for the modelers, but is it of consequence to anyone else?”70

System modelers were cast out from the city in the early 1970 s, but the discipline blossomed in the private sector, where it was effective at tackling the analysis of less complex systems than an entire city. As it turned out, their exile would not be permanent.

The year 2011 witnessed cybernetics redux when IBM resurrected urban dynamics and installed it in Portland, Oregon, a city of some half-million people. While the simulation efforts of the 1960s had to cope with severely limited computers and data-collection capabilities, with virtually limitless processing power and vast stores of digitized data at its disposal, IBM developed a computer model of Portland that dwarfed Forresters. “System Dynamics for Smarter Cities,” as the apparatus was blithely named, wove together more than three thousand equations. Forresters had used just 118 (only 42 of which, a subsequent analysis determined, really shaped the 77 results). On a website used to interact with the model, diagrams reminiscent of those in Urban Dynamics dissected the city into a spaghetti-like tangle of interacting variables. It was as if someone sauntered into an IBM lab, dropped off a copy of the moldering book, and said, “Give me one of these.” And in so doing ignored forty years of painstaking learning and progress in urban modeling and simulation.

Where IBM’s Deep Thunder simulation in Rio predicted rainfall up to forty-eight hours in advance, the one it built for Portland, grinding on ten years of historical data, was meant to project years into the future and inform long-term planning (much as Pittsburgh’s model in the early 1960s was meant to inform a master plan for 1980). Planners could ask the program questions by toggling different controls. “How would transportation policy investments affect K-12 education? How would parks and land-use decisions effect greenhouse gases?” explained Joe Zehnder, Portland’s chief city planner at the time.74 The software would spit out predictions in response. IBM touted it as a “decision support system,” a tool to help policy makers explore the ripple effects of different options, and the interdependencies of different systems in the city.

The idea to resurrect urban dynamics came from Justin Cook, an IBM strategist who himself was a graduate of the Sloan School where Forrester had once taught. By 2009 IBM had accumulated a deep reservoir of systems modeling knowledge from its work with industry. Cook saw an opportunity to apply it to the company’s new Smarter Cities initiative. Looking for a pilot, he said, “I decided that Portland might make a very good candidate ... they were at the very beginning stages of working out a twenty-five-year plan.” Late in 2009, he approached Mayor Sam Adams, a leading advocate of sustainable urbanism, with a proposal for what was not to be a traditional consulting engagement, but rather what Cook described as a “joint research project.”