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

Although “there was a good deal of skepticism” among local economists and planners “that this could be done because of the inherent complexity of a model like that,” according to Zehnder, the project moved forward anyway. Over the next year, IBM worked with Zehnder’s office and the local experts to develop the map of equations and an arsenal of historical data that would power the simulation. With the help of San Francisco-based Forio, a developer of business simulations, IBM began to weave a spiderweb of relationships that quickly ballooned to over seven thousand equations (a number that was deemed too complex), was pruned back to six hundred (too simple), and then eventually built back up to the roughly three thousand contained in the final revision.Given the suspect track record of system dynamics in cities, IBM’s decision to bring cybernetics back to urban planning was less reckless than it at first appears. As Zehnder described how the model was constructed, with a series of workshops and iterative designs, it appeared to be a vast improvement over Forrester’s process— which seems to have taken place mostly behind the closed doors of his laboratory after a cursory round of interviews with ex-mayor John Collins’s buddies. Although it’s not clear whether Cook was aware of the criticisms of Urban Dynamics before the project began, local experts raised those old concerns immediately. But, as Cook told me, the context for building and using systems models of cities had changed dramatically: “Now you can actually take a model like this and put a web interface on it and let people interact directly with the tool and even change some of the assumptions that are in it. That was pretty powerful.” In defense of system dynamics itself, the method “is very explicit about the relationships,” he convincingly argues, “instead of being a black box where people can’t see the logic. We thought this was especially important for working with cities and the constituencies that they could see into the guts of this and make sense of it.”

In the end, however, the Portland model, like Forrester’s, had little impact on policy. Unlike Forrester’s model, which spat out absurd contradictions that actually did stimulate debate, IBM’s predictions in Portland were reliably dull. Its greatest revelation, much ballyhooed in the company’s public relations campaign for the project, was a strong correlation between the adoption of pro-bicycle municipal policies and a decline in obesity. But no one in bike-obsessed Portland needed three thousand equations to know that. When I asked Zehnder what role, if any, the model played in the planning process, his response indicated that it was largely a sideshow. “It proved ... to be something where we weren’t really going to be able to maintain or use it—in a way that people were going to have confidence in—to illustrate these relationships.” But as Cook explained, and Zehnder concurs, the real benefit of building the model was teaching people that cities are “systems of systems,” to use a phrase Colin Harrison has advanced to explain IBM’s approach to the complexity of smart cities. As Zehnder explained, the result was “an increased awareness that, like all cities, [Portland] operates in silos,” a bureaucratic term for government departments that don’t cooperate effectively. “Their decisions affect other parts of the city.

After Portland, Cook turned the software over to another business unit within IBM to market it to other cities. At the time we spoke in late 2012, there were still no takers. The challenge for models like these in the future will invariably lie in better balancing the value gained (which is still too small) with the effort required by the city to maintain and operate it (still too high).

While IBM’s efforts in Portland may have avoided the kind of devastating consequences that resulted from the first wave of systems models of cities in the 1960s— partly because IBM built the model in a more responsible manner, and partly because the planners chose to ignore it—the project has again raised important, lingering questions about the value of computer simulations of cities.

Michael Batty founded and directs the Centre for Advanced Spatial Analysis at University College London, one of the world’s leading centers for urban modeling. Over a career that began at the University of Manchester in 1966, Batty has advanced the science of simulating cities through its dark ages, connecting those ambitious, failed early efforts with today’s more modestly successful ones. In a 2011 article, “Building a Science of Cities,” he explains the limits of systems models, and why they were abandoned in the first place. Systems models like Forrester’s, Batty argues, “treated cities as organised from the top down, distinct from their wider environment which was assumed largely benign, with their functioning dependent on restoring their equilibrium through various negative feedbacks of which planning was central.” Forrester’s methods for analyzing systems assumed a closed loop—everything that mattered to the system’s behavior was contained in the equations. There was no external environment, or at least not one that mattered. And it largely saw the process of change as a shift from one steady state, or equilibrium, to another, in response to some directed action. “As soon as this model was articulated, it was found wanting,” Batty counters. “Cities do not exist in benign environments and cannot be easily closed from the wider world, they do not automatically return to equilibrium for they are forever changing, indeed they are far-from-equilibrium. Nor are they centrally ordered but evolve mainly from the bottom up as the products of millions of individual and group decisions with only occasional top down centralised action.” In the decades since Forrester, the science of complex systems had taken a 180-degree turn. Mechanical metaphors had been replaced by biological ones, grand design by evolutionary processes, closed loops by open fields of influence. Hammering home the point, Batty concludes, “What has been realised in the last 50 years, is that this notion of systems freely adjusting to changed conditions is no longer valid, in fact it never was” Ecologists had long ago discarded the notion of stability in living systems. But this central tenet of cybernetics, equilibrium, remains firmly embedded in the popular imagination of how human and natural systems behave.

We can only hope that IBM and other would-be urban system modelers will learn from the missteps of cybernetics redux in Portland. Despite theoretical flaws and practical failures, Forrester and his disciples never gave up hope that their methods would one day revolutionize social science and policy analysis. More than twenty years after the publication of Urban Dynamics was met with harsh criticism, an unrepentant Forrester proclaimed the universality of systems; he lamented in a 1991 speech that “There is an unwillingness to accept the idea that families, corporations, and governments belong to the same general class of dynamic structures as do chemical refineries and autopilots for aircraft.”84 If there were shortcomings to systems models of cities, his disciple Louis Alfeld argued in 1995 they were “limited detail and limited resources ... [which] can be overcome by new hardware and software technology.”

In the meantime, a range of modeling techniques has supplanted system dynamics in urban research, including many Batty has helped develop. They are showing promise where system dynamics failed. Where system models tried to replicate macrolevel behavior, new techniques such as agent-based simulation use fast parallel- processing computers to simulate the minute interactions between individuals (or “agents”) at the microlevel iteratively over time, and calculate the aggregate impact of millions of simultaneous actions. One of the largest such models, developed at ETH Zurich, one of Europe’s leading technical universities, in 2004 successfully replicated the actual traffic patterns of Switzerland’s 7.2 million inhabitants. And unlike Forrester’s static equations, each individual agent can learn and adapt to changing conditions, such as congestion, from one cycle to the next just like real people.