What kind of problem is a city?

I ended the first essay with a question: is a city a machine that can be tuned, or a living organism that answers every intervention in its own way?

A scene that could easily come from a real case, but which I have made up here: in a traffic control center, screens watch hundreds of junctions — vehicle counts, speeds, signal timings. At one junction, every afternoon, something odd happens: the light is green, but the right-turn lane has stopped; the cars behind it are not moving either. Every day the system puts this junction at the top of its “bottleneck” list and proposes a fix.

What the screen does not show: a care home sits on the corner, and its residents go out for a walk in the afternoon. The car turning right has to give way to a pedestrian on the crossing, whose light is green at the same time. The pedestrian phase is timed for a walking speed of about one meter per second;¹ an elderly pedestrian crosses more slowly, the car waits, and so does everyone behind it.

The system’s proposal looks reasonable: shorten the pedestrian phase, or remove the crossing at that corner and send pedestrians further along. The data isn’t wrong, but the proposal could put people who already find crossing hard into greater danger. Anyone who got up and went to the junction would see it in five minutes; when there are hundreds of junctions on a screen, nobody goes to look.

As I said in the first essay, I had thought every solution I saw at the fair in Barcelona rested on the same loop: notice, decide, respond. In this scene the sensors notice; the system supports the decision; people decide; the lights respond. Some decisions, such as signal timings, can be left to the system — but with great care. The screen is the loop’s first step. But what is it noticing? The answer depends on what kind of problem a city is.

Three kinds of problem

In 1948 the mathematician Warren Weaver divided the problems science deals with into three.²

The first is the group Weaver called “problems of simplicity”: two, at most three or four variables. Problems that can be comfortably expressed in formulas — where a billiard ball will go. According to Weaver, the telephone, the radio, the automobile, the airplane and the diesel engine grew out of solving problems of this kind.

The second is disorganized complexity: millions of parts behave at random, but the whole averages out predictably — the pressure of a gas, or an insurance company’s annual accounts. You need not know the individual molecules; the average is enough.

The third is organized complexity: a large number of factors vary in ways that depend on one another.² Living things are like this. Here averages or statistics alone are not enough, because the parts do not behave at random; they behave in relation to one another. Weaver’s example is striking: of two molecules made of the same atoms, one may be a poison while the other, its atoms arranged in a mirror-image pattern, is completely harmless.²

Jane Jacobs read the 1958 version of Weaver’s essay and in 1961 wrote a single sentence: cities, like the life sciences, are problems of organized complexity. For her, the planners’ mistake was to treat the city as one of the first two kinds — either a simple mechanism that could be managed with a few variables, or a crowd that could be computed from averages.³

Even traffic

The most “statistical”-looking side of a city is traffic. Count the cars, find the average speed, and tune the road accordingly.

In 2008 Japanese physicists ran an experiment. They asked drivers to go round a circular track at the same speed and at equal spacing. There was no accident, no narrowing lane, no signal. After a while, a small slowdown grew, and a jam formed on its own; it traveled backward like a wave. The researchers’ conclusion: a bottleneck is only a trigger of a jam, not its essential origin.⁴

So traffic arises from drivers’ small reactions to one another. Average speed does not show this. And the screen in the control center was looking at an average.

By eye, or by data?

Jacobs’s proposal is a method: to think about processes, to reason from particular cases to the general, and to look for small “unaverage” clues. Aggregate statistics, she writes, tell almost nothing about how things work in organized complexity.⁵

I see this every summer in Datça, on Turkey’s Aegean coast. Children walk down the narrow street in front of our house to the school below. Both sides of the road are full of parked cars and motorcycles; there is almost no sidewalk, and nothing to slow the cars that pick up speed downhill. Every time I pass, I am afraid. A vehicle counter would probably see no problem here: few cars, free-flowing traffic.

A little further on, on the ring road I use often, it is the opposite: the sidewalk is wide enough to fit more than a lane, but there is hardly a pedestrian on it. The road is so narrow that when one vehicle stops, one side of the road is blocked. Almost every time I drive through, I have to cross into the oncoming lane.

One has no sidewalk but has children; the other has a sidewalk but nobody walking. “Narrow the sidewalk, widen the road” is dangerous on one and might be right on the other, which you can only tell by looking at that street. In Jacobs’s words, there is no substitute for knowing the particulars.⁵

But the eye is fallible too. A few examples seen on the street can give a misleading impression of the city as a whole: when a neighborhood’s face changes and new cafés open, we assume its old residents have left; yet a study in New York found that low-income families whose neighborhoods grew more expensive did not move more often than similar families in other neighborhoods.⁶ This is where data is needed.

According to urban scientist Michael Batty, cities are, for the most part, not designed from a center; they grow from the bottom up, out of countless small, local decisions — and yet their size obeys definite scaling laws.⁷ Another research team showed that many things, from the number of patents to income to the length of electrical cable, change in a regular proportion to a city’s population.⁸

Data sees the shape of the forest, not the trees. The eye sees the tree, not the forest. Neither replaces the other.

Can the loop run the city?

Shannon Mattern goes a step further: a city is not a computer. The presumption that every meaningful flow can be sensed and measured leaves no room to think about our stickiest problems.⁹ The screen in the control center does exactly that.

In 1956, the cybernetician Ross Ashby made this into a law. Put simply, to manage a system you must be able to give sufficiently varied responses to the different states the system produces. In Ashby’s words, “only variety can destroy variety.”¹⁰ Thinking with the Barcelona loop, the “decide” step cannot choose more than the options the “respond” step has. A junction’s light has a few timing settings; the situations the junction meets are countless: the school letting out, rain, match day, the care home on the corner. Even if the system saw all of them, its response is limited to one of those few settings. What I take from this is that regulation ought to be distributed across the city rather than concentrated in a single center — across neighborhoods, institutions, people.

This is where my own thinking began to change. I used to say “the city is not a machine,” but my remedy was to measure better. The idea that if we measure better we will decide better now feels like going back to the machine through the back door.

In 1973 Horst Rittel and Melvin Webber called planning problems “wicked”: even their definition is disputed, and there is no finishing line at which they are solved.¹¹ At the same junction three things are wanted at once: that the vehicles flow (efficiency), that the care home’s residents cross safely (justice), that the neighborhood has a say over its own street (autonomy). As one rises another may fall; there is no single right setting. Who decides which comes first is a separate question. I will come back to it in this series.

So what kind of problem is a city?

What I have understood so far: a city is neither a machine to be tuned nor an organism not to be touched.

It is an order in which countless small decisions are bound to one another; one that has to be known in its particulars; one that is tended rather than solved.

That does not mean nothing can be measured. If a building’s load-bearing columns are genuinely weak, that is not an opinion but a fact. But to think we can read the whole city from a single dashboard is to mistake the kind of problem it is. The screen saw a bottleneck; a care home sat on the corner.

What I understand now is this: the loop I thought about in Barcelona can help the parts of a city — a bus line, a junction, a water pipe. But it cannot run the whole city from a single center.

If a city is that kind of problem, what exactly did the idea of a “smart city” promise? That is the next essay’s question: What did “smart” promise, and who promised it?