Here Comes Everybody

by Clay Shirky

  The third kind of loss is the most serious. Networked organizations are more resilient as a result of better communications tools and more flexible social structures, but this is as true of terrorist networks or criminal gangs as of Wikipedians or student protesters. This third loss, where the harms are not merely transitional, leads to a hard question: What are we going to do about the negative effects of freedom? It’s easy to tell the newspaper people to quit whining, because the writing has been on the wall since the internet became publicly accessible in the early 1990s—their response has been inadequate in part because they waited so long to grapple with the change. It’s harder, though, to say what we should be doing about Pro-Ana kids or about newly robust criminal networks.

  It used to be hard to get people to assemble and easy for existing groups to fall apart. Now assembling latent groups is simple, and the groups, once assembled, can be quite robust in the face of indifference or even direct opposition from the larger society. (In some cases, that very opposition can strengthen the group’s cohesion, as with the Pro-Ana girls.) When it is hard to form groups, both potentially good and bad groups are prevented from forming; when it becomes simple to form groups, we get both the good and bad ones. This is going to force society to shift from simply preventing groups from forming to actively deciding which existing ones to try to oppose, a shift that parallels the publish-then-filter pattern generally.

  CHAPTER 9

  FITTING OUR TOOLS TO A SMALL WORLD

  Large social groups are different from small ones, but we are still understanding all the ways in which that is true. Recent innovations in social tools provide more explicit support for a pattern of social networking called the Small World pattern, which underlies the idea of Six Degrees of Separation.

  Imagine you are seated next to someone on a plane, and after a brief conversation you realize you have a friend or acquaintance in common. At this point you are both required to express surprise at this discovery, and one of you may even make the canonical remark: “What a small world!” After all, what are the chances that the two of you would know someone in common?

  The surprising answer is that the chances are actually quite good, for reasons having to do with the structure of social networks. Consider the most basic form of the problem. Given two people drawn at random from a population of six billion, each of you would have to know something like sixty thousand people to have a fifty-fifty chance of knowing someone in common. Even getting to a one-in-ten chance would require that each of you know twenty-five thousand people. Most of us don’t know tens of thousands of people, and yet we discover these small-world connections all the time. How is that possible?

  The first factor is something called “homophily,” or the grouping of like with like. The percentage of the world that rides in airplanes is small, so you are not, by definition, drawn at random from a pool of six billion, you’re drawn from a much smaller one. You have at least two other things in common as well (other than both being seated in row nine), and that is your departure and arrival cities, increasing the likelihood that people you know live in a town your seat mate visits and vice versa. The choices you both have made about where to live and work increase the chance that your friends and acquaintances will share a contact.

  Now consider your friends. You are probably moderately well connected—neither as social as Paris Hilton nor as reclusive as J. D. Salinger. (This says nothing about you personally—most people fall between those extremes, by definition.) And most of the people you know are (again by definition) in a similarly middling position. It is therefore tempting to assume that everyone is roughly average, but this assumption is wrong (for the same reason that “the average” is meaningless in power law distributions). Assuming that a social network is held together by its average members leads us to underestimate seriously the likelihood of sharing a link with someone we meet. In fact, social networks are held together not by the bulk of people with hundreds of connections but by the few people with tens of thousands.

  Consider the list of people you know. You are unlikely to know many recluses, since recluses by definition don’t have many contacts. At the other end of the spectrum, you are very likely to know one or more highly connected people, since to be highly connected in the first place, they have to know many people like you. The chance that you are a highly connected person is low, just as it is for everyone, but the chance that you know one is high. And the “knowing someone in common” link—the thing that makes you exclaim “Small world!” with your seat mate—is specifically about that kind of connection. When you are trying to find a link with someone else, you are unlikely to know any given contact of theirs, as we would expect in a sparsely connected environment. But you are very likely to know one of the most connected people they know. It is the presence of these highly connected people that forms the backbone of the social networks.

  All this seems like common sense, but it wasn’t until 1998 that anyone offered a convincing explanation of the pattern. Prior to that year sociologists understood that social networks somehow manage to be sparsely connected (most people have only a moderate number of connections), but that despite this sparseness the networks are both efficient (any two people are connected together by only a few links—the Six Degrees of Separation pattern) and robust (the loss of a random connection, or even several, doesn’t destroy the network). What they didn’t understand was how those networks were held together.

  In 1998 Duncan Watts and Steve Strogatz published their research on a pattern they dubbed the “Small World network.” Small World networks have two characteristics that, when balanced properly, let messages move through the network effectively. The first is that small groups are densely connected. In a small group the best pattern of communication is that everyone connects with everyone. In a group of friends Alice knows Bob, Carol, Doria, and Eunice, and each of them knows the others. In a cluster of five people there would be ten connections (by Birthday Paradox math), so each person could communicate directly with any of the others. If anyone drops out of the group, temporarily or permanently, none of the other links between people would be disrupted. (This highly connected pattern appears, among other places, in tightly connected clusters of friends using social networks like MySpace and Facebook, or weblogging platforms like LiveJournal and Xanga.)

  The second characteristic of Small World networks is that large groups are sparsely connected. A larger collection of people—one that ran from Alice to Zephyr, say—would have many more potential connections. As the size of your network grew, your small group pattern, where everyone connected to everyone, would become first impractical, then unbuildable. By the time you wanted to connect five thousand people (not even the size of a small town) you’d need half a million connections (Birthday Paradox math again). On the other hand, if you let everyone continue to maintain a handful of connections, then as the network grew, any two people pulled at random would have a long chain of links between them, far longer than six links, in fact. Such a network would be unusable, since the people in it would hardly be connected together.

  Figure 9-1: Two ways of connecting ten people. The left-hand network shows everyone connected to everyone else, which quickly becomes too dense to scale to even moderate numbers of people. The right-hand network keeps people connected but maintains a sparser network.

  So what do you do? You adopt both strategies—dense and sparse connections—at different scales. You let the small groups connect tightly, and then you connect the groups. But you can’t really connect groups—you connect people within groups. Instead of one loose group of twenty-five, you have five tight groups of five. As long as a couple of people in each small group know a couple of people in other groups, you get the advantages of tight connection at the small scale and loose connection at the large scale. The network will be sparse but efficient and robust.

  A Small World network cheats nature by providing a better-than-random trade-off between the number of links required t
o connect a network, and that network’s effectiveness in relaying messages. It occupies a sweet spot between the unbuildable and the unusable, and as a side effect, it is highly resistant to random damage, since the average person doesn’t perform a critical function. (By contrast, in a hierarchy almost everyone is critical, since the loss of any one person’s connections disrupts communication to everyone connected through that person.) A handful of people are extremely critical to holding the whole network together, because as the network grows large, the existence of a small number of highly connected individuals enables the very trade-off between connectivity and effectiveness that makes the Small World pattern work in the first place.

  When you list the participants in a Small World network in rank order by number of connections, the resulting graph approximates a power law distribution: a few people account for a wildly disproportionate amount of the overall connectivity. Malcolm Gladwell, in The Tipping Point, calls these people Connectors; they function like ambassadors, creating links between disparate populations in larger networks. Without these people large social networks would indeed face the trade-off between impractical and useless. With them everyone is connected to everyone else in six degrees of separation.

  Figure 9-2: A network of dense clusters. The network has many fewer connections than it would if everyone were connected to everyone, but it still puts everyone three degrees at most from everyone else. Note that some nodes assume disproportionate importance in holding together the whole.

  So far this is all straight sociology—Watts and Strogatz discovered a pattern common to modern societies, though the connections within and between the smaller groups vary. (Some societies are more tribal than others, with denser local connections and sparser global ones.) What’s happening now is that we have tools that both support and extend these patterns. Most Meetup members are members of only one group, but in any large town there are a few people who are members of lots of groups. Meetup is a Small World network, as is MySpace. (Among hundreds of millions of users, the average number of friends is less than sixty, while the median number is five, exactly the kind of disproportion we would expect.) Weblogs also exhibit the pattern—the most connected weblogs are thousands of times more connected than ordinary weblogs are, while ordinary weblogs, with a few readers, are far more likely to be part of a densely connected cluster.

  As an example of the way social tools can both rely on and extend the Small World pattern, consider dodgeball, a social networking service designed for mobile phone users, invented by Dennis Crowley and Alex Rainert (both former students of mine). Late last September I found myself with an uncharacteristically free evening, so I decided to stop in at the Magician, a bar on the Lower East Side that some of my friends frequent. Before I arrived, I sent a text message from my phone to dodgeball. The message I sent was simplicity itself: “@magician.” The dodgeball service recognized The Magician as a bar (it was previously entered into its database) and recognized me as a registered user. So it was able to send text messages out to my other dodgeball-using friends telling them where I was. They all received messages on their phones saying, “Your friend Clay is at The Magician on Rivington Street.”

  It also did something more complicated. Because every dodgeball user has a list of friends, dodgeball knows not only that Dennis is a friend of Clay’s but that Andy is a friend of Dennis’s, who is a friend of Clay’s. This is friend-of-a-friend networking (sometimes known as FOAF networking), and it’s how social networks like MySpace and Facebook work. But because dodgeball also knows something about the geographic location of its users, and because digital cameras are ubiquitous in the connected crowd that dodgeball targets, it’s able to use FOAF networking to broker introductions.

  So minutes after I checked in with dodgeball, I got a message back from the service saying, “Andy Krucoff is also at the Magician. You know Andy through Dennis.” The message was accompanied by a digital picture of Krucoff. It was small and grainy, but given the uncanny human ability to recognize faces (much of our brain’s visual processing apparatus is given over to face recognition), it was enough for me to locate him, even in dim light. Seeing him, I walked up, held out my hand, and said, “I’m Clay. If Dennis were here, he’d introduce us.” My meeting Krucoff was simultaneously less social than if no technology had been involved (Dennis, our mutual friend, was nowhere to be seen) and more so (without dodgeball I wouldn’t have been able to meet Krucoff at all, even though we were standing ten feet apart). Dodgeball used FOAF networking to take a latent link (in this case, between me and Krucoff running through Dennis) and make it real, or rather it gave me the information I needed to make it real. When I introduced myself, both Krucoff’s network and mine got one link denser, and a lot of people I know got one hop closer to him, and vice versa.

  The software didn’t actually introduce us; it simply provided me the tool to make the introduction myself. Because the number of people you could know at any given moment is always a tiny fraction of the people you do know, social tools have to help us decide when to actually make a connection. As a result, tools that rely on FOAF networking work better when they augment human social choices rather than trying to replace them. Hundreds of tools build on social networking, from Cyworld (the Korean megasite, with pictorial representations of users) to asmallworld (an intentionally exclusive community for the highly connected and well-off) to Dogster (for dog owners). All make the same underlying assumptions about human links, and all play in some way with the tension between homophily and the desire to meet new people.

  Once you’ve understood this pattern—that a larger network is a sparsely linked group of more densely linked subnetworks—you can see how it could operate at multiple scales. You could tie several few-person networks together into a network of networks. Connections in these larger networks are still between individual people, but now those individuals have become even more critical; in fact, the larger the network is, the more important the highly connected individuals are in holding the overall structure together. Even at seemingly absurd extremes, the pattern holds: random pairs of people from New York City, a pool of millions, are likelier to be connected in a shorter chain than random pairs drawn from the Northeast, and pairs from the Northeast are likelier to be connected in a shorter chain than random pairs from the United States. The layers are arbitrary, but the comparison isn’t: because the smaller networks are denser than the larger network of which they are part, the pattern repeats itself at many scales.

  Small World networks operate as both amplifiers and filters of information. Because information in the system is passed along by friends and friends of friends (or at least contacts and contacts of contacts), people tend to get information that is also of interest to their friends. The more friends you have who care about a particular piece of information—whether gossip or a job opening or a new song they like—the likelier you are to hear about it as well. The corollary is also true: things that none of your friends or their friends care about are unlikely to get to you. This pair of functions, amplification plus filtering, was at work in the huge MySpace protests in California in 2006. In late March of that year tens of thousands of students in the Los Angeles Unified School District walked out of school and down to City Hall, stopping traffic as they went, as part of a broader protest against a proposed anti-immigration law, HR 4437. Like the Belarusian and Filipino protests, the school boycott assembled quickly, using MySpace pages and mobile phones, and school officials didn’t see it coming. The walkout upset the administrators both because it was a threat to their ability to keep order and also because California pays its schools based on average daily attendance, so the student walkout also threatened the schools with a financial penalty. Unlike the old “advertise to everyone but reach a fractional population” model for protests in years past, the Small World network of MySpace and text messaging meant that the message mainly went to students who would already be interested in participating, without becoming public before the eve
nt itself.

  Small Worlds networks mean that people don’t simply connect at random. They connect in clusters, ensuring that they interact with the same people frequently, even in large networks. This in turn reduces the Prisoners’ Dilemma and helps create social capital. One reason the phrase “social capital” is so evocative is that it connotes an increase in power, analogous to financial capital. In economic terms, capital is a store of wealth and assets; social capital is that store of behaviors and norms in any large group that lets its members support one another. When sociologists talk about social capital, they often make a distinction between bonding capital and bridging capital. Bonding capital is an increase in the depth of connections and trust within a relatively homogenous group; bridging capital is an increase in connections among relatively heterogeneous groups. Think of the difference by considering the number of people to whom you’d lend money without asking when they’d pay you back. An increase in bridging capital would increase the number of people you’d lend to; an increase in bonding capital would increase the amount of money you’d lend to people already on the list.