by Clay Shirky
No matter how you display it, this shape is very different from the bell curve distribution we are used to. Imagine going out into your community and measuring the height of two hundred men selected at random. For anything like height that falls on a bell curve, knowing any one of the numbers—average, median, or mode—is a clue to the others. If you know the height of the median man, or the most common height among all the men, you can make an educated guess about the average height. And most critically, whatever you know about the average height can be assumed to be most representative of the group.
Now imagine height were described not by a bell curve but by a power law. If the average height of two hundred men was five foot ten; the most frequent (or modal) height would be held by dozens of men who were each only a foot tall, the median height would be two feet tall (a hundred men shorter than two feet, and a hundred taller). Most important, in such a distribution, the five tallest men would be 40, 50, 66, 100, and 200 feet tall respectively. Height doesn’t follow a power law (fortunately for the world’s tailors and architects), but the distributions of many social systems do. The most active contributor to a Wikipedia article, the most avid tagger of Flickr photos, and the most vocal participant in a mailing list all tend to be much more active than the median participant, so active in fact that any measure of “average” participation becomes meaningless. There is a steep decline from a few wildly active participants to a large group of barely active participants, and though the average is easy to calculate, it doesn’t tell you much about any given participant.
Any system described by a power law, where mean, median, and mode are so different, has several curious effects. The first is that, by definition, most participants are below average. This sounds strange to many ears, as we are used to a world where average means middle, which is to say where average is the same as the median. You can see this “below average” phenomenon at work in the economist’s joke: Bill Gates walks into a bar, and suddenly everyone inside becomes a millionaire, on average. The corollary is that everyone else in the bar also acquires a below-average income. The other surprise of such systems is that as they get larger, the imbalance between the few and the many gets larger, not smaller. As we get more weblogs, or more MySpace pages, or more YouTube videos, the gap between the material that gets the most attention and merely average attention will grow, as will the gap between average and median.
You cannot understand Wikipedia (or indeed any large social system) by looking at any one user or even a small group and assuming they are representative of the whole. The most active few users account for a majority of the edits, even though they make up a minority, and often a tiny minority, of contributors. But even this small group does not account for the whole success of Wikipedia, because many of these active users are doing things like correcting typos or making small changes, while users making only one edit are sometimes adding much larger chunks of relevant information.
Power law distributions tend to describe systems of interacting elements, rather than just collections of variable elements. Height is not a system—my height is independent of yours. My use of Wikipedia is not independent of yours, however, as changes I make show up for you, and vice versa. This is one of the reasons we have a hard time thinking about systems with power law distributions. We’re used to being able to extract useful averages from small samples and to reason about the whole system based on those averages. When we encounter a system like Wikipedia where there is no representative user, the habits of mind that come from thinking about averages are not merely useless, they’re harmful. To understand the creation of something like a Wikipedia article, you can’t look for a representative contributor, because none exists. Instead, you have to change your focus, to concentrate not on the individual users but on the behavior of the collective.
The power law also helps explain the difference between the many small but tightly integrated clusters of friends using weblogs and the handful of the most famous and best-trafficked weblogs. The pressures are reflected in Figure 5-2, which shows the relationship between a power law distribution and the kinds of communication patterns that can be supported.
Figure 5-2: The relationship between audience size and conversational pattern. The curved line represents the power-law distribution of weblogs ranked by audience size. Weblogs at the left-hand side of the graph have so many readers that they are limited to the broadcast pattern, because you can’t interact with millions of readers. As size of readership falls, loose conversation becomes possible, because the audiences are smaller. The long tail of weblogs, with just a few readers each, can support tight conversation, where every reader is also a writer and vice versa.
As is normal in a power law distribution, most writers have few readers. Such readers and writers can all pay similar amounts of attention to one another, forming relatively tight conversational clusters. (This is the pattern of small groups of friends using weblog or social networking tools, described in the last chapter.) As the audience grows larger, into the hundreds, the tight pattern of “everyone connected to everyone” becomes impossible to support—conversation is still possible, but it is in a community that is much more loosely woven. And with thousands of people paying attention, much less millions, fame starts to kick in. Once writers start getting more attention than they can return, they are forced into a width-versus-depth tradeoff. They can spend less time talking to everyone. (It’s no accident we call these interactions shallow and say that people who have them are stretched thin.) Alternatively, they can limit themselves to deeper interactions with a few people (in which case we call them cliquish or standoffish). At the extremes they are forced to adopt both strategies, to limit both the number and the depth of interactions. A wedding reception is a localized version of this trade-off. The bride and groom gather a room full of people they could talk to for hours, then talk to most of the guests for just a few minutes each so as not to be rude.
Why Would Anyone Bother?
Coase’s logic in “The Nature of the Firm” suggests that in organizing any group, the choice is between management and chaos; he assumes that it’s very difficult to create an unmanaged but nonchaotic group. But lack of managerial direction makes it easier for the casual contributor to add something of value; in economic terms, an open social system like Wikipedia dramatically reduces both managerial overhead and disincentives to participation. Even understanding how a wiki page comes into being does nothing to answer the hardest question of all: Why would anyone contribute to a wiki in the first place? The answer may be easiest to illustrate with a personal example.
I recently came across a Wikipedia entry for Koch snowflake, one of a curious bestiary of mathematical shapes called fractals (shapes that have the same pattern at many scales, like a fern leaf). The article had an animated image showing the snowflake in various stages of construction, accompanied by the following text:
A Koch snowflake is the limit of an infinite construction that starts with a triangle and recursively replaces each line segment with a series of four line segments that form a triangular “bump.” Each time new triangles are added (an iteration), the perimeter of this shape grows by a factor of 4/3 and thus diverges to infinity with the number of iterations. The length of the Koch snowflake’s boundary is therefore infinite, while its area remains finite.
This description is accurate but a little awkward. I rewrote it to read:
To create a Koch snowflake, start with an equilateral triangle and replace the middle third of every line segment with a pair of line segments that form an equilateral “bump.” Then perform the same replacement on every line segment of the resulting shape, ad infitum. With every iteration, the perimeter of this shape grows by 4/3rds. The Koch snowflake is the result of an infinite number of these iterations, and has an infinite length, while its area remains finite.
This rewrite describes the same shape but in a way that is a little easier to grasp.
Why did I do it? Nothing in my daily life
has anything to do with fractals, and besides, I was improving the article not for me but for subsequent readers. Psychological introspection is always a tricky business, but I know of at least three reasons I rewrote that description. The first was a chance to exercise some unused mental capacities—I studied fractals in a college physics course in the 1980s and was pleased to remember enough about the Koch snowflake to be able to say something useful about it, however modest.
The second reason was vanity—the “Kilroy was here” pleasure of changing something in the world, just to see my imprint on it. Making a mark on the world is a common human desire. In response to mass-produced technology with no user-serviceable parts inside, we install ringtones and screensavers, as a way to be able to change something about our inflexible tools. Wikipedia lets users make a far more meaningful contribution than deciding whether your phone should ring with the 1812 Overture or “Holla Back Girl.”
This desire to make a meaningful contribution where we can is part of what drives Wikipedia’s spontaneous division of labor. You may have noticed that I accidentally introduced a mistake in my edit, writing “ad infitum” when I should have written “ad infinitum.” I missed this at the time I wrote the entry, but the other users didn’t; shortly after I posted my change, someone went in and fixed the spelling. My mistake had been fixed, my improvement improved. To propose my edit, I only had to know a bit about the Koch snowflake; there are many more people like me than there are mathematicians who understand the Snowflake in all its complexity. Similarly, fixing my typo required no knowledge of the subject at all; as a result, the number of potential readers who could fix my mistake was larger still, and because the fix was so simple, they did not need to have the same motivations I did. (If you noticed that error as printed here and were annoyed by it, consider whether that would have been enough to get you to fix it if you could.) It’s obvious how Wikipedia takes advantage of different kinds of knowledge—someone who knows about World War II tank battles and someone who knows the rules of Texas hold’em are going to contribute to different articles. What’s less obvious is how it takes advantage of skills other than knowledge. Rewriting a sentence to express the same thought more readably is a different skill from finding and fixing spelling errors, and both of those differ from knowing the rules of poker, but all those skills are put to good use by Wikipedia.
The third motivation was the desire to do a good thing. This motivation, of all of them, is both the most surprising and the most obvious. We know that nonfinancial motivations are everywhere. Encyclopedias used to be the kind of thing that appeared only when people paid for them, yet Wikipedia requires no fees from its users, nor payments to its contributors. The genius of wikis, and the coming change in group effort in general, is in part predicated on the ability to make nonfinancial motivations add up to something of global significance.
Yochai Benkler, a legal scholar and network theorist and author of The Wealth of Networks, calls nonmarket creation of group value “commons-based peer production” and draws attention to the ways people are happy to cooperate without needing financial reward. Wikipedia is peer production par excellence, set up to allow anyone who wants to edit an article to do so, for any and all reasons except getting paid.
There’s an increasing amount of evidence, in fact, that specific parts of our brain are given over to making economically irrational but socially useful calculations. In one well-known experiment, called the Ultimatum game, two people divide ten dollars between them. The first person is given the money and can then divide it between the two of them in any way he likes; the only freedom the second person has is to take or leave the deal for both of them. Pure economic rationality would suggest that the second person would accept any split of the money, down to a $9.99-to-$.01 division, because taking even a penny would make him better off than before. In practice, though, the recipient would refuse to accept a division that was seen as too unequal (less than a $7-to-$3 split, in practice) even though this meant that neither person received any cash at all. Contrary to classical economic theory, in other words, we have a willingness to punish those who are treating us unfairly, even at personal cost, or, to put it another way, a preference for fairness that is more emotional than rational. This in turn suggests that relying on nonfinancial motivations may actually make systems more tolerant of variable participation.
We also have practical evidence that when a perceived bargain changes, previously contented volunteers will defect. America Online built its business as a user-friendly entry point into digital networks, and much of its friendliness came directly from AOL’s users, many of whom loved the service so much that they worked as volunteer guides. After AOL’s stock price rose into the stratosphere, however, a number of those guides banded together to file a class-action suit, claiming AOL had unfairly profited from their work. Nothing had changed about the job they were being asked to do; everything changed about the financial context they did it in, and that was enough to poison their goodwill. (Though the case is still pending, AOL has dropped the volunteer guide program.)
Social Prosthetics
The question every working wiki asks of its users is “Who cares?” Who cares that an article on asphalt exists? Cdani does. Who cares that it include photos? SCEhardt does. Who cares that the Koch snowflake description be clear? I do. Wikis reward those who invest in improving them. This explains why both experts and amateurs are willing to contribute—the structure of participation is not tied to extrinsic rewards, so people capable of adding to the technical explanation of complex mathematical shapes end up working alongside people who only know enough to be able to proofread descriptions of same. This reward, and the loyalty it creates, help explain one of the most complex questions about Wikipedia’s continued success: How does it survive both disagreement and vandalism? Openness, division of labor, and the multiple motivations of its users drive its rising average quality, but none of those things explain why articles on contentious subjects aren’t damaged by editing wars among rival factions, or simply destroyed by vandals, who can delete an entire article with the click of a button. Why don’t these sorts of things happen? Or to ask the same question in the language of economics: Why doesn’t Wikipedia suffer from the Tragedy of the Commons? Why haven’t free riders and even vandals destroyed it?
The wiki format is another version of publish-then-filter; coercion is applied after the fact rather than before. All edits are provisional, so any subsequent reader can decide that a change to an article is unacceptable, to be further edited or to be deleted. This capability is universal; any edit or deletion can be further edited or undone (“reverted”), changes that are themselves then held up for still more scrutiny, ad infinitum. Every change to a Wikipedia article is best thought of as a proposed edit; it shows up the minute it is made, but it is still subject to future review and revision. (I checked back on the Koch snowflake article later and was pleased to see my changes had survived such review.) In the case of obvious vandalism, the review process happens astonishingly quickly. Martin Wattenberg and Fernanda Viegas, researchers at IBM who study Wikipedia, have documented a number of contentious articles on subjects like abortion and Islam where complete deletions of the articles’ contents have been restored in less than two minutes.
Like everything described in this book, a wiki is a hybrid of tool and community. Wikipedia, and all wikis, grow if enough people care about them, and they die if they don’t. This last function is part of any working wiki, but it isn’t part of the wiki software, it’s part of the community that uses the software. If even only a few people care about a wiki, it becomes harder to harm it than to heal it. (Imagine a world where it was easier to clean graffiti off a wall than to put it there in the first place.) When a vandalized page reappears as if nothing has happened, it creates the opposite of the “Kilroy was here” feeling of a successful edit—nothing is more frustrating to a vandal than investing energy to change something and then have that effort disappear in seconds. E
vidence that enough people care about an article, and that they have both the will and the tools to quickly defend it, has proven enough to demoralize most vandals.
As with every fusion of group and tool, this defense against vandalism is the result not of a novel technology alone but of a novel technology combined with a novel social strategy. Wikis provide ways for groups to work together, and to defend the output of that work, but these capabilities are available only when most of the participants are committed to those outcomes. When they are not, creating a wiki can be an exercise in futility, if not an outright disaster. One notable example was the Los Angeles Times “Wikitorial” effort, in which the content of the paper’s editorial pages was made available to the public. The Times announced the experiment in a bid to drive users there, and drive them they did. A group of passionate and committed users quickly arrived and set about destroying the experiment, vandalizing the posted editorials with off-topic content and porn. The Wikitorial had been up for less than forty-eight hours when a Times staffer was told to simply pull the plug. The problem the Times suffered from was simple: no one cared enough about the contents of the Wikitorial to defend it, much less improve it. An editorial is meant to be a timely utterance of a single opinionated voice—the opposite of the characteristics that make for good wiki content. A wiki augments community rather than replacing it; in the absence of a functioning community, a wiki will suffer from the Tragedy of the Commons, as the Wikitorial did, as individuals use it as an attention-getting platform, and there is no community to defend it.
