A Group Is Its Own Worst Enemy: Social Structure in Social Software Clay Shirky's talk at ETCON 2003. Cory Doctorow doctorow@craphound.com -- Social Software: software that supports group interaction The Internet supports lots of communications modes: 2-way, 1-many, many-1, many-many Pre-Internet: had same models Pre-Internet, the last important tech that changed the way we interact in groups was the table (to a lesser extent, the conference call -- but they suck) "Software that supports group interaction" -- bad definition, too broad, includes spam and emailers and stuff. Email can support group patterns, but it can, same with blogs. Instapundit is a broadcaster -- closer to MSNBC than a conversation; OTOH LiveJournal is more like a conversation. But this definition recognizes that groups are a run-time effect: you can't anticipate what they'll do. WR Byaim (sp?) did group theory with nuerotics -- discovered that neurotics were working as a group to defeat therapy. He couldn't resolve whether this was a group activity or individual activity. Humans are fundamentally individual AND fundamentally social. It's easy to see how cohesion comes out of guilds in MMORPGs, but Byam says that this is much deeper and happens sooner than you think Example: You were at a party and you got bored. They you don't leave. Why don't you leave? But 20 min later, someone gets his coat and everyone leaves. Everyone else was bored too, but the triggering event let the air out of the group. This is called "the paradox of groups." There are no groups w/o members, but there are also no members w/o groups. How Byat defined the behaviours that defeated the therapy: * Sex talk: flirting. Every IRC channel is full of this, too. Flirting is always in scope in human conversation. * Vilification of external enemies: OSS movement in the mid-90s; you could always get a conversation going about MSFT and BillG. Creates cohesion. Members who are most paranoid are best at identifying enemies. * Veneration: Veneration of a religious symbol [[audience chants "CLAY CLAY CLAY" -- reference to Orlowski's dumb-fuck hatchet-job in yesterday's Register UK]]. Go into a Tolkein newsgroup and diss Two Towers -- see what happens. Even though it's supposed to be about discussing Tolkein, they'll just flame you, 'cause you're interfering with the religious text. Group structure (Rob't's Rule, Constitution, etc) are necessary, to draw a circle around acceptable behaviours. Group structure keeps a group on target on message, focused on its sophisticated goals and from sliding into these patterns. It defends the group from the patterns of its own members. In the 70s, Communitree (an early BBS) threw off structure to see what patterns arise. Incredible things do happen when this happens. Over time, new patterns emerge. The users included boys with modems at high-schools who told fart jokes, posted 4-letter words, nyah nyah nyah. The adults were overrun. The open access place had too much openness. They couldn't defend themselves against their own users. They had too much free speech. The site was shut down. Was their inability defend themselves from being overrun a technical or a social problem? Did the software not allow it, or was the group constinutionally incapable of censoring? It doesn't matter: the two problems are intertwined. Attack from within is the pattern that matters. Communitree didn't die because of DoS or hacking -- it was taken down by logging in and posting. This story has been written many times, and you'd hope that people would read these accounts and fix it. Call it "learning from experience," but learning from experience sucks compared to reading: i.e., "Don't go there, you will be eaten by an alligator," is better than discovering the alligator first-hand. The same thing has happened at Lucasfilm's online community, LamdaMOO, etc. On LambdaMOO, the wizards said that they would only address technical, not social problems. But they discovered that they couldn't change the two of them. Social software people are more like economists than programmers. Groups of people are a runtime phenomenon. This is a computer Constitutional Crisis. The first crisis is the worst: not just, "We need rules," but "we need rules for making rules." The chance that any unmoderated group will have a flamewar about whether to have a moderator approaches 1 over time. (Geoff Cohen) Social software is getting bigger. It's not the dotcom boom because Yahoo was about how manyu users you can support -- which means that dense interconnected patterns that drive interaction can't be supported at large scale. Small groups have interactions that are lost to large groups. Mid sized groups are bigger than 10 and smaller than 1000, and they have their own interactions. The development cycle is faster because we have a good platform. Ludicorp built ConFab in two weeks. Why now? Because it's time: when blogtime came, it had nothing to do with tech. We had all the tech we needed for blogs when Mosiac came out, but instead we got Geocities. Phil Gyfford's Pepyslog implied that blogs will be around for 10 years -- that's how long Pepys's diaries are. The new apps are web-native: not a light Web-frontend on LOTUS DREADNAUGHT. Wikis and blogs live on the Web. RSS is web-native syndication. We can now have a "small pieces, loosely joined" pattern -- see Joi Ito's Emerging Democracy. The Emerging Democracy happening made conference calls valueable with interrupt logic moved into a chatroom and annotated in a wiki. Broadband confernece call made out of three loosely coupled software apps. The other factor: ubiquity. Most people have web-access. In many situations, ALL people have access to the net. ALL is different from MOST. For some people, everyone they work with, know, are related to is online. New software assumes that ALL offline groups will have an online component. That's different from the old, separated online and offline lives. Another kind of ubiquity: whenever a group gathers together, they can be online and offline together and do new things at the same time: all my meetings have chatrooms and/or Wikis. People who used the Wiki had never used it before. If these assumptions are right (a group is its own worst enemy and social software is increasing) what can we say? There's something supernatural about groups as a runtime phemomenon -- failure is the default mode. Most YahooGroups are dead. There's nothing you can do that will make it come out right every time. Lessons: If you're going to design a piece of social software, accept: * You can't completely separate social and technical issues. Separate mailing lists, etc, are dumb. This doesn't work. See "LambdaMoo Takes a New Direction." The conversation can't be forked. * Software determines what people do -- up to a point. You can't program social issues completely. Groups always assert their rights. Put the responsibility for describing group values into the group's hand. * Members are different from users. There is always a group that cares more than the average about the integrity of group as a whole. That's the core group. The software doesn't always allow the core group to express itself, and if it doesn't they'll invent new ways. On alt.folklore.urban, the core group set up a mailing list called the Old Hats for coordinating activity. Then they started a "Young Hats" list -- like Anonymous Cowards, logged-in users and high-Karma users in /.. * The core group has rights that trump individuals in some situations. This is at odds with liberatarianism and one-user-one-vote. There was a vote for a soc.culture.tibetan newsgroup that got voted down by Chinese university students. This was the wrong answer. Contentious groups all got voted away by people who opposed them, but the people who matter to the creation of the group are the people who want to discuss it. See WikiPedia volunteer fire department: roll back graffiti, stay up despite repeated attacks. If you're going to design a piece of social software, design for: * All groups have a Constitution: formally instantiated in code, informally instantiated in social norms * Handles for the user that matters. Anonymity doesn't work well for large groups, neither does weak pseudonymity. The world's best reputation system is in the emotional center of the brain. Almost all reputation systems are trivial or bad or both. Some people cheat on their wives but not at cards. eBay has done us all a disservice: non-iterated atomic transactions are the opposite of social situations. A good repuatation system just needs to let me know who you are. There has to be a penalty for switching handles. * A way for good works to be identified -- member since, Karma, etc. A music sharing group that FedExes 180 GB HDDs back and forth gives new members the username, $SPONSOR'S_UID_MEMBERNAME * There needs to be a cost to participate: moderators have to be around for a while. It needs to be hard to do somethings on the system: otherwise the core group won't be able to defend itself. This is anti-ease-of-use, but only for individuals, but the user of social software is the GROUP, not the USER. * Spare the group from scale: conversations require dense, two-way conversations. Metcalfe's law is a drag. The value of a group is inverse to its size: you'll give a kidney to a smaller group than you'd give a kiss to. MeFi shuts off new users when it gets too big. People who use your software have rights, even if you own the software.