danah boyd Stanford February 4, 2005 "Revenge of the Social Network: Lessons from Friendster" AGENDA In this talk, i want to make three points: 1) Building technology on misinterpretations and simplifications of social theory will not result in expected outcomes. 2) Resistance should be expected with social technologies and it is essential for developers to embrace, understand and be respectful of it. 3) Understanding the connection between theory, practice and technology is the next step for those concerned with Human-Human Interaction, with or without technology or computers. Let me begin by introducing Friendster. FRIENDSTER Friendster was developed in late 2002 as a dating site. It grew in popularity throughout 2003 as millions of people around the globe joined the service. The early adopters fit primarily into three communities: urban gay men (mostly in NY), Burning Man attendees and bloggers (many in Britain). As early adopters invited friends and those friends invited their friends, it spread to a diverse collection of hipsters, geeks, teenagers, etc. By early 2004, the site had attracted large communities from the Philippines, Malaysia, Singapore and Indonesia. Structurally, Friendster looks a lot like a dating site - there are profiles and you can search based on location, interest, sexuality, etc. Yet, two features separate Friendster from other dating sites: Friends and Testimonials. On Friendster, users can connect to their friends, forming an articulated public social network. This Friend network affects the visibility of users on the service, the way in which Profiles are constructed and the culture in general. From February thru October of 2003, i did extensive ethnography on Friendster. As a participant observer, i had access to over 2 million accounts. I interviewed or surveyed over 250 participants and ran 6 focus groups on various aspects of Friendster. Also, by being enmeshed in two of the key early adopter groups, i had regular access to how Friendster fit into broader cultural contexts for these groups. What is interesting about Friendster is not the technology, but how it was used, abused, challenged, adored and dismissed by different user groups. In this talk, i do not discuss much of the details of the Friendster phenomenon, although i allude to them. What is important to know is that while it spread like wildfire amongst certain populations, it has faded in popularity. [Identity Construction] [Activities] SOCIAL NETWORKING While it was built as a tool for dating, few people used it primarily for this. Instead, it was an opportunity for voyeurism as people figured out who knew who. People found old friends and engaged in social play. The tool itself was built on assumptions of social network behavior. Researchers espouse that social networks are a critical part of people's daily lives. People use their networks to acquire support of all different kinds. They use the structure of their network to find jobs, loved ones, get and make meaning out of recommendations. Enamored with the potential value of networks, Friendster and similar services sought to exploit the structure, believing it would be the key to improving sites that emphasized relationship development. These systems concerned themselves with what social network theory purportedly told us about our networks - Weak ties are key to getting jobs! The best dates are friends of friends! Networks give the best recommendations! They took these glossy statements and built technology to magnify the hype. Sites were constructed to model people's social networks so that users could more *efficiently* engage with their networks. If friends of friends are so valuable, why not make people aware of who each others' friends are? And why not *optimize* how people can connect? In building these systems, the creators made assumptions about both the networks and human behavior. Creators had a set of assumptions about how their systems would be used. They also assumed that social networking can and should be optimized. Needless to say, not all users agreed with, or even understood, the philosophies of the creators. As a result, users repurposed the tools available to support their own needs and desires. Thus, what users did did not resemble the outcomes predicted by the social science. In some cases, these new uses horrified the creators who actively attempted to construct boundaries of acceptable behavior, further aggravating the users. Let's begin by talking about four theories that were used. MILGRAM Consider for a moment the argument of 6 degrees of separation. Unbeknownst to pop culture, this isn't a term coined by Milgram. His research was called "Small Worlds." It's a term coined by a playwright. What Milgram did was question how many hops it would take to get from one person in Nebraska to a person in Massachusetts. He asked people in Nebraska to send a packet along to someone in Massachusetts. If they knew the target personally, they were asked to send the packet to the target. If they didn't, they were asked to send the packet on to someone who might know the target. Each person was asked to continue passing on the packet until the target was reached. Thus, someone in Nebraska might send it on to their neighbor who used to live in Massachusetts, who would send it on to someone who lived in the appropriate town, etc. Only a fraction of his letters made it to their target, although more than Milgram expected. He found that the average number of steps of those that did was around 5.5. Thus, everyone professed that people are, on average, separated by 6 hops! Using Milgram, if we build a model of a network, the average person will be separated by 6 people! There's a crucial missing step in this logic. What Milgram asked people to do was GUESS who would be the likely next candidate in passing on the message. The person did not have a whole networks view. She did not KNOW who should be the next candidate based on shortest path. She only guessed. Without a whole networks view, we don't actually know the average path length between two people. We only know the how long it took for some messages to get through a string of guesses within the network. Seeing the network is fundamentally different than making a next step towards a target based on what you know about your network. And seeing the network is not something that everyone exposes - some Nebraska residents may have been closer to their target, but didn't know it because their acquaintances did not reveal everything about themselves. DUNBAR Next, take the magical number 150. I've heard over and over again how the average person only knows 150 people. This is referring to research done by Robin Dunbar. He was interested in understanding how many people humans could "groom." In other words, how many people did a human keep in their social network at any one point in time. Dunbar was interested in how monkeys groomed each other to keep up their social network. What he found was that there was a similarity between monkey grooming and human gossip. Just as monkeys groomed to maintain their networks, humans gossiped to maintain theirs! He found that the MAXIMUM number of people that a person could keep up with socially at ANY GIVEN TIME, a.k.a. gossip maintenance, was 150. This doesn't mean that people don't have 150 people in their social network, but that they only keep tabs on 150 people max at any given point. Like many services, Friendster used the magical number to cap the number of Friends you could have, assuming that anyone who had more than 150 listed was trying to spam the service. Unfortunately, most of us actually have name recognition with or once knew far more than 150 people. On Friendster, it is not just a representation of who we currently keep tabs on, but who we recognize, were friends with, have to be socially polite to, etc. GRANOVETTER In his seminal work called "Strength of Weak Ties," Mark Granovetter found that the majority of people found out about jobs through weak ties, not strong ones. First, these services have corrupted the definition of weak tie to equate it with friends of friends. This is not accurate, but it's also not egregious. The idea is that with a strong tie, you are aware of all of their strong ties. What kind of best friend are you if you aren't aware of your best friend's other best friends? These people are not inherently your strong ties and, thus, they become weak ties in your world. You have a connection to them; that connection is because they are important in your friend's life. But there's a problem here. Not everyone that you know is a strong tie. And not everyone that you know's friends are weak ties. Furthermore, the term 'Friend' is highly culturally constructed. From Brazilians to ravers, there are groups who call everyone they've ever met Friend. It's socially appropriate! But this is not what the sociologists mean. As such, most of the people in our Rolodexes are weak ties, not strong ones. We do not have rich interpersonal connections with them. Yet, on Friendster, we call them all Friends. Before you object, think who you would call if you were in an emotional crisis; think of who knows all about the ups and downs of your anxieties. There's no doubt that some folks love to blog their woes to the world, but most of us only have a few people we would call and who we could guarantee would support us through our troubles. Those are strong ties. Pulling back from tie strength, let's also look at the logic of the argument. When Granovetter found that people found their jobs through weak ties, he did not find that all weak ties help us find our jobs. Furthermore, he did not say that our weak ties had the power necessary to get us those jobs, simply that they helped us find them. There's a reason for this. And this is the most overlooked issue in the phenomenon: Context. Our relationships have a context to them, not just a strength. That context is crucial for many distributions of information, support and trust. Consider a weak tie that you party with; he's dating someone who runs a big tech company. Your connection to him is a weak tie; his connection to his boyfriend is a strong one. He hears that the tech company has an opening. As Granovetter acutely noted, when he hears that you're looking for a job in that field, he's likely to tell you about that job, even though he knows nothing of your expertise. Why? The favor of spreading information gives him power as a bridge. But this does not mean that he's going to risk his reputation with his strong tie over a weak tie. Say that you want the chief engineering position. He doesn't know you in that context; he has no way to vouch for your worthiness. Thus, he's taking a much bigger risk in hyping you to his boyfriend. Even if he lets his boyfriend know that some guy from the clubs wants the job, he'll make it very clear that he doesn't really know you, but will just offer that information just in case. He'll do it in passing, to see if his ears perk up. If they don't, he'll let it pass. Take the dating realm. Have you ever had a close friend set you up on a date? Now, have you ever had a weak tie set you up on a date? One of the things you will find is that when a weak tie sets you up on a date, you often have NOTHING in common with this person. Again, context. Or better yet, focus. FELD Focus is a researched concept in social networks. Scott Feld talks about the power of foci in understanding networks. You and your strong ties have things in common, the foci of the relationship. Often, the closer you are, the more you share in common. This is why you often have things in common with friends of friends. You and your friend have lots of overlapping things; she and her friend have lots of overlapping things; there's a high probability that you and that person have some things in common. As Feld noticed, the more you have in common with someone, the more likely you are to be a strong tie, particularly when the things in common are rare. But, pulling back, shared foci between friends of friends does NOT mean shared foci between a string of weak ties. Foci are not transitive and the less you have in common with someone, the lower the probability will be that you have something in common with someone they know. Take two hops down a weak tie chain and the probability is random. STRUCTURAL EQUIVALENCE Take a look at this image. In this model, A & B share a lot of unconnected friends in common. Feld would look at this and say that there's a high probability that these people should know one another - they probably have a lot in common. In other words, they might be dating compatible. Now, on Friendster, i have seen this formulation frequently, only there's a huge twist to the process. If A & B have so many people in common, they are probably aware of each other by at least name recognition. Thus, on Friendster, they would be likely to mark each other as Friends even though they do not know each other. This configuration explains precisely one type of relationship: Breakups. A & B won't connect to one another because they dono't like each other any longer. ARCHITECTURE Current social network theory is built on architectural assumptions about social life, namely the face-to-face reality of negotiating life in the context of a city, suburbia or a village. Everyone in this room knows that technology has changed this. We can feel it, even if we don't understand it. Our social life has changed; the types of people we talk to are different; the ability to find people like us is visceral. There are critical differences in everyday social interaction for those of us who've gone digital or otherwise engaged in mediated communication. Even if we can't articulate the differences, we can feel them. In reflecting on Mitch Kapor's argument that architecture is politics, Larry Lessig noted that code is law. In fleshing this out, Lessig equates code to architecture. This is a critical realization. It allows us to think about how what we build structures the underlying elements to the ways people can interact. This realization doesn't negate social theory, but it puts it into context. It means that rather than translating everything that we learned about social life into the digital, we must critically reflect what will and won't work as new social structures are built. The core of people isn't different, but how they interact might be. This is a critical piece to always keep in mind when thinking about how social theory applies to technology - if the architecture changes, so does all behavior that sits on top of it. PRESENTATION OF SELF 1) Friends on these sites are not close ties. In fact, they're barely weak ties! Thus, anything that can be assumed about transitivity across ties is 100% lost. This only gets worse as we go down the chain. As one of my informants reminded me, why would i want to date my hairdresser's brother's drug dealer's second-cousin? 2) Asking favors is fundamentally different than offering them. People gain by being bridges. Thus, to be able to tell you about a job gives me reputation points in our relationship. Feeling pressured to connect you to someone makes me uncomfortable. In all of the networks described above, the bridge got to control the information flow. In Milgram's "Small Worlds," if you didn't know that i knew the target person, you may not have tried to pass it on to me. If you don't know that i am dating someone who has something that you want, you won't try to pressure me into giving you access to it. Thus, i can choose when to reveal my connections in a situation where i can come across as being helpful, rather than being put in a position to feel cornered. Revealing the network shifts the power. Of course, that's part of the point, right? All of these sites want you to USE your friends to gain access to jobs, dates and recommendations. But what's in it for your friends? To answer this, let's go back to who your "Friends" are. After Friendster came out, you'd go to a club in San Francisco or New York and people would be talking about their Friendsters. They weren't referring to their Friends... they were referring to the people that they connected with on the system. When sociologists asked people about their network, they did not pass on how people described their relationships. Thus, people could speak candidly about how well they knew people, in what context, and what they really thought of them. Thus, when they were building maps of relationships, they could make sense of these variations. AUTISTIC SOFTWARE There are three problems with these sites: 1) Publicly articulating who you know and how you feel about them is bloody well uncomfortable. Try saying no to that boss of yours you hate but need to play nice with. You would express this as a negative relationship to a sociologist, but you have to put them up on Friendster as equivalent to your lover. Your evil boss and your lover are not the same! Public articulation means that you can't distinguish them without awkward feelings. 2) People aren't good at formalizing their relationships into categories. Sociologists didn't ask people who their weak ties were; they asked about all of their ties and then came up with a uniform metric to gauge everyone's network. People can't honestly say how "sexy" someone is on a scale from 1-3 such as is encouraged on Orkut. Nor can they discuss trustworthiness without a context. 3) Not all relationships are bi-directional. This has to do with fandom and power. Just because you believe you really know Angelina Jolie does not mean that she knows you. On a more local level, not everyone knows the same about someone they know. For example, your shrink probably knows more about you than you know about her. And frankly, the kinds of questions that your superior can ask about your family and personal life are vastly different than the kinds of question you can ask her. Furthermore, just because you read someone's blog does not mean that you know them. Even if you comment. It has to be reciprocal and dive beyond the actual public forum. When you build technology that does not take these factors into consideration, you are building autistic software. In other words, you are asking all of the nuances and subtleties of social behavior to be turned into a simplistic formula that both person and computer can compute. These networks don't look real. They're too socially awkward. They're not built to give us a way to express the subtleties of how we know people, the power differentials, the contexts, the strengths. Furthermore, they expose more about us to different groups of people than we would ever do in real life. All of a sudden, we have to reconcile the bar-hopping facet of our identity with the proper work facet. Face management is a key pillar to sociability. Issues of face management became apparent to Friendster users because they typically signed on with one group of friends before running into another cluster of their network. This was most apparent when Burning Man attendees joined in connection with other Burners. Burning Man is an arts festival in Nevada where people don ridiculous clothing and run around the desert. On Friendster, it was quite common for Profiles to include half-naked images with artistic getups, images that were not appropriate when removed from the Burning Man context. Faced with work colleagues as well as festival attendees, people were faced with a decision - Do you shift your profile to look like them? or Do you find a middle ground? It doesn't matter, really... Because your colleagues can see that all of your friends are Burners. Guilt by association. Worse, Frendster exposes separate facets to each other without giving you the option to mediate. If the focus of our interactions between two groups are similar, we would comfortably expose them over time. If you find out that your colleague likes jazz, you might take him with you to meet your jazz-going friends. But if he hates jazz, you probably won't think to introduce him to the jazz aficionados. On Friendster, your ability to connect people because of their similarities is lost. The only similarity that matters is you. Furthermore, they get to interact through the system without you even negotiating whether or not they should meet. All of a sudden, your drunken friends are asking your boss out on a date cause she's hot. Yikes! Not only does this disempower you, remove the ability for you to connect them as need be, but it now makes you have to deal with the consequences of two different groups with two different standards of social norms. During my ethnography, i heard many stories of social awkwardness but none highlight these issues better than the story of a young teacher in San Francisco. Her friends found her Profile on Friendster and came in to accuse her of being a pedophile. She was horrified. She realized that one of her friends had a picture of him wearing a Catholic school girl costume and his friends had written testimonials about his love of small girls. What was a party joke amongst adults became a site of fear for youth. The code of ethics for teachers requires them to keep their personal lives separate from their work lives, but it was the technology that collapsed this. Nothing that she could do would help - leaving the service would indicate guilt. What was a funny joke amongst friends might appear offensive to outsiders. Jokes have context and part of that context involves audience. Repositioned, there are consequences. RESISTANCE AND CONFIGUREMENT Faced with unreal networks and peculiar social awkwardness, users went in a million different directions. Some of these directions can be marked as points of resistance as users challenged the expectations put forward by the service. Most notably, users created Fakesters - or fake characters. These personas represented everything from pop culture icon Ali G to Jesus to Stanford University. Fakesters had multiple purposes. For some, they allowed users to hide behind a pseudonym to escape the awkwardness. For others, it was a way of marking interest and finding others with similar interests. Fakesters turned Friendster into a social game, where people tried to exercise creativity and make others smile. Unfortunately, the creators of Friendster did not appreciate the Fakesters and sought to prevent them from existing. Fakesters slowed down the system, resulted in terrible search quality and ran counter to any efforts to create a serious dating service. Here is where Friendster made a problematic decision - they employed a tactic known in STS as 'configuring the user.' When technologies are built, the creators often have a very limited scope of desired and acceptable behavior. They build the systems aimed at the people who will abide by their desires. Often, their users don't have the same views about how the technology should be used. They use it differently. Creators get aggravated. They don't understand why users won't behave. The demand behavior. First, the creator messages the user, telling them that this isn't what is expected of them. Then, the creator starts carrying a heavier and heavier stick. This is called configuring the user. And y'know what... it doesn't work. Sure, Friendster stopped some bad behavior, but at what cost? They succeeded in getting rid of most of the people playing games with the site, but they also lost the foci elements that let people find people like them (often a better mechanism for dating) and the ability for people to safely guard themselves from others. Their heavy handedness didn't make the network any more real. For most users, it simply made it less fun and equally socially awkward. A small number of users reacted with outrage and went from creating light Fakesters to impersonating the creators themselves, staging protests outside of their office and encouraging everyone to leave the service, all the meanwhile discussing a Fakester Revolution. Fakesters reacted to being configured by creating increased chaos. One Fakester told me that the whole point of Fakesters was to remind people that nothing about Friendster is real. Profiles become outdated and meaningless, connections have more to do with social protocols than with actual friendships and articulations of self are usually utopian in nature anyhow. Fakesters saw their actions as political, reminding everyone that 'none of this is real.' Yet, the more designers try to force users into desired behavior, the less they pay attention to why users do what they do. Users are reacting to the design of the service, their experiences using it and how it fits into their cultural expectations. When their values are not aligned with the designer, trouble emerges. There are two ways to explain almost every aspect of user behavior - from the users' point of view and from the creators' point of view. The users are making sense of the technology dependent on how it fits into their lives, not dependent on how the creators want them to see it. In the case of Friendster, users showed what they wanted by forcing the system to do unexpected things, much to the dismay of the creator. THEORY + PRACTICE + TECHNOLOGY Embedded in all of this is a kernal that i believe is essential for future HCI development, both in industry and academia. You cannot implement theory into technology and expect practices foretold by theory. It doesn't work that way. You also can't force practice to behave like theory through technology. Resistance emerges when there is a disconnect between the technology and the practice. Practice is critical because it's how a particular tool fits into people's lives, values and culture. Social theory has been addressing practice since its origin and often talks about how technologies are employed, but in order to build new technology we need to question the theories, resituate them in architectural changes and be constantly aware of the practices that emerge. In other words, HCI, when concerned with the social domain, needs to constantly triangulate between three different things - theory, practice and technology. Each inform the others and it is essential to understand and balance the biases of each.