RSS

Pinboard Blog

The Social Graph is Neither

I first came across the phrase social graph in 2007, in an essay by Brad Fitzpatrick, though I'd be curious to know if it goes back further.

The idea of representing relationships between people as networks is old, but this was the first time I had thought about treating the connections between all living people as one big object that you could manipulate with a computer.

At the time he wrote, Fitzpatrick had two points to make. The first was that it made no sense for every social website to try and recreate the same web of relationships, over and over, by making people send each other follow requests. The second was that this relationship data should not be proprietary, but a common resource that rival services could build on as a foundation.

Fitzpatrick subsequently went to work for Google, and his Utopian vision of open standards and open data became subsumed in a rivalry between Google and Facebook. Both companies now offer their version of a social graph API, and Google (which is trying to catch up) has taken up the banner of open standards and data portability.

This rivalry has brought the phrase 'social graph' into wider use. Last week Forbes even went to the extent of calling the social graph an exploitable resource comparable to crude oil, with riches to those who figure out how to mine it and refine it.

I think this is a fascinating metaphor. If the social graph is crude oil, doesn't that make our friends and colleagues the little animals that get crushed and buried underground?

But right now I would like to take issue with the underlying concept, which I think has two flaws:

I. It's not a graph

The idea of the social graph is that each person is a dot in a kind of grand connect-the-dots game, the various relationships between us forming the lines. Some of these lines may point in one direction (Ted works for Sylvia), while others are reciprocal (Bob is my neighbor). All of them taken together form a mathematical structure called the social graph. Facebook even has a pretty picture!

We nerds love graphs because they are easy to represent in a computer and there is a vast literature on how to do useful things with them. When you ask Google for directions from Detroit to Redwood City, for example, you're interacting with a graph that represents the US road network. The same principle applies any time a site tells you people who bought object X might also be interested in book Y.

In order to model something as a graph, you have to have a clear definition of what its nodes and edges represent. In most social sites, this does not pose a problem. The nodes are users, while edges means something like 'accepted a connection request from', or 'followed', or 'exchanged email with', depending on where you are.

The way you interpret this is another matter - does clicking 'follow' imply you're friends with someone in real life? But at least what the data model represents is unambiguous.

But when you start talking about building a social graph that transcends any specific implementation, you quickly find yourself in the weeds. Is accepting someone's invitation on LinkedIn the same kind of connection as mutually following them on Twitter? Can we define some generic connections like 'fan of' or 'follower' and re-use them for multiple sites? Does it matter that you can see who your followers are on site X but not on site Y?

One way to solve this comparison problem is with standards. Before pooling your data in the social graph, you first map it to a common vocabulary. Google, for example, uses XFN as part of their Social Graph API. This defines a set of about twenty allowed relationships. (Facebook has a much more austere set: close_friends, acquaintances, restricted, and the weaselly user_created).

But these common relationships turn out to be kind of slippery. To use XFN as my example, how do I decide if my cubicle mate is a friend, acquaintance or just a contact? And if I call him my friend, should I interpret that in the northern California sense, or in some kind of universal sense of friendship?

In the old country, for example, we have two kinds of 'friendship' (distinguished by whether you address one another with the informal pronoun) and going from one status to the other is a pretty big deal; you have to drink a toast with your arms all in a pretzel and it's considered a huge faux pas to suggest it before both people feel ready. But at least it's not ambiguous!

And of course sex complicates things even more. Will it get me in hot water to have a crush on someone but have a different person as my muse? Does spouse imply sweetheart, or do I have to explicilty declare that (perhaps on our 20th anniversary)? And should restrainingOrder be an edge or a node in this data model?

There's also the matter of things that XFN doesn't allow you to describe. There's no nemesis or rival, since the standards writers wanted to exclude negativity. The gender-dependent second e on fiancé(e) panicked the spec writers, so they left that relationship out. Neither will they allow you to declare an ex-spouse or an ex-colleague.

And then there's the question of how to describe the more complicated relationships that human beings have. Maybe my friend Bill is a little abrasive if he starts drinking, but wonderful with kids - how do I mark that? Dawn and I go out sometimes to kvetch over coffee, but I can't really tell if she and I would stay friends if we didn't work together. I'd like to be better friends with Pat. Alex is my AA sponsor. Just how many kinds of edges are in this thing?

And speaking of booze, how come there's a field for declaring I'm an alcoholic (opensocial.Enum.Drinker.HEAVILY) but no way to tell people I smoke pot? Why are the only genders male and female? Have the people who designed this protocol really never made the twenty mile drive to San Francisco?

What happens to dead people in the social graph? Facebook keeps profiles around for a while in memoriam, so we probably shouldn't just purge dead contacts from the social graph immediately. But we certainly don't want them haunting us on LinkedIn - maybe there should be a second, Elysian social graph where we can put those nodes to await us?

You can call this nitpicking, but this stuff matters! This is supposed to be a canonical representation of human relationships. But it only takes five minutes of reading the existing standards to see that they're completely inadequate.

Here the Ghost of Abstractions Past materializes in a flurry of angle brackets, and says in a sepulchral whisper:

“How about we let people define arbitrary relationships between nodes…”

(subject,verb,object)

“Maybe even in XML…”

<Person "john">
    <likesToShareRecipesWith "susan" />
</Person>

“Of course, we'll need namespaces…”

	
  <ns:Person rdf:about="http://www.example.org/#john">
    <ns:likesToShareRecipesWith 
     rdf:resource="http://www.example.org/#susan" />
  </ns:Person>

And RDF rises lurching out of the grave to infect the brains of another generation of young developers.

But even if we go ahead and build the Semantic Web, 2004 edition, and populate it with information about all our connections to other people, it still won't be expressive enough.

One big sticking point is privacy. Do I really want to find out that my pastor and I share the same dominatrix? If not, then who is going to be in charge of maintaining all the access control lists for every node and edge so that some information is not shared? You can either have a decentralized, communally owned social graph (like Fitzpatrick envisioned) or good privacy controls, but not the two together.

There's another fundamental problem in that a graph is a static thing, with no concept of time. Real life relationships are a shared history, but in the social graph they're just a single connection. My friend from ten years ago has the same relationship to me as the friend I dined with yesterday. You're left with forcing people (or their software) to maintain lists like 'Recent Contacts' because there is no place in the model to fit this information.

"No problem," says Poindexter. "We'll add a time series of state transitions and exponentially decaying edge weights, model group dynamics as directional flows, and pass a context object in with each query..." and around we go.

This obsession with modeling has led us into a social version of the Uncanny Valley, that weird phenomenon from computer graphics where the more faithfully you try to represent something human, the creepier it becomes. As the model becomes more expressive, we really start to notice the places where it fails.

Personally, I think finding an adequate data model for the totality of interpersonal connections is an AI-hard problem. But even if you disagree, it's clear that a plain old graph is not going to cut it.

II. It's Not Social

The social graph project has roots in something called Friend of a Friend, or FOAF (disclaimer: I worked on a rival project called LOAF, and you may feel free to ascribe everything I say here to purest bitterness).

The idea of FOAF was that everyone would create little XML snippets that represented their interests. For example, if you liked burgers and had a huge crush on your neighbor Matt, you could generate an RDF file that said so and stick it in to your Geocities page.

The problem FOAF ran headlong into was that declaring relationships explicitly is a social act. Documenting my huge crush on Matt in an XML snippet might faithfully reflect the state of the world, but it also broadcasts a strong signal about me to others, and above all to Matt. The essence of a crush is that it's furtive, so by declaring it in this open (but weirdly passive) way I've turned it into something different and now, dammit, I have to go back and edit my FOAF file again.

This is a ridiculous example (though it comes up with strange regularity in the docs), but we run into its milder manifestations all the time. Your best friend from high school surfaces and sends a friend request. Do you just click accept, or do you send a little message? Or do you ignore him because you don't want to deal with the awkward situation? Declaring connections is about as much fun as trying to whittle people from a guest list, with the added stress that social networking is too new for us to have shared social conventions around it.

OkCupid was one of the first social sites to understand that every visible action sent a signal. While other dating sites nagged you to upgrade to an expensive 'Gold' status, which branded you as a foreveralone pariah, OkCupid took pains to make sure people had stuff to do on the site that was unrelated to dating. A popular activity was building and taking personality quizzes. The quiz feature removed some of the stigma from hanging out on the site (I'm just a cool guy having some fun with these quizzes!) while creating a whole new avenue for meeting people.

Social graph proponents seem uninterested in the signaling problem. Leaving aside the technical issues of how to implement it, how does cutting ties actually work socially? Is there any way to be discreet, for example, or have connections naturally degrade over time? In real life, all relationships fade naturally if you don't maintain them, but right now social networks preserve ties in amber until we explicitly break them. Is my sister going to resent me if I finally defriend her annoying husband? Can I unfollow my ex now, or is that going to make her think I'm still hung up on her?

There's no way to take a time-out from our social life and describe it to a computer without social consequences. At the very least, the fact that I have an exquisitely maintained and categorized contact list telegraphs the fact that I'm the kind of schlub who would spend hours gardening a contact list, instead of going out and being an awesome guy. The social graph wants to turn us back into third graders, laboriously spelling out just who is our fifth-best-friend. But there's a reason we stopped doing that kind of thing in third grade!

-:-

You might almost think that the whole scheme had been cooked up by a bunch of hyperintelligent but hopelessly socially naive people, and you would not be wrong. Asking computer nerds to design social software is like hiring a Mormon bartender. Our industry abounds in people for whom social interaction has always been more of a puzzle to be reverse-engineered than a good time to be had, and the result is these vaguely Martian protocols.

But let's say an inspired mathlete proves me wrong. There's a brilliant hack that fixes all the issues I've raised and we go ahead and build the Platonic social graph. What can you actually do with it?

Well, one thing we've seen is that machine-readable lists of friends make it much easier to launch social sites. Letting a thousand startups bloom is one of the big justifications in Fitzpatrick's essay. But is removing this friction a good thing? It is admittedly annoying to have to re-follow people every time you sign up for something, but it also forces the authors to make the site appealing enough to get us over that hurdle. We're already starting to see apps whose first act is to suction down our contact list and spam our various accounts with invites without bothering to woo us at all. I can't imagine having open API access to the social graph is going to improve that.

In other domains, a big graph would be good for recommendations, but friendship is not transitive. There's just no way to tell if you'll get along with someone in my social circle, no matter how many friends we have in common.

But one thing you can do is mine a huge amount of information about my friends and infer things about their interests, income, social status and tastes. And then maybe you can use that information to bring them valuable news and offers, or help them digitally engage with their favorite brands.

Imagine the U.S. Census as conducted by direct marketers - that's the social graph.

Social networks exist to sell you crap. The icky feeling you get when your friend starts to talk to you about Amway, or when you spot someone passing out business cards at a birthday party, is the entire driving force behind a site like Facebook.

Because their collection methods are kind of primitive, these sites have to coax you into doing as much of your social interaction as possible while logged in, so they can see it. It's as if an ad agency built a nationwide chain of pubs and night clubs in the hopes that people would spend all their time there, rigging the place with microphones and cameras to keep abreast of the latest trends (and staffing it, of course, with that Mormon bartender).

We're used to talking about how disturbing this in the context of privacy, but it's worth pointing out how weirdly unsocial it is, too. How are you supposed to feel at home when you know a place is full of one-way mirrors?

We have a name for the kind of person who collects a detailed, permanent dossier on everyone they interact with, with the intent of using it to manipulate others for personal advantage - we call that person a sociopath. And both Google and Facebook have gone deep into stalker territory with their attempts to track our every action. Even if you have faith in their good intentions, you feel misgivings about stepping into the elaborate shrine they've built to document your entire online life.

Open data advocates tell us the answer is to reclaim this obsessive dossier for ourselves, so we can decide where to store it. But this misses the point of how stifling it is to have such a permanent record in the first place. Who does that kind of thing and calls it social?

III What, then, is to be done?

The funny thing is, no one's really hiding the secret of how to make awesome online communities. Give people something cool to do and a way to talk to each other, moderate a little bit, and your job is done. Games like Eve Online or WoW have developed entire economies on top of what's basically a message board. MetaFilter, Reddit, LiveJournal and SA all started with a couple of buttons and a textfield and have produced some fascinating subcultures. And maybe the purest (!) example is 4chan, a Lord of the Flies community that invents all the stuff you end up sharing elsewhere: image macros, copypasta, rage comics, the lolrus. The data model for 4chan is three fields long - image, timestamp, text.

Now tell me one bit of original culture that's ever come out of Facebook.

Right now the social networking sites occupy a similar position to CompuServe, Prodigy, or AOL in the mid 90's. At that time each company was trying to figure out how to become a mass-market gateway to the Internet. Looking back now, their early attempts look ridiculous and doomed to failure, for we have seen the Web, and we have tasted of the blogroll and the lolcat and found that they were good.

But at the time no one knew what it would feel like to have a big global network. We were all waiting for the Information Superhighway to arrive in our TV set, and meanwhile these big sites were trying to design an online experience from the ground up. Thank God we left ourselves the freedom to blunder into the series of fortuitous decisions that gave us the Web.

My hope is that whatever replaces Facebook and Google+ will look equally inevitable, and that our kids will think we were complete rubes for ever having thrown a sheep or clicked a +1 button. It's just a matter of waiting things out, and leaving ourselves enough freedom to find some interesting, organic, and human ways to bring our social lives online.

—maciej on November 08, 2011



Pinboard is a bookmarking site and personal archive with an emphasis on speed over socializing.

This is the Pinboard developer blog, where I announce features and share news.




How To Reach Help

Send bug reports to bugs@pinboard.in

Talk to me on Twitter

Post to the discussion group at pinboard-dev

Or find me on IRC: #pinboard at freenode.net