The Social Graph is Neither

I uploaded a bubble sort in prolog and a b-tree in Haskell ("I'm an enlightened far-reaching hipster hacker").

I quoted Allen Ginsberg on facebook. "Look at me I'm smart and read poetry".

MySpace collapsed and Facebook's clean UI won because most people don't have that much identity.

The most popular person in high school was almost certainly not a nerd. Nerds take a while to maxing out their social skills, but once they do I find they are almost as good at it as sales guys or MBA gals.

From my experience the most popular kid almost never had the time to dedicate to crafting the technical skills necessary to build a mega-successful online business, Social or otherwise. Nerds think about things far more than the typical prep or jock.

I can calculate pi to "enough" precision, whether I'm making a hula hoop, or a piston ring, or sending a probe to Saturn - I suspect a lot of what I'd like out of a functional "social graph" probably doesn't need much more precision than "three and a bit", 3.1416+-0.0004, or 3.14159265+-0.000000005.

A few comments: In general, what data you capture, always depends on what you want to do with the data. Any data model is an abstraction, which discards a lot of information. This simplification is what makes the data model useful.

Of course, storing social relations as an undirected, graph with a single type of edge, as is done in Facebook, is a gross simplification of real social structure. Everyone knows that; Facebook know that; but the fact is that you can build a lot of useful stuff, with even such a simple model. Is it perfect? No - I'm sure we will do better in future - but you've got to admit its amazingly successful, for such a simple representation.

So, we aren't going to be able to define a format for the social graph that is so rich, that it will capture all possible uses with it - but I don't see why we couldn't define a format that captures enough detail to do many a great many of the things we might want to.

Next point: graphs are amazingly flexible data structures; if you have a multigraph or hypergraph representation, or a set of graphs, you can represent an awful lot of rich information. Don't knock graphs too much. You could build an awful lot of cool communication functionality, if you had really any rich social graph representation, not necessarily the 'perfect' one.

The author also writes: "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." Of course, friendship isn't strictly transitive; otherwise the giant connected component of a social graph, would be one big clique; but friendship is highly transitive. That's why recommendation works so well in Facebook. If A is friends with B, and A is friends with C, then there is a vastly higher chance that B is friends with C, than with some other random person. Sociologists have called this 'triadic closure' and studied it for years: http://en.wikipedia.org/wiki/Triadic_closure

So, while the recommendations won't be perfect, as things like friendship aren't perfectly transitive, they'll potentially be very good. But what recommendation engine is perfect, anyway?

Finally, the author refers to the problem of finding an adequate model as 'AI-hard' - well, as I've said, any model - even the most perfect one (unless we go and make a copy of the entire state of the world) - is going to sacrifice something. So I see why coming up with a perfect model is impossible. But as to coming up with a model that is adequate for a great many of the possible things we might want to build; well, I'd say 'AI-hard' is an exaggeration.