In some cases relationships are just binary: you are either a "friend" or not on services like Twitter or MySpace.
Sometimes it's binary with some added metadata, like Facebook, where you can describe how you met a "friend" choosing from a limited list; or Flickr, where you can decide that a "contact" is also a "friend" and/or "family".
In other cases it can be more refined, for example within PeopleAggregator we let users pick one from a list of relationship types defined by the network administrator (i.e.: from "never met" to "best friend") plus a "family" flag.
Then there are different methods to make connections reciprocal.
But in all cases it's the definition of the different kinds of relationship is set by the network within which you are creating these relationship, and until recently I was convinced that this had to be the case because otherwise the complexity would be too high.
But maybe I was wrong, maybe it would be possible to manage a much more granular and personalized relationships system.
In an unbundled social environment, where I would be managing my relationships on my own tools and use them across the different networks I would belong to, I could be defining my own type of relationships, exposing them to the people in my personal networks and to the content management systems I would be using in order to let them distribute various elements of my digital lifestyle.
Hmm... it's still somewhat fuzzy, but I think there might be something interesting down here...