19 Sep 2004
This is the first post in a series where I will discuss challenges in social software design and potential solutions.
Problem: In social software, how can you decide what to disclose to whom?
Attempts: One common approach is to divide relationships into classes, like LiveJournal's "friends" list or FOAF's relationship types. Another approach is to use your existing articulated social network and choose by degree: some items are open to your immediate friends, some to their friends too, some to the entire network.
Failures: Both of these approaches are broken, precisely because they are automatic. Human relationships are both finely nuanced and ever changing, and any automatic system effectively moves the question of "to whom should this be sent" up to the point of content creation. This is reasonable for a party invitation or personal ad, but inspires no confidence for the sharing of actual secrets. One could argue that this is because current a priori neftwork specifications are too coarse, but any style of relationship specification even close to the required level of complexity would be far too difficult to articulate—and then far too brittle. So until computers can read minds, if we want to facilitate the sharing of truly privileged information, we have to do it differently.
Solution: Rather than attempt to predict disclosure from the context of content creation, facilitate disclosure from the context of existing interpersonal contact. This reflects real life: you won't send updates on your romantic life to your old girlfriends, but when you see one at a party you might discuss that for a while. You might do this for some ex-girlfriends but not all. You might want to discuss this and change your mind when you hear she's just been dumped. But you can almost picture going up to her with a big list of potential things to talk about, and then screening some away as you communicate.
The software should provide a similar interface. In the context of authoring a message or other piece of content with a known audience, offer the author a way of reaching into his palette of private content and tossing something special into the communication. Select potentially relevant items based on data mining of past, similar interactions; and offer full-text search. Present this alongside the ongoing dialogue: