Kottke points to a great article by Columbia's Duncan Watts...
The social context of content has everything to do with it's meaning. It's one of the reasons that I think that a purely pixel-based algorithmic approach to, say, image recognition is doomed. In optimistic moments, I've said that the computer vision community may produce a 98% reliable dog detector... But what we really want is a "funny" detector... or "cool" detector... that's gonna be a long-time coming... or maybe it's already here but involves analyzing people's actions around the pixels v. just the pixels in isolation.