In a comment to Don Dodge, Robert Scoble explains Riya as:
"It's a photo service where you upload your pictures. It recognizes faces and puts an entry box around them. You tell it a face, then it goes and looks for other versions of that same face. It works pretty well right now (I'd guess it recognized about 80% of the times my wife was in a picture, for instance)."
I understand better what Riya is doing now. I guess in a way you could describe Riya much more generically as an auto-caption generator for a photo-album. That would be a cool feature.
In fact, I'd like to see more photo-catalog formatters/generators that display photos nicely on a map, label what the photo is based on geo-location/sensed data from the camera, and the like.
I'm not convinced that the concept is worth tens of millions, but the more generic idea sounds cool.