I think they have come up with an idea, which hasn't been thought about before and extends and builds on the principles of Web 2.0.
In short: Riya (formerly OJOS) = Flickr + Tagging + Human Learning + Visual Processing + AI = Automated Photo Tagging.
The software and Computer Science research to get computers to recognise a visual image has been improving in relation to the processing power available. Yet with all these advances nothing could bootstrap the machine learning process better than the ability for humans to 'teach' a computer on how to identify an image. This is where the collaborative tagging capability comes into play.
We have seen examples of where collaborative filtering to use the wisdom and knowledge of crowds to accurately determine whether an e-mail message is SPAM.
I think Riya is trying to take advantage of all the images that have been human "tagged" and correlate this with its own AI / Pattern Recognition engine.
In short... after the 1000th image of an iPod is uploaded on a blogger site or Flickr, Riya will have learned not only physical dimensional qualities about an image -- but also tagged data about it... -- Anyway thats my take on what these folks are up to...