Last month, I've read an article on barrapunto about a photosharing website: riya
This flickr competitor is boasting an interesting feature:
It can recognises faces and text within the photos. Like an OCR software, the service need to be trained to recognise faces of person appearing on your photos.
I think the idea is you can then search your and others' photo using these criteria (who is in it?, what photos haved this text?)

Although innovative, I think an online service is not the best place for this technology.
I'd rather see it in my desktop photo management software.

It's definitely a competitive advantage over flickr, but when one day someone will create a mashup that use any photo sharing service with that kind of technology - which is the way not uncommon for "Web 2.0" apps - it won't be anymore.

In the other hand, in a web of integration of data and services, one service has to differentiate itself from the others. But these differentiation need to be presented via APIs and not as end user features.

Now riya has got the attention of everybody including competitors. If this feature was presented as an API to developpers, the benefits for riya would have been greater, has third parties would exercice it while keeping a low profile.