Version history is built into OS X, and will be a part of Windows 8, if I recall correctly. It's effectively file journaling, but with an elephant's memory. So you can go back to any saved version of a given file, modulo certain constraints (momentary copies are only kept for a day, hourly copies are only kept for a month, daily copies are only kept for three months...)
But as for remote device file linking, that sounds like an identity and verification nightmare. The best-case scenario is a cloud file system, like Dropbox. (or iCloud, for that matter) But that's less file linking and more reducing each user's device to a smart terminal operating on a common data store.
He's describing metadata, which most file formats have always had, but extended to content relevance. Metatags are a user-accessible variety of metadata, but it's somewhat problematic (see below).
The challenge of well-formed metadata is the demand on the user that it makes. Somebody's got to mark all those files with "My loved one", "my cat", "lunch", and so on. There's been a moderate amount of research about this problem as a human factors issue and the upshot is that very few people will continue to make a diligent effort at it, since it basically amounts to investing time and brainpower into a lot of random work, and the rewards never seem to come (how often will you be searching your photo library for, say, bees?)
Mac OS X has been slowly and kind of awkwardly evolving its file search to infer metadata automatically, but sometimes being not quite good enough is about as useful as not being good at all. And that's a large part of the problem. (Google Images has been increasingly clever in its image searches, but they have a lot of resources, both in terms of data to throw at the problem and in terms of clever people who can be paid to spend all day thinking hard about the problem.)
There are many different kinds of search, and many kinds of search strategies. For example, search often involves recovering a specific previously-experienced thing ("I remember this photo, it had a dog, and it was sunny, and there was a blotch of some kind in the lower right corner, and it had my late aunt in it..."). In the case of automated metadata generation, if the system fails to identify your late aunt, and you've misremembered the breed of dog, and it didn't know the blotch was relevant to your recollection of it, it will have failed your search. If you had manually tagged it, and misspelled your late aunt's name, couldn't remember the breed of dog at the time and just put "dog", and then went on to the next photo to tag, you will have failed at preparing the image for your own search.
I've been on the periphery of some of the research about how users search for information -- on the web and elsewhere -- and I find it fascinating stuff.
