(Warning: blue sky speculation and impratical idea-slinging follows)
So, I was watching a downloaded copy of an interesting talk sponsored by Google (and available on Google Video, natch: http://video.google.com/videoplay?docid=-8246463980976635143&q=Human+Com... ) called "Human Computation". It was given by [[Luis von Ahn]], and it was largely about his ideas on [[Human-based computation]], with a lot of attention given to [[ESP Game]] and PeekaBoom.
ESP Game, to summarize, is a quest to label images on the Web by making describing the image the product of a game in which two players negotiate the description.
In his talk, I think he estimated that 5000 players could label every image on the web in about 2 months.
Now, what struck me about this (aside from the ingenuity of figuring out a way to put human cycles to work) was how currently *pointless* this quest was. So Luis now has a database with descriptions of a portion of the web's images. He can make an alright image search engine. Maybe sell the data to Google who could obviously make use of it. Image search might improve a few percent (let's be optimistic). He even admitted as much in the questions section, that the loop isn't closed, but that it's just an "engineering perspective" (should I make a joke about academics here?).
The essential point is that this is an instance where the famous "Read-only Web" or Web 1.0 is a serious barrier to actually using the gathered info on a large scale, since you can't go in and add the devised descriptions to those images lacking them (which is useful for among other things, screen readers). The HTML and images are static. You can't analyse the relevant object and improve it. The source is closed.
But we here are fans of wikis, good ol' Web 2.0, the RW web. Why not apply these ideas to Wikipedia? Disambiguations are one possibility; categorization is another; image tagging or pace ESP Game, descriptions (I know we have to have a bunch of images on en or Commons which need descriptions; even a few disjointed words are an improvement on nothing). It's too bad Luis probably wouldn't want to use Commons instead of Google's image search, since he has everything all set up already. I mean, already we've got plenty of bots and rather complex software specialized for various obscure tasks. Why not a game?
I mean, critics are always saying Wikipedia is a crappy text-based MMORPG masquerading as an encyclopedia... why not prove them wrong and show that Wikipedia is a mediocre text-based puzzle game masquerading as an encyclopedia? :)
~maru