O’Reilly nails it in their description of it.
The launch of Google Image Labeler, a “game” that asks people to label images, and figures that images given the same label by multiple people are likely to be correct, continues the Web 2.0 trend towards bionic software, that is, software that combines machine and human intelligence. This is really just another version of the web 2.0 principle, harnessing collective intelligence, but with an emphasis on “harnessing” rather than on “collective.”
Like Distributed Proofreaders (the granddaddy in the space), Amazon’s Mechanical Turk, and mycroft, but unlike, say, a Flickr tag cloud as a reflection of collective labeling of images, Google Image Labeler puts people explicitly to work.
There’s a spectrum of ways to put humans to work refining computer results, from the implicit to the explicit. The most explicit, of course, is going to be when the third world job shops now engaged in making booty for World of Warcraft start offering their services for more general hire.
[UPDATE : O’Reilly continues their investigation looking at the roots of the Image Labeller in the ESP Game]