Want to help Google beef up its
images index? Want to kind of have some fun in the process?
Google Image Labeler is a new pseudo-game from Google designed to add meaningful keywords to the millions of images in its index while keeping you kind of entertained. It works like this: When you start the game you're assigned a random partner. Then you and your partner are shown the same random image, for which you must enter keywords. So, if it shows you a picture of a dog, you could enter dog, pet, terrier, etc. For every keyword you enter that matches a keyword your partner entered, you get 100 points. As soon as you and your partner enter a match, you go on to the next level, and you play until the two minute timer runs out. When the game is done, you're ranked against other players. So far I haven't been able to break the top 100--success seems to depend a lot on how good your partner is (or to what degree they think like you) and how good the images you get are. Some of the images in the game are pretty challenging to identify, and "mysterious industrial structure" doesn't make a very good keyword.
It's easy to see how this benefit's Google's index: If two randomly-selected people happen to enter the same keyword for the same image, it's a pretty good bet that it's an accurate keyword for that image. This is a pretty smooth move on Google's part, harnessing the brainpower of bored Google fans to do what computers can't currently do well--glean meaningful data from images. If Google can get--and keep--enough people playing the game, it could really strengthen their image search offering.
[Via
Google Blogscoped]