Here’s how it works: You and another player come up with labels or tags for a randomly-selected image from the GIS database. Neither player can see the other player’s suggestions at any time. Once you and your partner agree on a keyword, you get 100 points and the game selects another picture to label. Each game lasts just one and a half minutes, and in that time you have to match as many tags as possible to get a high score.
Google Blogoscoped explains more about its implications for GIS:
More than a game, for Google this is a way to tag images using human brain power… to improve their image search results. Two people finding the same tag can serve as validation the tag makes sense. I suppose for Google it’s not important that two people find the same keywords at the same time – they can simply let people tag the images and then add any threshold they want (like “4 people must have chosen this tag for it to become a confirmed tag”).
As TechCrunch reports, it’s a variant of the popular ESPgame which does pretty much the same thing, with a similarly ambitious goal of tagging the whole web. The ESPgame website claims that if enough people played this sort of game, all images on the web could be categorized in “a matter of weeks”. I’m not so sure, but the “just one more go” factor for Image Labeler is undeniably strong, making this the latest addition to my advanced work avoidance tactics. Yeah, this is the kind of thing lifehacker warned me about.
I’ve been playing on and off since yesterday, and my current highest score is 800 for a single game and around 8000 in total. Of course, this being the Internet, some players are regularly getting 2000-plus per game. The player at the top of the leaderboard already has 1.2 million points. By my calculations that means he or she must have correctly tagged over 12,000 images, not including passes. Since last Friday. Google Image Labeler: Serious business.
It’s not perfect, though. One problem is that the images are regularly too small to see. Often I’ll have to lean in and squint at the screen just to work out which way up it is. This is obviously for copyright purposes, and it’s the same reason that GIS itself only stores thumbnail-sized versions of any searched pictures. But when the whole point of the game is to accurately describe what you’re seeing, it’s a bit of a design flaw.
I also wonder whether the timed nature of the game will make the results less reliable. Since you’ve only got one and a half minutes to tag as many images as possible, you’ll inevitably think of only the most general descriptions to ensure a match. I’ve lost count of the number of images where the matching tag was “man”, when the point of the image was what the man was doing (digging a garden, climbing a ladder, giving a speech), not just the fact that it had a man in it.