Well, like many, many other folk, I got temporarily hooked on the new Google Image Labeler before Digg and other flash mob assemblers managed to slow the site down to a crawl. This addictive little toy is wonderful in its simplicity yet frankly quite profound in it's implications.
The "theory" behind the tool is discussed at some length by Luis von Ahn of CMU in a Google engEDU video entitled "Human Computation". That video is almost as much fun as the Google Image Labeler is to play...
The basic idea behind Labeler is to create a two player online game similar to Ahn's "ESP Game." Two players are randomly connected and both shown an image. Then, each player inputs keywords that describe the image being displayed. If both players type the same keyword, a new image is shown and the keyword guessing starts again. This continues for two minutes until the game is up. The players then get points based on how many matches they had.
Based on initial results and measurements that showed that large numbers of people were willing to spend as many as 12 hours a day playing the game, Ahn estimates that 5,000 players would be needed to label all the images in Google Images in less about two months. This is just one example of what can be done with "Games with a purpose"... Ahn classifies the Labeler and ESPGame as "symmetric verification games" and in his talk describes the principles for creating "asymmetric verification games" as well.
SETI@home, folding@home, BOINC and similar systems showed us how to convert the otherwise wasted spare cycles of millions of computers into useful work by offering public-spirited computer owners a way to contribute those cycles to useful projects. Amazon's Mechanical Turk showed that by offering people money, spare human processing cycles could be captured and managed for commercial purposes on demand. The StarDust@home project showed that you don't have to pay people for spare brain-cycles if they think they are contributing to something worthwhile. Now, Ahn and Google are demonstrating that building games and entertaining people might be an exceptionally powerful way of capturing spare human cycles. And there are quite a few of them available!
Ahn estimates that during 2003, 9 billion human-hours were consumed by just by people playing Solitaire on their computers. Clearly, there's quite a resource there to be captured. To provide some scale to that number, Ahn shows that the Empire State Building in NYC took only 7 million human-hours to build (6.8 hours of Solitaire play) and the larger Panama Canal took only 20 million human-hours to build (less than a day of Solitaire.)
Perhaps we will see the coining of a new unit of measure:
1 "Sol" equals the amount of human processing devoted to Solitaire during one hour in 2003...
If Ahn's predictions are correct, and if the inevitably large number of people "play" with the Google Image Labeler, it would seem that Google will once again lengthen its lead over Microsoft, Yahoo and others by providing the world's best and most comprehensive image search capability. Google's competition is currently limited to indexing images based on surrounding text, alt text, or primitive image understanding algorithms. I can't help wondering if we'll soon see the Google's competitors soon release their own "games" in an effort to improve their indexes. (How will they make them good enough to attract users from ESP Game and Labeler? Is there a way to combine these games with Second Life or Halo? Can Yahoo! expand on what they have already been doing with exploiting Flickr tags?) Will the "best" search engine one day be the one that offers the most effective and most fun indexing "games"?
I think it goes without saying that "Games with a purpose" and "Human Computing" will be concepts that we'll be hearing quite a bit about in the future.
bob wyman
Comments