What if games could do something practical while they entertain us? What if by playing games you weren't simply entertaining yourself and others, but adding to the grand sum of human knowledge? This is the idea behind an ongoing academic project entitled "Games With A Purpose." The project, which focuses on the work of a young assistant professor at Carnegie Mellon, Luis Von Ahn, has one specific objective: to create games with useful computational side-effects.
Von Ahn observed that billions of hours of our time are sunk into games as minor as Windows solitaire. He then speculated that this wasted time could be put to practical use. In fact, Von Ahn calculated, more man-hours go into solitaire each month than were expended on building the entire Panama Canal between 1904 and 1914. If someone could come up with a methodology for harnessing these boundless intellectual energies, who knows what could be accomplished?
Of course, it's one thing to point out the "wasted human cycles" being lost to games, but it's quite another to find something humans could compute better than artificial systems. Von Ahn, as it turns out, had a pretty good idea about what humans could do better, and he had plans to put it to use. The human brain has a number of talents that remain beyond the remit of artificial intelligence, especially when it comes to things like visual processing. Describing what we see in a picture, for example, comes naturally to us, but it is still too complex for computers. This is where Von Ahn's concept first found application: in image labeling for Google image search. It's a splendid example of the kind of thing humans are good at, and where our brains are the only way to come up with useful data. Automated web-trawling programs can only read file names, alt-text or captions attached to images. Only an image that has been directly labeled by a human being can carry the kind of accurate information we'd need to know what it was a picture of. Labeling can make image searches more accurate and therefore far more useful to users. Getting real people to manually label images is the best way to have a useful image search on the web.
However, getting people to carry out massive and repetitive tasks is costly. Google would need a small nation of workers to get even a fraction of the images on the web properly labeled. Von Ahn's solution had gaming's compulsive power on its side: Not only would humans be manually labeling images for Google, they'd be doing it for free. In fact, gamers would want to do it, because the labeling process was all part of a guessing game called The ESP Game.
This online game (which you can go and play right now) throws up a random image that two players can both see. These two players cannot confer or communicate in any way, but must nevertheless come up with the same word to describe the image. They must also do it without using any of a list of banned words. So if it was a picture of a flowery dress, the words "flowers" and "dress" might be off-limits, and the pair would have to agree on another description, perhaps "gown" or "embroidered." Finding slightly more esoteric matches and collating this data across thousands of games, and therefore tens of thousands of images, allows The ESP Game to generate useful, applicable tags for vast numbers of images. This in turn makes image search vastly more accurate, and it's all done for the cost of putting together a rudimentary online game.