Building Better Computers With Games

Last week news spread of a game, Foldit, that had players helping scientists unlock the patterns of new proteins in hopes of discovering cures for cancer and AIDS. This week the games with a purpose news covers slightly less exciting ground.

In 2006 Google licensed the ESP game, later renamed "Google Image Labeler," developed by Carnegie Mellon researcher, Luis von Ahn. The game shows the same picture to two randomly selected players and asks them to label it. You get points if your partner chooses the same label, and Google gets a more accurate description of the image to help improve its image searches.

This week von Ahn released a website devoted to games with a similar purpose, called, naturally enough, Games with a Purpose. The site includes five games, the original ESP game, and four new ones working with the same idea. There's a music labeler, that asks players to identify the best mood or setting for a song, in the hope that, for instance, a search for, “drunk and alone,” will produce the best mix, rather than just tracks with those words in the title. There are also games aimed at creating better AI, including a game that has players tracing an outline around a specific figure in an image, or a word game, where one player helps the other guess a secret word. The last game seems like an extension of the ESP image labeler, only this one has players ranking pictures on their aesthetic quality, to create more aesthetically pleasing search results and to also throw the old assumption, “beauty is in the eye of the beholder,” out the window.

Speaking to MSNBC's unfortunately named, CosmicLog (why not 'blog?' This sounds like a divine #2), von Ahn describes the two aims of his site, “One is to train computers to do the things that humans do. And another thing we're trying to do is just use a different approach to experiments. For the first time, we can get hundreds of thousands or millions of people working on the same problem at the same time. ... We're seeing how we can coordinate millions of people to solve tiny bits of the problem." The answers he produces will be shared freely and could aid in AI research.

It's no cure for cancer, but could make your Google searches better. Though, at least with the Google Image Labeler, shouldn't they pay us to play that game?