You really have to check out this presentation on Google Video. Not because the presenter is named “Luis,” but because the presentation he gives is really fascinating. I got wind of it from the Google Blog, and appropriately enough, it is related to search engines. Specifically, it focuses on computer/human intelligence, for example, labels given to photographs on the web.

Professor Luis von Ahn has worked out a very clever way of labeling images on the web, using “wasted” computational cycles… of human brains. He points out that billions of human work hours are spent on playing Solitaire, thousands of times more each year than were used to build the Empire State Building or the Panama Canal. He also points out that most images on the web do not carry labels, which are (a) the only way that visually impaired people can know what an image on a web page is, and (b) necessary for good search engine results when looking for images.

He solves this problem by making a game called The ESP Game (go ahead, give it a try!). In this game, two complete strangers from the Internet, logged on to the game at random, are paired together, but they cannot communicate. They are both shown the same image and asked to type words that best describe the image. If they type the same word, they get points; the faster they guess the same word, the more points they get. As a result, the answers they give tend to be more accurate and generated in less time. As words become verified as belonging to the image more and more, they are added to a “taboo list” of words the players are not allowed to guess, ensuring that the labels for the image are extensive and complete.

Of course, even with protections against cheating (many people could conspire to log on at the same time and give the same answer to all images, for example), common misconceptions or even certain opinions tend to creep in. An image of Walter Matthau with a big mustache was not labeled with the actor’s name, but did get labeled “Saddam” and “Mr. Wilson” (Matthau played that character in Dennis the Menace). An image of George W. Bush, on the other hand, received the labels “President,” “George,” “Bush,” “Man,” “Dumb,” and “Yuck.”

One side benefit of this game: It helps people learn English. When you think about it, that makes great sense–enough so that I am going to suggest this game to my students as a way to increase their vocabulary–an area they often ask for help with and is difficult to do.

Another game that gives even further data on images is “Peekaboom,” which has one player slowly reveal an image to another based upon a keyword given for the image; as the image is revealed to the second player, they have to correctly guess the word the first player has been given. This lets the database know where in each photo any particular keyword is located. In addition to also being a good English-language game, Peekaboom also helps programmers design image recognition software.

There are other games, and a lot more very interesting stuff in the 51-minute presentation. If you have a free hour, I’d recommend giving it a look.