Putting party games to good use
Oct. 16th, 2007 12:02 amOther than just playing them, I mean.
Linky time. I recently happened across http://www.sirlin.net/ - which has a bunch of interesting articles about game design - but in particular, this video. It's about an hour long, so, to summarize: There are many problems that currently cannot be effectively solved by computers, such as labeling the key components of an image. Web image searching currently relies on file names and nearby text; obviously an image labeling system would help immensely. Obviously, there are many artificial intelligence researchers working on this problem, but rather than wait for them to come up with something, Luis von Ahn proposes we get humans to do it instead. He has set up "The ESP game" - a simple online game, which works as follows: Each player is paired up with another player, their partner. Both are presented with the same image, and are asked to type words that their partner will also type; no communication between partners is provided. When among the words guessed, there is one both partners have entered, both players get points - and that word is taken as a labeling of the image (as it came from two independent sources). Pre-existing labels on the pictures are "taboo", which both avoids redundancy and increases the challenge of the game. The thing is, the game is actually pretty fun (or apparently so, I haven't played it[0]); it is, after all, not that different from your average party game (and the "asymmetric verification games" he introduces later resemble party games even more). There are more details - preventing people collaborating to introduce false labels, similar games to accomplish other tasks and two general frameworks for doing so, actual numbers on how much this could compute, people's more detailed reactions to the ESP game (some are pretty interesting); watch the video for the whole thing. It's really cool.
-Harry
[0] ADDENDUM: Having now played it a few times, I'm not certain I'd exactly call it "fun", but it's certainly addictive.
Linky time. I recently happened across http://www.sirlin.net/ - which has a bunch of interesting articles about game design - but in particular, this video. It's about an hour long, so, to summarize: There are many problems that currently cannot be effectively solved by computers, such as labeling the key components of an image. Web image searching currently relies on file names and nearby text; obviously an image labeling system would help immensely. Obviously, there are many artificial intelligence researchers working on this problem, but rather than wait for them to come up with something, Luis von Ahn proposes we get humans to do it instead. He has set up "The ESP game" - a simple online game, which works as follows: Each player is paired up with another player, their partner. Both are presented with the same image, and are asked to type words that their partner will also type; no communication between partners is provided. When among the words guessed, there is one both partners have entered, both players get points - and that word is taken as a labeling of the image (as it came from two independent sources). Pre-existing labels on the pictures are "taboo", which both avoids redundancy and increases the challenge of the game. The thing is, the game is actually pretty fun (or apparently so, I haven't played it[0]); it is, after all, not that different from your average party game (and the "asymmetric verification games" he introduces later resemble party games even more). There are more details - preventing people collaborating to introduce false labels, similar games to accomplish other tasks and two general frameworks for doing so, actual numbers on how much this could compute, people's more detailed reactions to the ESP game (some are pretty interesting); watch the video for the whole thing. It's really cool.
-Harry
[0] ADDENDUM: Having now played it a few times, I'm not certain I'd exactly call it "fun", but it's certainly addictive.
