We were asked to write a description for a YouTube movie, called Human Computation. My description is pretty long, but just deal with it. :0
If you want to see the movie yourself, it's located here --> http://www.youtube.com/watch?v=qlzM3zcd-lk
The movie, if you want to call it that, starts out by talking about "Captchas", which are those little scrambled words or sentances that people have to decode when they register on websites. They have them so that it's a lot harder for hackers/spammers to mass produce accounts on pages. The real aim of the presentation as a whole is to talk about human computation, or the things that humans can do but that computers can not yet do on their own, like the "captcha" problem.
The next example the speaker brings up is labeling images with words. Describing and identifying something visual with a written response is very hard to do for a computer. Basically, one of the ways a computer can translate between images and text is with filenames or tags or something similar, like when you do an image search on Google. This method is obviously not very accurate. :0 The speaker wants a method for labeling all the images that appear on the web. And accurately so, I persume. Fast and cheap, too.
Humans will need to lable the images, as computers don't have the ability to do so. The speaker proposes an online game where players will see images, and then get to suggest words that describe that image. Then, the words are matched up together and the words the players had in common become describers for the picture. Pretty damn smart, if you ask me. It's a little more complicated than that, but that's the basic idea. It's apparently called "The ESP Game." They have apparently already been running it for a while to great success.
Next, the speaker talks about locating actual objects in the images, like distinguishing what exact areas in an image a specific object is located. This problem is also solved with a game, this one called "Peekaboom", and is apparently linked with "The ESP Game". One player clicks in an image, and thus reveals an area in the picture to the other player, which can only see the selected area. Player 1 can see a word describing the image, and player 2 has to guess what the word is. Once again, simple yet ingenious. :) Using this method, and by combining a bunch of testresults, you end up with a computer that can locate a specific word in an image.
After this, the speaker moves on to common-sense facts. Ideas and things people have in their heads, billions at a time, like "water is wet" or whatever. To make a computer able to understand such ideas, the speaker suggests ANOTHER game, this one called "Verbosity." Here, player 1 has a word, and adds a sentance to it, then only shows the other player the sentance without the word. Like, "X is wet", and player 2 can guess that "X" is water, and they would be correct, or something.
The speaker calls these types of games: "Asymmetric Verification Games"
The basic idea is, Player 1 is given an input, and they give an output on that. Player 2 is shown the output, and needs to guess what the original input that Player 1 was given.
Apparently, just the first game was a SYMMETRIC game, rather than asymmetric, cause both players are given an input and need to match eachothers outputs.
Then the speaker goes on to compare the two things to eachother. :0
In symetric games: Constraint is number of outputs per input.
In asymetric games: Constraint is number of inputs that yield the same output.
The general idea he suggests is that when a computer can't solve a problem on it's own, you could provide a test for humans to interact with so that the computer and the human can solve the problem together.
He keeps hinting that this will eventually turn into The Matrix. =P
Then it's question time! And I can't be arsed to write the rest down. :|
Now I'll go enjoy my time off from school instead! :D
-Fredrik "Leo" Gustafsson