Human Computation

Just learn about this great idea from Dr. Luis von Ahn presentation here. The presentation is based on his PhD theses. I want to try to describe here, briefly, what he is doing. His idea has turn on lots of bulbs in my brain.

He start by telling the story of CAPTCHA (Completely Automated Public Turing test to tell Computer and Humans Apart). We all should by now probably know about CAPTCHA, and how does it help preventing spammers to get millions of free email account.

CAPTCHA is a computer-generated product that currently only human can solve but not other computer program. See the paradox? This is the basic idea of Human Computation. There are still areas where humans do better job than computers. And we do it well, way better than computers currently capable of.

What some of us may not know is that spammers have found a way to solve CAPTCHA. There are 2 basic ways (I myself knows about the first way, but only learn the second way from Luis' presentation).

First, spammers pay people to actually solve CAPTCHAs. But this has turn out costly for them. Second clever thing they done, is by using p*rn-sites. Spammers create p*rn sites that when visitors want to see more, they would have to enters words of a CAPTCHA (which behind the screen will then be submitted to Yahoo's email registration form). Being passionated p&rn lover, more often than not, those visitors will as fast as the can typed in the CAPTCHA :-)

The second example taught us a very interesting idea (in my words): We can use human computational power to solve computer-unsolved problems in a way that's actually "fun" for them, plus we don't need to spent a lot of money to pay them :-) And to know just how many wasted cycles of human computation is, Luis gave a figure: In 2003 estimated that there are > 30 billion man-hours wasted on playing the game Solitaire.

So what kind of problems that we may solve? Turn out there are a lot of them. Luis gave example of giving better descriptions to images on the web, so that a search for images could give better results.

To solve that problem Luis created (not another p*rn site, no!) 2 kinds of game he called symmetric and asymmetric game. One of the game he has created basically just ask people to put on words that describe images. The results are then used by google images search to help them provide correct images to people when they are searching for, say, 'cat'.

But that would take long time, right? you may ask. Well, you gotta see Luis' presentation to learn much about interesting statistics, shortly I can only say it does worked!

You can see, and try, the games at www.espgame.org and www.peekaboom.org. I just checked espgame site, and there it shows that they have been giving > 33 million image labels since October 2003. Wow.

Luis himself seems doesn't just stop there. He continued to do research on how to solve more interesting problems.

I noted one interesting question in the video: Would it be possible that for every boring job we do, we could create fun ways to do it? Of which Luis answered, I don't know, it would be great if we could figure out how to do this for every problem. But yes, this is an open problem.

An interesting open-problem,
So let's use up those wasted computation cycles :-)