Friday, January 23, 2009

Now That's Using Your Noggin

Remember that SETI screensaver from the early days of the Internet? Brilliant, right? Harness unused CPU cycles from desktops around the world to solve a hard problem broken into bits. But we didn't find aliens. Or if we did, no one told us. And we probably sucked more electricity out of the wall causing global warming which may have increased our need for an alien lifeline. But that's not the point.

There's another company who's business is founded on applying the exact same principle. Google. They have hundreds of thousands if not millions of cheap servers breaking down search and other computational heavy problems into tiny bits and then reassembling them to get the answer. They call the algorithm MapReduce and have given a watered down version to the world in the open-source Hadoop. And when you get hired by Google they ask you, "What problem would you solve if you had limitless computational power?" Now that's employee onboarding!

But there are lots of problems computers are terrible at solving. Image recognition. Language recognition. Complex pattern matching. But the human brain is pretty good at these tasks. (Great book on this is On Intelligence, by Jeff Hawkins). What if you could coordinate a massive human effort to solve these problems? What if you could get everyone in line more than the North Koreans? What problem would you solve?

Well, like many great technological movements, the pornography industry has figured it out. You can now get free, quality porn on the Internet if you just solve some "Captchas" for them. Captchas are those letter scrambles that make sure you are human when you go to Ticketmaster or register for a new service. Here is an example:

Well, if you can get someone to crack these, you can start doing nefarious things like infiltrate and spam people on other services or reserve those tickets I want. Since that makes money, it's a different revenue stream for the porn industry. Not quite legal though. And certainly not admirable.

But look to our scientists. They have taken a similar principle and applied to protein analysis. And it's an app called FoldIt. How cool is that? By getting users to solve 3D problems folding proteins you are actually contributing to a medical database about how to target proteins with drugs.




http://www.youtube.com/watch?v=lGYJyur4FUA

Imagine if every turn of your Rubik's cube contributed towards a cure for AIDS? Now that's pretty cool. It would be even better if you abstracted the problem so you didn't know that it was a protein problem.

What would you do if you had unlimited human analysis and a way to incentivize them to apply it?

  • Find anomolies in the US budget? The new accounting test and game brought to you by Obamatronics
  • Choose the hottest person in America. Facebook profile challenge creates a trillion page views and makes the company profitable through contextual ads for beauty enhancement products. Ok. Less cool. More fun.
  • Transcribe all the video in the world? Language learning game brought to you by the Google Library project.

There are some cool problems we could solve and we have barely touched the surface.

1 comment:

Brian Dalessandro said...

Hey Trevor,
Check out Amazon's Mechanical Turk if you haven't heard of this already. Its a marketplace for distributing labor intensive but easy tasks to millions of participants.

Also, one of the hurdles of pattern recognition is being able to labeling the entries in your database for eventual pattern discovery. This form of labor distribution enables researchers to gather the data that only the human mind can really process in a cheap and efficient manner. The protein case you site is an excellent example.