Home > 1. technology > Stumbling onto human computation

Stumbling onto human computation

September 24th, 2007

I have recently become a big fan of StumbleUpon. With a single click, you can transported to wild and wonderful things on the Internet based on recommendations from other web surfers. Using this service, I stumbled on this amazing video on human computation. After watching the video, I realize that StumbleUpon itself is an example of human computation. Here’s professor Luis von Ahn‘s 60 minute or so presentation:

What if we just think about humans as a very specialized kind of processor? Human computation is a novel take on artificial intelligence (AI) problems like classification and ontologies. Traditionally, AI would use edge detection or some other image processing technique to classify a picture. But no one has been successful in using these technique to find common objects like cars, celebrities, etc. Human computation asks: why ask computers to do what people can do automatically and instantly? Is there some way to harness the 9 billion man-hours wasted playing games like tetris to do accomplish something productive?

Professor Ahn’s first game is called ESPgame. You sign up and become teamed up with someone else from across the Internet. Simultaneously, you enter words to describe a random image. If you type in the same word, you get points for a “match.” The most common words to describe an image become “taboo” as the game advances and the players have to search for more subtle words that describe the image.

So far, hundreds of millions of images have been classified by hundreds of thousands of people. He has two other notable games: Peekaboom, a game for locating objects within an image, Verbosity, a game for collecting common-sense facts about the world.

I actually like the idea of human computation because it applied to an even wider class of internet services than the games that Ahn has created. For example, google’s search engine is really human computation: it relies on links from sites like this blog that were chosen by real people to determine the order of the search results. Similarly, both Amazon’s recommendation feature (“people who bought this also bought..”) and even the user reviews can be considered human computation. And the tagging services like del.icio.us, my music service last.fm also fits into the category as well as StumbleUpon, the place I discovered this idea originally.

1. technology , ,

  1. September 25th, 2007 at 12:23 | #1

    And don’t forget the Mechanical Turk from Amazon. It goes one step further and creates a marketplace for buyers to purchase human intelligence cycles and sellers to get paid for expending those cycles.

  2. September 25th, 2007 at 16:48 | #2

    Yeah, I saw a mention of that site in a crummy NY Times article on Sunday on artificial intelligence. I finally checked it out because of your reminder. It’s an interesting concept, but I wouldn’t necessarily say “takes it one step further.” Von Ahn’s work has two important features that mturk misses: (1) a fun aspect and (2) built in verification. And mturk may also be plagued by the fact that such web markets for unstandardized goods and low prices are tough to work effectively.

    Thanks for the comment and the reference!

  1. No trackbacks yet.
Olark Livehelp