Humans vs Computers

I’ve been searching through Google Video looking for good presentations that I can use as resources.  I am mainly looking for presentation styles, but much of the content has been interesting.  Here is one I stumbled across today.

Luis von Ahn, a professor at Carnegie Mellon University, is talking about Human Computation.  He looks at several things that computers just can’t figure out yet.  One of those things is a captcha.  This is a term coined by Prof Ahn.  You have probably seen a few of them during your browsing.  They look like this:

 

Humans can read the letters, but computers cannot.

Here’s the first really interesting statistic from the presentation. 

9 Billion Human-Hours of Solitaire were played in 2003. 

That is billion with a B.  In comparison, the Empire State Building took 7 million (with an M) human-hours to build.  That is equivalent to 6.8 hours of the world playing Solitaire.

Ahn’s goal is to put this wasted Solitaire time to good use.  He has devised a program that is fun and competitive that will help correctly label pictures on the web.  Before showing the game, he went to Google Images and searched for “dog”.  There were a lot of dogs, a rabbit and a guy wearing a blue suite.  This is because the image identification process cannot analyze the picture to identify a dog.  Instead, Google relies on the file name and words around the picture to try and identify a picture.

The ESP Game solves this problem.  The game pairs two anonymous people and shows them both the same picture.  As quickly as possible both players (without seeing what the other is typing) must try to type a word that identifies the picture.  As soon as the two players have a common word, points are scored.  Here’s the twist.  Certain words are “taboo” words.  Generally those are the obvious things in the picture.  It requires the players to look at all the items in the picture in choosing descriptive words.

The end result is a very accurate list of descriptive words for each picture in the database.  Plus the players can get millions of points (the top ten are listed on the site).

Ahn estimated that all the pictures in Google’s database could be accurately tagged in about two months if about 5000 people were to play this game around the clock.  He pointed out games in Yahoo and MSN that average 5000 players all the time.  The whole project is an interesting way of using human brain power.

Once you master the ESP Game, give Peekaboom a try.

This entry was posted in edtech and tagged . Bookmark the permalink.