Folksonomy and the ESP Game

Somebody in the physical world told be about Carnegie Mellons ESP Game. (I know it was a physical person because I found the url on a piece of paper on my desk — I’m not sure who though, if it was you, speak up in the comments below.) It’s fun itself, and also relevant to a new meme that’s been floating about lately: “bottom-up social classification”.

In the same week as I was tipped to the ESP Game, I attended a brownbag lunch at work discussing meta data, primarily as it applies to photos. Of note was the “community classification” offered by the photo service Flickr. Obviously, classification and meta data are crucial as the amount of information continues to skyrocket.

How to do it is much more difficult than identifying the need.

This “bottom-up social classification” idea has been generating a bunch of conversation lately, nicely summarized by Gene. This concept has been coined a folksonomy by Thomas.

You can read more about Folksonomy in thoughtful posts by Victor, Jess, Stewart and Alex.

So, the ESP Game is pretty similar, which is why I’ve combined these posts. The ESP Game is a Java-based game that facilitates two people viewing a single image at the same time. You can read about how to play, but basically two people are looking at the same image. They both type labels describing the picture. When they’ve both identified the same label, a match is declared and both players earn points. The points are the reward to the humans — the high-quality, double-human-confirmed keyword/label associated with an image is the reward to the system that CMU is building.

(Previously identified labels for a specific picture are then taboo words, and aren’t allowed to be used in the game. This ensures that additional — generally more precise — labels are continually identified and captured.)

They’ve already collected 3.726 million human-confirmed labels. I’ll let them describe the value of this:

Labeling an image means associating word descriptions to it, as shown below. Computer programs can’t yet determine the contents of arbitrary images, but the ESP game provides a novel method of labeling them: players get to have fun as they help us determine their contents. If the ESP game is played as much as other popular online games, we estimate that all the images on the Web can be labeled in a matter of weeks!

Having proper labels associated to each image on the Internet would allow for very accurate image search, would improve the accessibility of the Web (by providing word descriptions of all images to visually impaired individuals), and would help users block inappropriate (e.g., pornographic) images from their computers.

Help out and have fun.