Games with a Purpose: GWAP.com

Checking over my feeds revealed that Luis von Ahn's group has released their anticipated Games With A Purpose website gwap.com (cheers O’Reilly Radar). The site provides an umbrella for web-based games, from which the solutions contribute to solving AI-complete computer vision problems. The current suite of games include:

• The ESP Game: Match tags for an image with a partner.
  
• Tag a Tune: Describe a tune and guess whether your partner is listening to the same tune.
  
• Verbosity: A word game, where you describe a secret word by giving clues.
  
• Squigl: Outline parts of an image and match with a partner.
  
• Matchin: Select favorite images and match with a partner.
  

Two of von Ahn's products not released under the umbrella (for whatever reason) include Peekaboom (reveal parts of an image to a partner) and Phetch (find images on the web that match a description). Related products include the implementation of ESP for Google called the Google Image Labeler, and reCAPTCHA.

The tag line for the site is: You play the games, Computers get smarter, Everyone benefits! There is also a promotional video for the site on YouTube. All games are clear, simple, fun and centered around computer vision problems that the current state of algorithms find difficult. Signing up provides a general GAWP account with a user profile that includes an aggregate score, a level that reflects experience, and a break down of contributions to each game in the suite. The site is pretty in that is very graphics heavy.

All in all, the portal is impressive, and the games are great, although as a technical kind of guy I am more interested in how each game contributes to their image database (beyond my wild assumptions), and how they plan on exploiting (harnessing) this information. They site is branded Carnegie Mellon University like von Ahn's other products, suggesting ownership and financial interests beyond von Ahn's affiliation.

I suspect that all of these products have an academic driving force rather than commercial, which raises the question as to how such products could be exploited toward commercial means (basic monetisation). I have little doubt that such database or perhaps the data collection frameworks would have a cash value to someone outside of academia. Interestingly, a similar human computation based computer vision project called Polar Rose (that grew out of a research project from two Sweden universities) have $5.1M in venture capital.

I will continue to study von Ahn's work, towards improving my notions of a human-powered NP-complete problem solving service.