An Interface for Targeted Collection of Common Sense Knowledge Using a Mixture Model

Robert Speer, Jayant Krishnamurthy, Catherine Havasi, Dustin Smith, Henry Lieberman and Kenneth Arnold worked on developing a user interface to help build a "common sense" knowledge database. The main purpose of this interface was to get users to enjoy using this interface so that they are more comfortable and provide better training for the system.

In their system, common sense knowledge is represented as concepts and features. For example "door is part of a house". To build this data base, the team created a "20 Questions" interface. Their hypothesis was that creating a user interface that was interactive and enjoyable would be better at retaining users than a static data entry form. For their user study, they had some users use their 20 Questions system and other users just used a data entry form. They found that those who used 20 Questions were able to complete their task much faster than those who used static data entry. Users also felt that the 20 Questions interface was much more enjoyable and that they felt it adapted itself to their use a lot more.

There are just somethings that are easier done by humans. The problem is that most humans don't want to do these things. Interfaces like this one, interfaces that turn a task into a game, would be very effective at "tricking" people into accomplishing these tasks. As time goes on, I think human computing will more and more become an effective way of generating knowledge.