Knowledge Collection from Volunteer Contributors
Papers from the AAAI Spring Symposium
Timothy Chklovski, Pedro Domingos, Henry Lieberman, Rada Mihalcea, and Push Singh Cochairs
Many AI tasks depend on having large amounts of knowledge and data. There are knowledge bases to be constructed, corpora to be tagged, long training sessions, and so forth. Such resources are critical to our success, but building them can be difficult and time-consuming. What if we could farm out most of that work to thousands of volunteers on the internet?
Collecting knowledge from volunteer contributors (unstructured or with contributors organized into tiers and classes) can potentially both enable new, more knowledge intensive approaches to the current open problems in many subfields of AI and allow the field to tackle new challenges. We emphasize applications collecting semantic data—from knowledge that aids reasoning about everyday world to linguistically annotated corpora to bring knowledge-rich processing to NLP applications. Additional emphasis is on applications that benefit from having such knowledge available—including both aspects of their construction and their role as motivator and validator of users’ contributions.
There are many open challenges in turning to the general public for help. On one hand, the systems need to collect useful knowledge (often, knowledge usable with current reasoning methods, sufficiently unambiguous and cross-validated, of useful breadth and depth). On the other, collection needs to have some payoff to the user that exceeds the effort of providing the knowledge. Collection needs to be fun and engaging to attract enough volunteers, by, among other things decomposing hard problems into “bite-sized” chunks that the average person could solve quickly.