Bruce Porter Rich Mallory Peter Clark Art Souther Fred Prado Charles Callaway and (not shown above): Carl Andersen, Steve Correl.
During the past eight years, we have built a large knowledge base in one area of biology, and developed methods for automatically answering a variety of questions using the knowledge base. Containing about 180,000 facts concerning 30,000 concepts, our knowledge base is one of the largest of its kind (i.e. its content is structured and formally represented). In addition to expanding this knowledge base, we are also beginning to construct similar knowledge bases in other domains, most notably, the domain of Distributed Computing.
We are especially encouraged by the results from using our knowledge base for a variety of AI tasks. Most recently, James Lester used the biology knowledge base to test his system for explanation generation. The system generated about 60 explanations, expressed in English, concerning biological objects and events. In a controlled experiment, domain experts found little difference between these explanations and those written by their colleagues.
Currently, we are extending the types of questions that can be answered using automated reasoning with a large knowledge base. Jeff Rickel developed a method for "compositional modeling", the task of constructing a model appropriate for answering a prediction ("what-if") question. Performing this task well requires building the simplest model that can adequately answer the question - a daunting requirement since knowledge bases like ours implicitly contain MANY models at numerous levels of detail. The Qualitative Process Compiler and QSIM are used to simulate the models built by Jeff's program.
Finally, we are testing the generality of our research results by building a knowledge base in another domain - distributed computing environments (focussing on OSF's DCE) - to construct a help-desk assistant for automatically answering a proportion of customer's questions which would otherwise be phoned in to a normal help-desk.