Automated User-Centered Reasoning and Acquisition System

Principal Investigator:  Vinay K Chaudhri Mailing address: AI Center SRI International 333 Ravenswood Avenue Menlo Park, CA 94025-3493 AIC Program:  Representation and Reasoning

Digital Aristotle is an application that can understand and reason over a growing number of scientific disciplines, and explain its knowledge in a user and domain appropriate manner. To contribute to this goal, SRI International (SRI) proposes to develop an Automated User-centered Reasoning and Acquisition System (AURA) to allow subject matter experts (SMEs), for example, biologists, physicists, and chemists, and scientific educators to formulate knowledge and questions to query that knowledge, with an ever-decreasing reliance on knowledge engineers.

We will address the design and implementation challenges as a team consisting of SRI International, University of Texas at Austin (UT), Human Computer Interaction (HCI) Institute at Carnegie Mellon University (CMU), and the Boeing Company (Boeing). SRI, UT, and Boeing were the top performing team in the Halo Phase I. The HCI institute at CMU is a leading center doing pioneering research at the intersection of AI and HCI.

The innovative aspects of this project are as follows.

Module-centered knowledge acquisition: AURA will feature a novel document-centric knowledge acquisition interface. This interface will be a radically new application of the concept of graphical assembly of knowledge from components [1] and will test the hypothesis that circumscribed knowledge fragments, in particular paragraphs from a textbook, can indeed be formulated by subject matter experts in a modular fashion using self-contained and well-packaged components. If successful, this will substantially reduce the cost of knowledge formation as compared to what we saw during Halo Pilot Phase. The success of this hypothesis will also set the stage for the large scale construction of a knowledge base that is central for realizing the dream of Digital Aristotle.

Natural Query Communication: Human domain knowledge is naturally expressed in a variety of forms, e.g., graphs, text, and diagrams, in sharp contrast to the cryptic and rigid input formalisms which a machine typically requires. To begin to bridge this gap, AURA will feature a novel, multi-modal interface for question-asking, allowing users to query the knowledge base in a graceful combination of controlled natural language, diagrams, and graphs. This will significantly enhance the usability and accessibility of the final system to users. The question formulation interface like the one proposed here will enable end-users to naturally state their questions to Digital Aristotle.

Usability design by a marriage of technology, behavioral science, and fine arts design techniques: We will develop innovative designs of AURA using the informed intuition of professional designers steeped in the knowledge of both SMEs and the KR developers on the team. We will use computational models of the skilled user to design efficient interfaces for the SMEs who are expected to encode vast amounts of knowledge, even before system prototypes are available for user testing. We will use a new approach to software architecture design that supports common usability concerns in the very earliest stage of implementation and allows us to continue UI design activities further into the implementation stage. This approach to design is the hallmark of the HCI Institute of CMU, is unique in the world, and will ensure wide acceptability of the Digital Aristotle amongst user communities.

AURA will build on our proven success in developing the knowledge acquisition system SHAKEN for DARPA’s Rapid Knowledge Formation project and in using KM as an inference engine during Vulcan’s Halo Pilot project. By enhancing our already proven team with world class user-centered design experts, and domain experts, and substantial engineering talent, we are well positioned to take on the challenge outlined in the CFP.

AURA: Automated User-Centered Reasoning and Acquisition System The AURA system enables graduate students in Physics, Chemistry, and Biology to author knowledge from a science textbook, and support high school students to formulate questions against that knowledge to obtain answers.

• Vinay K. Chaudhri, Mark Greaves, Daniel Hansch, Anthony Jameson, Frederik Pfisterer. Using a Semantic Wiki as a Knowledge Source for Question Answering. AAAI Spring Symposium, 2008.  [PDF, Details]

• Spaulding, A. and Chaudhri, V. K. and John, B. E. and Prevas, G. and Mishra, S. and Pacheco, J. Using Equations in Concept Maps to Graphically Build Knowledge Bases. 2007.  [PDF, Details]

• Chaudhri, V., John, B.E., Mishra, S., Pacheco, J., Porter, B., Spaulding, A. Enabling Experts to Build Knowledge Bases from Science Textbooks, in to appear in Proceedings of The Fourth International Conference on Knowledge Capture (K-CAP), October 2007.  [PDF, Details]

• Chaw, S.-Y., Fan, J.J., Tecuci, D.G., Yeh, P.Z. Capturing a Taxonomy of Failures During Automatic. to appear in the Proceedings of The Fourth International Conference on Knowledge Capture (K-CAP), October 2007.  [PDF, Details]

• Clark, P., Chaw, S.-Y., Barker, K., Chaudhri, V., Harrison, P., Fan, J., John, B., Porter, B., Spaulding, A., Thompson, J., Yeh, P.Z. Capturing and Answering Questions Posed to a Knowledge-Based System. to appear in the Proceedings of The Fourth International Conference on Knowledge Capture (K-CAP), October 2007.  [PDF, Details]