The High Performance Knowledge Bases (HPKB)
project demonstrated that the teams of knowledge engineers working together
could create knowledge bases (KBs) roughly at the rate of 10K axioms/year for a
pre-specified task and evaluation criteria. The HPKB effort showed that it is
possible to create KBs by reusing the content of knowledge libraries, and it
demonstrated reuse rates ranging from 25% to 100%, depending on the application
and the knowledge engineer. It was acknowledged that the ability of a subject
matter expert (SME) to directly enter knowledge is essential to improve the KB
construction rates.
The SRI team is developing a system for direct knowledge entry by SMEs as an
integrated team of technology developers. The SRI team includes Boeing, Information
Sciences Institute (ISI) at University of Southern California, Northwestern
University, Pacific Sierra Research (PSR), Stanford University, University of
Massachusetts at Amherst, University of Texas at Austin, and University of West
Florida. Knowledge Systems Laboratory at Stanford, Pragati Systems, and
Massachusetts Insititute of Technology joined the team after the contract
award.
The claim of this effort is that SMEs, unassisted by AI technologists, can
assemble models of mechanisms and processes from components. These models are
both declarative and executable, so questions about the mechanisms and
processes can be answered by conventional inference methods (for example,
theorem proving and taxonomic inference) and by various task-specific methods
(for example, simulation, analogical reasoning, and problem-solving methods). A
related claim is that relatively few components, perhaps a few thousand, are
sufficient for SMEs to assemble models of virtually any mechanism or process.
We claim that these components are independent of domain, and that assembly
from components instantiated to a domain is a natural way for SMEs to create KB
content.
The research in this project exploits and extends previous work in the HPKB
project, as well as work in process description languages, qualitative physics,
systems dynamics, and simulation. One scientific innovation, and the principal
extension to Cyc and the "HPKB standard" of knowledge bases, is the
idea of declarative and executable models (DEMs) assembled from components. The
declarative aspect of DEMs supports conventional inference, whereas the
executable aspect supports reasoning by simulation. For example, the
declarative part of a model of aerosols is sufficient to answer questions like,
"Will a 5-micron filter afford protection against this aerosol?"
while the executable part is necessary to model the dispersal pattern of the
aerosol.
The development of libraries of components made available to SMEs via
restricted natural language based, graphical, or templatized interfaces is the
principal means by which logic-oriented knowledge representation formalisms
become accessible to ordinary users. Every modeling technology shows this
progression: Spreadsheets, finite-element packages, statistical packages, chemical
synthesis software, Macsyma and Mathematica, architectural and CAD packages,
graphics and HCI systems, etc., are accessible to ordinary users because they
offer libraries of components. As a practical matter, then, it makes sense to
provide SMEs with libraries of modeling components. As a scientific matter, we
believe we can develop components that represent how humans think about
mechanisms and processes.