Cyc Design Challenges and Solutions
                                Keith Goolsbey
                                 Cycorp Inc.
                         3721 Executive Center Drive
                         Suite 100, Austin, TX 78731


                                      Abstract
    Cyc comprises a large, contextualized, common sense knowledge base (KB)
which is encoded in an expressive representation language (essentially FOL with a
few key extensions) and paired with an inference engine optimized for the classes of
queries we most frequently encounter. These queries tend to mix relatively shallow
reasoning within one of a large number of idiosyncratic subtheories with relatively
deep reasoning within one of a very small number of stylized subtheories. The con-
straints of these queries in a large and expressive KB combined with the need to
efficiently react to KB elaboration together provide a unique set of design challenges
that are extremely stressful for the solutions provided by the current state of the art
FOL theorem provers. The solutions to these challenges currently adopted by the
Cyc inference engine will be presented within the context of a new suite of TPTP
problems that are derived from Cyc’s KB and typical queries and are intended to
demonstrate Cyc’s design challenges for investigation by the wider community.