How Models of Creativity and Analogy Need to Answer the Tailorability Concern - Computational Creativity Research: Towards Creative Machines - page 99

kind of everyday analogical thinking that may not be goal-oriented in nature. PAI,
however, provides a tool for measuring those abilities which, at least at the surface,
don't rely on directed problem-solving, such as reading comprehension. Addition-
ally, it is difficult to imagine that any research program in AGI would be able to
demonstrate clear progress without showing increased performance in an ability
that can be measured according to some psychometric test. Non-goal-oriented ana-
logical reasoning is a good example of this principle: If the cognitive processes
underlying normal analogical reasoning when it is non-goal-oriented (as in everyday
reasoning) and when it is goal-oriented (as during psychometric testing) are largely
the same, then an artificial system capable of performing the latter has a strong
claim to performing the former. A system that only has sporadic performance on
psychometrically-measurable tasks is difficult to defend as generally intelligent.
One might ask: Can a system do well on psychometric tests and still be subject
to claims of tailorability? The answer, if the requirements in TCA 4 are not met, is
certainly Ye s . PAI is not meant to be a replacement for the input format and large
database requirements we have been developing in this paper; rather, it is only one
possible sharpening of the ambiguous concepts in TCA 3 . Other possibilities may
exist, but we do not at present know of any satisfying alternatives.
5.3 Case Mapper: A Step in the Right Direction?
TCA 3 and TCA 4 reflect a growing realization among the AI community that if some
cognitive system is to ever move closer to AGI, it needs to be able to demonstrate
its performance with a large knowledge base. Case Mapper , from Northwestern
University's Qualitative Reasoning Group, is a system currently in development that
allows users to test SME [ 17 ], MAC/FAC, and other analogical reasoning tools,
along with large databases that are included. Among these databases is a version of
OpenCyc [ 35 ], which comes with millions of facts connecting tens of thousands of
concepts and relations. Although in our opinion the database is not yet large enough
to match the performance of an 8-year-old child on analogy problems that require
recollection of arbitrary source cases, the way it gives the user access to such a
collection of powerful tools suggests it can become very useful to anyone interested
in the field.
Case Mapper works with cases , which are collections of facts that can be con-
structed automatically by the program given some concept in the database. After
mapping two cases, a set of candidate inferences can be drawn from the source to
the target case. We can explore these candidate inferences manually in order to have
some idea of the system's current ability. For example, having Case Mapper con-
struct cases for Dog and Cat and setting them as source and target cases respectively,
among the candidate inferences produced are: