How Models of Creativity and Analogy Need to Answer the Tailorability Concern - Computational Creativity Research: Towards Creative Machines - page 93

questions, one of them being: How can we find good analogies? The TC is
especially problematic because it forces analogy researchers to prove that their the-
oretical process is the answer to this question, and although it can be promising to
see that some particular approach produces good analogies in some limited domain,
no approach can constitute a completely satisfying answer to this question unless it
is versatile enough to perform well in many domains.
Any system that can answer the challenge of the TC will instantly distinguish
itself from every other extant analogical system, since (at least, to our knowledge)
the only one that has been able to do this with some degree of success is the SME-
based family of systems [ 20 , 24 , 33 ]. Later in this chapter we will briefly discuss
SME (particularly, the CaseMapper software) and point out several of its features
which we believe constitute partial answers to the TC. But first, it is important to
clarify what it means to answer this challenge and discuss why it is such a non-trivial
feat.
5.2.1 Answering the TC
Gentner and Forbus [ 24 ] suggest that there are two possible ways to answer the TC.
One applies to visual domains, and involves using automatic encodings of visual
representations. The other more generally applicable direction involves two key fea-
tures: first, the use of pre-existing databases; second, an automated or semi-automated
parsing process that goes from input text to a sufficiently rich semantic representa-
tion. A first attempt at a precise statement of what it means to answer the TC is as
follows:
TCA 1 A computational system of analogy answers the TC if, given no more than a pre-
existing database and an unparsed input text, it is able to consistently produce good analogies
across many domains.
At least one general intent behind TCA 1 is clear: it attempts to place emphasis
on the filtering process (whose job is, as we said, to either select some subset of
available source analogs from a large database and recommend only some of them for
the more computationally expensive step of analogical matching, or to automatically
construct structured representations from sensory data). By removing the reliance
on human intervention, TC ensures that the filtering is not done manually in such
a way that guarantees desired results. However, in order to truly answer the TC
in a satisfactory way, we must be precise about its purpose and motivations: What
concerns are behind the TC in the first place? Furthermore, TCA 1 is hopelessly
vague: the words 'unparsed' and 'good', if left open to interpretation, make it too
easy for anyone to prematurely claim victory over TC. Also: Why is it important that
the database be pre-existing? What degree of separation must there be between the
creators of the database and the designers of the analogical system? For example,
does the database's underlying knowledge-representation philosophy need to overlap
with that of the analogical system?