IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information - Computational Creativity Research: Towards Creative Machines - page 123

neurophysiological proposals as to where these functions might be operationalised
in the wetware, and to evolutionary arguments that motivate the thinking. We will
show how the simple architecture we propose addresses several questions which are
conventionally studied separately in the literature: it is important to remember that
the point is to draw these together under our unifying architectural umbrella.
The rest of the chapter is structured as follows. Section 7.2 introduces our method-
ology. Section 7.3 summarises the background to our central proposal; this is exten-
sive, so only a superficial survey can be given here, and we apologise to authors who
may feel that their work has not been given adequate coverage. Section 7.4 presents
themodel itself, beginning at an abstract level of detail andworking down. Section 7.5
explains the most interesting predictions of the theory, and, finally, Sect. 7.6 explains
how it is will model spontaneous creativity.
7.2 Methodology: Structured Abstract Computational Modelling
For the reader who is unfamiliar with the philosophy of computational modelling
research, it is worth briefly summarising our approach here; other readers may safely
skip this section.
Computational models provide a means of developing a theory about a complex
process that is not directly observable. The idea is to write a program that embodies
the theory, and then test that program across all its possible parameters, looking for
predictions that are testable, and the more unexpected the better [ 22 ]. These unan-
ticipated predictions are important, because, otherwise, one can fall into the trap of
building a model that works perfectly well, but which tells one nothing about the
phenomenon being modelled: instead, it merely confirms that the modelling tech-
nique used is capable of modelling the data supplied, without necessarily shedding
light on the process that produced it.
Wiggins [ 51 ] introduces the distinction between descriptive and explanatory mod-
els: the former is the kind of model that describes a phenomenon without attempting
to explain how it works (e.g., the Gestalt grouping principles) and the latter is the
kind that provides a mechanism by which the phenomenon actually works, at some
level of abstraction. It is usually the case that a descriptive model is developed before
corresponding explanatory ones, and it is often the case that an explanatory model
functions as a descriptive model at a less abstract level of modelling. For example,
the finite element analysis techniques used in modern weather modelling give an
effective model of the physics of the weather, but abstract away the detail of the
movements of individual molecules in the atmosphere. Similarly, one may often
usefully think of the function of a program without considering the operation of the
bits in the computer's memory, and one may consider the operation of the proces-
sor on the bits without considering the operation of the electronics in the chip that
implements it. This last example is similar to—but not the same as—Marr's levels
of description [ 26 ], which have been debated by McClamrock [ 29 ]. Wiggins [ 52 ]
expands on these ideas at greater length.