A Personal Perspective into the Future for Computational Creativity - Computational Creativity Research: Towards Creative Machines - page 389

19.2.3 Current Efforts in Computational Creativity
with Respect to the Conceptual Schema
The conceptual schema presented above is intended as a compilation of the set of
decisions required for the instantion of a process of artefact generation, including
a refinement to account for cases where the suitability of the artefacts is based on
their ability to create an impact on a hypothetical human audience. This schema can
now serve as a yardstick for how far Computational Creativity has progressed along
the road to developing models of the type of process considered. It has already been
argued in Sect. 19.2.1 how existing poetry generators tend to focus on a very small
subset of the processes involved in the simple generative act. Process 8 introduced
in Sect. 19.2.2 is not better covered. Some effort at addressing this point was made
at the theoretical level in [ 8 ].
The process most commonly addressed in systems designed to generate artefacts
traditionally considered creative tend to focus on processes 5—application of a con-
structive procedure to a selected set of ingredients to obtain a set of candidates—and
6—application of selection criteria to the resulting candidates to establish a pre-
ferred result. The motto of the ICCC 2012, “Scoffing at mere generation for more
than a decade” highlighted the generally accepted need to go beyond construction to
consider evaluation as part of the remit of creative systems.
Very few systems can be said to include instantiations of process 1 of selection of
a constructive procedure. The poetry generator described in [ 4 ] includes a process
for defining what is called an aesthetic, which involves selecting between a number
of different ways of constructing the final poem, based on a set of pre-established
criteria. This could be considered as a basic approximation to the process of selecting
a computational procedure to apply. Systems based on evolutionary [ 9 ] or genetic
programming [ 13 ] do in some way include some means for the system to determine
dynamically how the artefact is constructed, in terms of combinations of very basic
evolutionary or genetic operators.
Process 2 of selection of the ingredients to be employed is also poorly represented
in existing systems. The MEXICA story generator [ 17 ] received as input a set of
previous stories, from which it built the internal resources it would use to generate
further stories. Some recent poetry generators [ 4 , 10 ] fetch texts from the Internet
that they use to extract the ingredients they will later use to construct new poems. The
PoeTryMe system [ 12 , 15 ] is designed as a framework that can consider different
linguistic resources as sources for its ingredients. These are instances of systems
designed to be able to consider various inputs, but the decision of which input set to
consider in each case is usually taken by the designer or programmer.
Processes 3 and 4 involve establishing selection criteria for identifying valid
candidates. A fundamental work on this subject is the paper by Graeme Ritchie
[ 18 ]. This work combined criteria for validity of candidates—referred is the paper
as typicallity—and for novelty. The criteria proposed by Ritchie cover together
processes 3 and 4 for establishing lower and upper bounds on quality and process 8
for determining the degree of novelty of the artefact. In his paper, Ritchie presented