A Personal Perspective into the Future for Computational Creativity - Computational Creativity Research: Towards Creative Machines - page 388

publication), so that only one of the top performers produced in a given run is chosen
to be “published” or made public. Some kind of record of published artefacts—or
their relevant characteristics—should be kept, so that it can be taken into account
when deciding what are valid candidates for publication. This involves additional
processes to be considered during artefact generation. As a result of this additional
constraint, perfectly acceptable candidates in terms of the criteria resulting from
selection processes 3 and 4 may be rejected, not on the grounds of their intrinsic
quality but on the grounds of not being different enough from previously produced
artefacts. In addition, the degree of similarity with previous results may also be
applied during reflection (process 7) to trigger modifications of the constructive
procedure (process 1) or the set of ingredients (process 2). Acceptable candidates
produced but not “published” may be mined for valuable features not already present
in previously published results, with a view to guiding the modification process. The
consequences of this change are significant. Even when a single producer of artefacts
is considered it implies a transition from a stateless process to one that needs to
keep track of what has been produced before. But if the system is to operate in a
wider context, artefacts produced by other creators would need to be considered as
well. This corresponds to the difference identified by Boden between p-creativity
and h-creativity [ 1 ]. Whereas artefact generation as considered previously could
be optimised for a particular target, and then satisfactory instances could be mass
produced, the new situation requires that the set of previously produced results be
taken into consideration. The production of one instance of an optimal solution in
principle rules out all other solutions built along similar lines. This shifts the focus
of the overall procedure much more onto the reflection process.
From a computational point of view, the need to consider the state of a given
generation procedure introduces the need for an additional process that filters results
based on their similarity with previously seen results. The similarity used here may be
established at various levels, from the physical to the conceptual, including similarity
in terms of procedures or ingredients employed, emotions induced, or effects pro-
duced. The already argued complexity of modelling the reactions of a human mind is
therefore implicit within this process, making it a significant challenge for the field.
By analogy with the other ingredients of the conceptual schema, we should consider
two separate process that a creator may need to contemplate: one for identifying the
criteria for this filtering operation—we can identify this as process 8 of selection
novelty-based filtering criteria specifically—and one for applying these criteria to
the set of results—which may in fact be considered an extension of process 6 of
selection, where, therefore, the various selection criteria resulting from processes 3,
4 and 8 are applied together to the results of the production process 5.
It is important to note that reflection (process 7) may also need to be extended to
refine process 8, in view of the accumulated results of all the other processes.