A Personal Perspective into the Future for Computational Creativity - Computational Creativity Research: Towards Creative Machines - page 387

The situation changes significantly if one starts attributing relative merit to the
construction process itself, beyond and above the intrinsic value of the artefact in
terms of practical use. This is very much what happens in the realm of intellectual and
artistic endeavours, such as for instance literature, music, painting, or art in general.
In fields of this nature, the value of artefacts becomes related to their ability to provide
stimuli to the mind of the perceiver that enrich their experience of the real world.
This is very much what happens in the case of the artist, and it plays a fundamental
role in setting the differences described between the process for the engineer and
that of the artist. From the conceptual point of view that we want to consider, this
peculiarity introduces a very dramatic perturbation into the schema that needs to be
considered for computational implementations of solutions to artefact generation in
these cases.
For a start, the creation process that we were considering can no longer be con-
sidered in isolation. The specification of the goal or the task of the artefacts to be
generated under this new approach is relative to a concept of mind that reacts to a
stimulus, intended as an abstraction of the audience that receives the artefacts pro-
duced and provides some feedback on whether they prove satisfactory. Although this
characterization is but one of many possible abstractions to describe the situation, it
captures the basic idea that the specification of the desired result is no longer phrased
in terms of specific properties of the artefact—as, in for instance, requiring that an
apple be red and shiny—but instead it is phrased in terms of possible reactions from
a very complex entity—a mind. Now a model capable of reproducing the reactions
of a mind is in itself a challenge even greater than that of artefact generation. This
premise underlies the main argument presented by Bringsjord and Ferruci [ 2 ]against
the possibility of ever achieving computational models of literary creativity. Accord-
ing to these authors, as machines will always be unable to model the extremely rich
range of reactions of a mind—in terms of imagery and evocation of sensual or emo-
tional experiences—to the simplest stimulus provided in textual form, creativity as
carried out by humans will always be beyond their reach. This is certainly an impor-
tant obstacle that cannot be dismissed outright. It is clear that modelling the type of
reaction being considered here would introduce significant complication into the set
of criteria as considered in processes 3 and 4 above for establishing lower and upper
bounds to the quality of artefacts desired. This complication may be challenging
enough for these processes to start meriting some kind of breakdown into simpler
processes. This is clearly an open problem at present, and one that has received little
attention from the computational creativity community. Some recent efforts point at
the interactions between computational creativity and cognitive computing [ 19 ]. The
need for more detailed research along these lines has recently been argued for the
case of story generators in [ 11 ].
A second issue is introduced by the fact that the perceived stimulus often arises
as a contrast between a newly produced artefact and previous instances of similar
artefacts. When repeatedly exposed to similar stimuli, the mind gets bored and reacts
less and less. This implies that a model is required to account for the moment in which
the audience is exposed to a particular result, as this determines which other results
have been seen before. This may involve a further selection process (process 8 of