On the Role of Computers in Creativity-Support Systems - Computational Creativity Research: Towards Creative Machines - page 118

domains, it may be possible to make certain assumptions about the audience, and so
IDEA model can be useful in a limited way.
6.4 Conclusions
To summarize the main arguments of this chapter, we would like to rearticulate
them in another way. It is generally accepted that the two main characteristics of
creativity are originality and intelligibility: the product must be novel or the process
must generate a new perspective; and the product or the generated perspective must
be intelligible in order to be useful for at least some audience [ 2 , 56 , 67 ]. For
novelty, research on real-world creativity shows that it is difficult for people to step
out of their conventional and habitual conceptual associations. To overcome this
inertia, several methods like making the familiar strange [ 18 ], concept displacement
[ 58 ], bisociation [ 36 ], lateral thinking [ 14 ], estrangement [ 54 ], conceptual blending
[ 15 ], and so on, have been proposed in the literature. However, computers do not
have this inertia, and so they can be very effectively used to generate novel ideas.
This argument has been presented in more detail elsewhere [ 31 ]. Our experience in
developing creativity-assistive systems (reviewed in Sect. 6.2 ) lends supports to this
hypothesis.
However, when it comes to incorporating usefulness of the generated perspective
or idea, we have argued that, in general, it is not possible to capture this aspect of
creativity algorithmically. The reason is simply that when a new object or style is
introduced, people react to it in different ways. Sometimes they adapt to it right away;
at other times they do not find it interesting or useful at first, but the same object
or style introduced at a later time becomes a big success; and sometimes they do
not find it useful at all, in spite of the efforts made by the creators to convince them
otherwise.
Nevertheless, one cannot rule out the possibility that in limited domains we might
be able to characterize usefulness algorithmically, and to design and implement com-
puter systems that can generate statistically a larger number of useful and interesting
artifacts and ideas. So combining this with novelty-generating systems, we can have
computer systems that are creative. Systems like Aaron exemplify this approach.
However, even in a limited domain, once usefulness is characterized algorithmi-
cally, it loses its novelty, and gradually ceases to be creative. (See, for instance, the
model of literary style change proposed by Martindale [ 42 ].) So while, we may be
able to model some aspect of creativity within a style (with respect to usefulness),
it remains doubtful whether creative changes in styles can be modeled successfully
in a universal way. Again, to emphasize, novelty can be modeled—it is relatively
easy to computationally generate new styles, but the problem is to incorporate which
styles will be successful (meaning people will adapt to them and find them useful).
Therefore, we claim that this usefulness aspect of creativity will always remain the
last frontier for computational modeling techniques.