Information Technology Reference
In-Depth Information
•
It fosters cooperation among different working groups, allowing, for instance, the
collaboration between research groups working on the development of AJS and
groups that focus on the development of image generation systems;
•
The same AJS may be incorporated with different systems allowing it to be used
for various creativity supporting tasks.
This chapter focuses on AJS validation. The next section discusses some of the is-
sues related to AJS validation and presents several validation methods based on psy-
chological tests, users' evaluations, and stylistic principles. Section
11.3
describes
the evolution of an AJS through time, from a heuristic based system to a learning
AJS. The results obtained in the system validation by means of the approaches pro-
posed in Sect.
11.2
are presented and analysed. Finally, we draw overall conclusions
and indicate future work.
11.2 Validation Approaches for AJS
Performance comparison of two AJSs is a delicate task. The existence of a vali-
dation task to which both can be applied is a prerequisite for comparison. Unless
the systems are applicable to the exact same task (which includes using the same
datasets) the comparison may lead to erroneous conclusions. The validation method
must be reproducible and the results should be numerically quantifiable. All compo-
nents of the validation task (e.g. datasets) should be made accessible to the research
community. Furthermore, it is also recommended that the datasets come from an
external source (i.e. that they are not specifically made for a given AJS) and have an
unbiased character. There are tasks, e.g. author identification, that despite not being
directly related to the ability to make aesthetic assessments, can be useful due to
their objectivity and can, potentially, complement other validation methods.
The characteristics of human aesthetic preferences—e.g. subjectivity, individual-
ity, cultural biases, change through time, etc.—create an additional difficulty. Simi-
larly, the interpretation of the results is also problematic and, in many circumstances,
it is difficult to determine what constitutes a good result.
In this section we will explore three different ways to validate an AJS: based
on psychological tests related to aesthetics, based on user evaluation, and based on
stylistic classification.
11.2.1 Psychological Tests
There are several psychological tests aimed at measuring and identifying aesthetic
preferences (Burt
1933
) and aesthetic judgement (Savarese and Miller
1979
, Furn-
ham and Walker
2001
). Some of them are employed on professional guidance, to-
gether with other psychological tests, in order to advise students about potential
careers.
