Information Technology Reference
In-Depth Information
simulate. For instance, the ability to produce smooth colour gradients through paint
strokes is something that would certainly enhance the quality of the pieces produced
by the software, and other physical simulations such as the use of a palette knife,
the ability to spatter paint, etc., would all add value.
The software is more lacking in appreciative and imaginative behaviours than in
skillful behaviours. We have argued that with the emotional modelling projects, the
software is exhibiting some level of appreciation of its subject material and its paint-
ing styles. The fact that The Painting Fool cannot assess its own artworks and those
of others against various aesthetic considerations is a major gap in its abilities. We
have implemented abilities for the software to calculate objective measures of ab-
stract art, for instance the location of symmetries, the distribution of colours, regions
of high and low texture, etc. However, it is difficult to imagine training the software
to appreciate works of art without essentially training it in a singularly subjective
way (i.e. to become a “mini-me” for someone). In such circumstances, it would be
difficult to argue against the software simply being an extension of the programmer,
which we clearly want to avoid. An alternative approach is to build on the project to
use mathematical theory formation to invent fitness functions, as described above.
Rather than inventing a single fitness function, we hope to show that it is possible
for software not only to invent more wide ranging aesthetic considerations, but also
adhere to them, change them and discuss and possibly defend them within cultural
contexts.
One aspect of this may involve getting feedback from online audiences, which
will be used to tailor the image construction processes. However, as mentioned in
Sect.
1.2
, we are keen to avoid creativity by committee, which could lead to the
software producing very bland pieces that do not offend anyone. Instead, we propose
to use a committee splitting process, by which The Painting Fool will judge the
impact that its pieces have on people, and choose to develop further only those
techniques that led to pictures which split opinion, i.e. those which people really
liked or really hated. Enabling the software to work at an aesthetic level will also
involve endowing it with art-historical and cultural knowledge to allow it to place
its work in context and to learn from the work of others. We are in discussions with
artists and art educators about the best way to do this. In addition, we will draw on
texts about creativity in both human and computer art, such as Boden (
2010
).
With the scene generation and collage generation abilities, we claim that the soft-
ware is very slightly imaginative, and we aim to build on these foundations. Firstly,
once the teaching interface is finished, we will deploy it by asking numerous people
from academia, the arts, and from the creative industries to train the software. The
payoff for people using the tool will be the production of pictures which hopefully
characterise the ideas they had in mind and would paint if they were using more
traditional methods. The payoff for The Painting Fool project will be a collection of
potentially hundreds of scene descriptions. We plan to use these in order to perform
meta-level searches for novel scenes in a more playful way than is currently pos-
sible. An important aspect of the teaching interface is the tagging of information,
which is passed from screen to screen in order to cross-reference material for use in
the overall scene construction. Hence, the software will in essence be taught about
