Stakeholder Groups in Computational Creativity Research and Practice - Computational Creativity Research: Towards Creative Machines - page 22

Fig. 1.4 The progression of an evolutionary art program through eight system epochs, taken from
[ 47 ]

produce offspring sets. He/she also has an idea ( E p )fora wrapper routine which
can use such a set of functions to produce images. He/she then uses the program to
generate ( C g ) a set of functions and employ the wrapper to produce ( E g ) an image
which is sent to the (P)rinter. The crossover and subsequent image generation is
r epe ated multiple times in system 2, and then the programmer—who has invented
( A g ) their own aesthetic—chooses a single image to print. In system 3, as in the
poetry example above, the programmer translates their aesthetic into code so the
program can select images. This is a development similar to that for the NEvAr
system [ 51 ].
Following the second line of development, in system 4, the programmer selects
multiple images using his/her own aesthetic preferences, and these become the posi-
tives for a machine learning exercise as in [ 52 ]. This enables the automat ic invention
( A g ) of an aesthetic function, which the programmer translates by hand T from

the machine learning system into the software, as in [ 53 ], so the program can employ
the aesthetic without user intervention. In syst em 5, more automation is added, with
the programmer implementing their ide a ( C m ) of getting the software to search for
wrappers, then implementing this ( E m )

( a g )

, so that the software can invent ( E p )new
example generation processes for the system.
Following the final line of development, in sys tem 6, we return to aesthetic
generation. Here the programmer has the idea ( A p ) of getting software to math-
ematically invent fitness functions, as we did in [ 54 ] for scene generation, using
the HR system [ 4 8 ] t ogether with The Painting Fool [ 43 ]. In system 7, the pro-
grammer re alis es ( C m ) that crossover is just one way to combine sets of functions,
and gives ( E m ) the software the ability to search a space of combination methods
( C p ). The software does this, and uses the existing wrapper to turn the functions
into images. System 8 is the end of the line for the development of the software,