Karl Sims used the genetic programming method in combination with selection based on the aesthetic criteria of the user to evolve abstract images [34,35]. This represents a step toward loosening the definition of the ``fitness function''. In traditional GAs and GPs, the fitness function is hard-coded into the computational system. In Sims's genetic images, the fitness function is provided by the aesthetically based selections of the human user in each generation of images. These selection criteria are whimsical and change in each generation as the genetic operators generate new arrays of choices.
Figure 1: Karl Sims' system for evolving images by aesthetic selection.
An array of twenty choices. These are the products of mutations
and or recombinations of the previous aesthetic selection(s). The user
can choose one or two of these to be the parent(s) of the next generation.
An example of images evolved through aesthetic
selection.
An example of images evolved through aesthetic
selection.