Tonight (May 13) at HacDC’s Natural Language Processing and Artificial Intelligence Group (NARG), Todd Fine gave a talk about a 1991 paper by Karl Sims called “Artificial Evolution for Computer Graphics.” It elaborated how genetic algorithms using the human aesthetic as a fitness function could generate beautiful, pleasant 2D textures and animations. Sims’ program created LISP S-expression trees which contained simple mathematical operations and graphical functions along with randomized scalar and vector constants. After the intervening selection of the most pleasing images, the expression trees would evolve in simple ways and create new generations.
Todd implemented a basic version of this approach using Pyevolve and the Python Imaging Library. He showcased some of his results (slides are available here), and the group discussed future elaboration that could involve cloud computing, Amazon’s Mechanical Turk, and web applications. NARG meets on every Thursday, and we encourage interested parties to join us.