Evolving Digital Art Using Grammatical Evolution

ABSTRACT The evaluation of aesthetic appreciation in evolutionary algorithm-generated digital artwork poses significant challenges. Prevailing computational metrics frequently exhibit misalignment with human aesthetic preferences, constraining the practical implementation of Agent-Based Design (ABD) systems in creative industries. This study aims to investigate the key attributes driving human preference in evolutionary art, and the efficacy of fitness functions based on these attributes, and then evalute the generated patterns. Therefore, a people-based preferences online survey was conducted to identify critical aesthetic attributes. After analyzing the survey results, it was concluded that Complexity and color Harmony are the most important elements in determining beauty in artworks. A NetLogo programming language used to devise a fitness funtction to measure these attributes. However, the metrics did not correlate well with people's preferences, indicating that further work is needed in this area. The devised fitness function has been used to demonstrate as proof of concept how digital art can be evolved using Grammatical Evolution (GE), a novel type of evolutionary algorithm uses a simple context-free grammar. Two types of generated patterns were observed during the end and middle stages of the evolutionary runs. One is after the end of the evolutionary experiments and the other during the middle stage of the evolutionary runs. Therefore, another survey was conducted to evaluate the two types of the generated patterns. The results showed that people preferred the patterns generated during the middle stage of the evolutionary runs, suggesting that more work is needed to improve the fitness function. This indicates a need for further development Keywords: Digital artwork, Evolutionary Algorithm (EA), Fitness function, NetLogo, grammatical evolution (GE), Agent Based Design (ABD)