MU_PSYC : Algorithmic music composition with a music-psychology enriched genetic algorithm

Recent advancement of artificial intelligence (AI) techniques have impacted the field of algorithmic music composition, and that has been evidenced by live concert performances wherein the audience reportedly often could not tell whether music was composed by machine or by human. Among the various AI techniques, genetic algorithms dominate the field due to their suitability for both creativity and optimization. Many attempts have been made to incorporate rules from traditional music theory to design and automate genetic algorithms. Another popular approach is to incorporate statistical or mathematical measures of fitness. However, these rules and measures are rarely tested for their validity. This thesis is aimed at addressing the above limitation and hence paving the way to advance the field towards composing human-quality music. The basic idea is to look beyond this constrained set of traditional music rules and statistical/mathematical methods towards a more concrete foundation. We look to a field at the intersection of musicology and psychology, referred to as music-psychology. To demonstrate our proposed approach, we implemented a genetic algorithm exclusively using rules found in music-psychology. An online survey was conducted testing the quality of our algorithm’s output compositions. Moreover, algorithm performance was analyzed by experimental study. The initial results are encouraging and warrant further research. The societal implications of our work and other research in the field are also discussed.