Co-creativity between music producers and ‘smart’ versus ‘naive’ generative systems in a melody composition Task

Research in computational co-creativity frequently focuses on technical performance of computational systems and subjective quality of end-products. How computational co-creative systems impact the creative process of users has received less attention. This paper reports on a two-way double-blind crossover study to investigate how the creative processes of thirteen electronic music producers were impacted while interacting with two computational co-creative systems providing melody suggestions. The two systems shared a common user interface, however, the ‘smart’ co-creative system suggested melodies that expanded on producers’ input melodies, while the ‘naive’ co-creative system produced melodies unrelated to the producers’ inputs. To capture participants’ subjective experience, aspects of creativity were rated on Likert scales and further explored with semi-structured interviews. Each system’s output and producer’s intermediate melodies were compared for change (in compositions), dissimilarity (new AI generated elements), and adoption (into melodies). Producers considered the ‘smart’ co-creative system to produce the most novel and valuable contributions to their process. Outputs from the ‘naive’ co-creative system were judged to be more dissimilar than smart expansions. Nonetheless, the changes that producers incorporated into their intermediate melodies were similar between the systems. This study suggests that co-creative interactions can stimulate the creative process by offering both related and unrelated musical suggestions.