The human touch and the algorithmic mind: A study on the perception of AI-generated music

Details

Supervisors

Verardi, Vincenzo

Faculty

Louvain School of Management

Degree label

Master [120] en sciences de gestion, à finalité spécialisée

Abstract

enThis thesis investigates how listeners from a specific community perceive music when told it was created either by a human or by artificial intelligence (AI). As AI tools like SUNO become more advanced in generating music that imitates human creativity, it is crucial to understand whether listeners react differently based on the label attached to the music. The main question examined is whether a framing effect influences preferences and perceptions depending on whether the music is labelled as human-composed or AI-generated. The study involved 93 participants who listened to two musical excerpts, one composed by a human and one by AI. These excerpts were presented across three conditions: blind (no label), true disclosure (accurate label), and false disclosure (misleading label). Participants selected which excerpt they preferred and rated each piece on four dimensions: authenticity, creativity, emotional connection, and audio quality, using a 7-point Likert scale. The experiment followed an A/B testing logic. Differences between conditions were assessed using paired t-tests and confirmed through Wilcoxon signed-rank tests, to ensure reliability despite possible non-normal data distributions. Findings indicate a consistent and statistically significant framing effect. Music labelled as human was rated more positively across all perceptual dimensions, even when the actual composition source was AI. This effect was particularly strong for authenticity, creativity, and emotional connection. Audio quality showed a smaller but still significant difference. Moreover, the same musical piece received lower ratings when it was presented as AI-generated compared to when it was labelled as human-composed, regardless its true origin. These results support the idea that listeners’ perception from this community is influenced by framing and labelling, regardless of the actual origin of the music. The study contributes to ongoing discussions about trust, creativity, and value in AI-generated content. It also opens avenues for further research into public acceptance of generative AI in the arts, the role of transparency, and how attribution affects appreciation across creative domains.