When Music Meets Machine: Perceptual Analysis of AI Music

In the last few years there has been an evident surge in the use of AIs among the general population. The public release of image generators powered by AI (DALL E, MidJourney...) allowed the general public to produce impressive images starting from a simple text prompt. Social networks quickly flooded with AI images and, suddenly, a powerful spotlight was shining on AIs. In this newfound attention, highly proficient language models started to claim their place (ChatGPT, Elicit...), offering to aid us in research, writing, coding and so much more. In this rapidly evolving environment, there has been much talk about the value of AI-generated content. This paper focuses on an aspect often overshadowed by the pleasantness of AI art: recognition. In what can be assimilated to an artistic Turing Test, this paper aims to determine whether listeners can distinguish between AI-generated music and music composed by humans. Building upon recent experiments that have revealed a positive correlation between high musical expertise and enhanced perceptual abilities in music (Castro and Lima, 2014), this paper also investigates whether specific characteristics such as music sophisticatedness and familiarity with a particular music genre can impact the ability to discriminate between AI-generated music and human-composed music. In an effort to break away from the tradition of classical music as a staple of music psychology, a more holistical approach was used, considering multiple genres and instruments.