“AI-Generated Music. Critical Analysis of Musical Validity,” the dissertation of Martin Lucas De Carvalho, a Piano and Keyboard Diplomate at Saint Louis.
A thesis questioning the future of music, this was the theme with which Martin Lucas De Carvalho addressed the final discussion of his Level I academic course in Piano and Pop- rock Keyboards., under lecturer Alessandro Gwis . The Saint Louis College of Music committee deemed this work worthy of publication for the quality of the analysis and the relevance of the topic addressed.
His thesis explores one of the most debated knots in contemporary musical aesthetics: can music generated by artificial intelligence be said to be artistically valid? A question that does not settle for easy answers, and which the student addresses with theoretical rigor, review of scientific literature and an original empirical experiment involving 112 listeners.
Music and artificial intelligence: a territory yet to be explored
Music generated by artificial intelligence is now at the center of global cultural and technological debate. Instruments such as Suno, Udio and other AI music generation systems can compose songs in seconds, imitating styles, genres and harmonic structures with increasing precision. Yet, the question that remains unanswered definitively is always the same: can a machine make art?
It is in this scenario that Martin Lucas de Carvalho’s work fits. His thesis addresses with scientific method and musical sensitivity the relationship between artificial intelligence and artistic value in music, contributing to a debate involving composers, researchers, producers and listeners worldwide.
The abstract of the dissertation
The artistic validity of music generated by artificial intelligence represents one of the most pressing aesthetic issues of our time. While music generation platforms such as Suno have achieved impressive technical sophistication, capable of producing tracks in seconds, a fundamental question remains open: can an algorithm generate music that is not merely technically competent, but authentically valid artistically?
The present research addresses this question through a path that integrates theoretical foundations, empirical literature review and direct experimental verification. The question is not whether AI music is “good” or “bad” in an absolute sense, but rather to understand what the relationship is between machine competence and perceived artistic validity for the human listener.
First, let us establish what we mean by artistic value in music: not a fixed value, but dependent on the meeting of formal coherence and perceptual effectiveness. Second, we examine what artificial intelligence has already learned to do well, and where it still fails. Recent literature suggests that AI has gained significant ground in replicating the formal codes of music. Fractures remain, however: micro-signs of artificiality, the subconscious recognition of an absence of conscious intentionality, and the crucial role of the listener’s cognitive bias as he or she discovers-or suspects-that he or she is listening to a machine.
The present research answers these questions not only through theory and literature, but through a direct empirical experiment: 112 listeners evaluated three songs generated with Suno, unaware of their origin, revealing a paradox that will be the focus of this research.
