Argument
Recent developments in generative artificial intelligence and its massive arrival in the musical world have revived longstanding philosophical debates and produced new ones. While one can now generate a piece of music in a few seconds using tools such as Suno, the use of algorithms to generate music—whether paper-and-pencil algorithms or computer programs—has a longer history and runs through part of the musical tradition: from eighteenth-century musical dice games (Musikalisches Würfelspiel) to deep neural networks, via stochastic models (e.g. Xenakis), generative grammars, or Markov chains, which are still used in current systems for artificial improvisation (e.g. Somax2). All these examples may be understood as belonging to the category of artificial music, that is, music in which a non-trivial part of the creative process is delegated to a technical object—understood broadly as covering algorithms, mathematical models, and analog or digital machines—capable of functioning with some degree of autonomy and thereby escaping, at least in part, the control of a human creator.
The tradition of algorithmic music, which aims to design explicit and interpretable algorithms for generating music, has long been the primary provider of artificial music. But the rise of machine learning changes the game by making it possible to employ implicit but opaque algorithms—trained on datasets but whose details, by design, elude direct human understanding. It is this new tension that we aim to explore in this international conference. What changes does machine learning, and especially deep learning, bring to the ways in which musicians, listeners, and philosophers create, appreciate, and think about music?
Topics include, but are not limited to:
-
Definition and ontology
-
Does artificial music, whether produced with explicit or implicit algorithms, fit the definitions of music debated in the philosophical literature? Do deep-learning techniques raise specific difficulties—especially since these algorithms are not completely the result of explicit human design?
-
What kinds of objects do we find in artificial music? Are they similar to those identified in learned and popular traditions (compositions, tracks, improvisations), or are they entirely new objects? To which ontological category or categories do they belong? What are their criteria of identity and their conditions of persistence over time?
-
Value and appreciation
-
Are there specific reasons to value or devalue artificial music in virtue of its artificiality? Some empirical studies show that listeners judge music more harshly when they believe it was generated by AI rather than by humans: what philosophical implications can we draw from such findings?
-
Are there aesthetic norms associated with artificiality: perfection? glitch? the synthetic? kitsch, in the case of deep networks that excel at imitating existing genres?
-
Is there a form of expressivity proper to artificial music—for instance, the musical expression of things other than human mental states?
-
Authorship and creativity
-
Who are the authors of pieces of artificial music? Human authors? Nonhuman ones? Hybrid? Or should we dispense with the notion of authorship altogether?
-
How should we characterize the artefacts involved in artificial music? Are they mere tools, or collaborators? And what does it mean to “collaborate” with a neural network?
-
Can machine learning be understood as a compositional method? What possible role could data collection or model fine-tuning play in the compositional process?
-
Ethical and legal issues
-
Are current concepts and tools for protecting intellectual property adequate in light of the proliferation of online musical content and of artificial music trained on these materials? Should we abandon copyright altogether?
-
Are the musical objects produced by deep models to be assimilated to derivative works (covers or mash-ups), to stylistic plagiarism? Or do they establish a new type of relation to pre-existing musics? If so, what is the ethical status of this relation, and how could it be legally regulated?
-
How can we identify, describe, and counterbalance the biases resulting from dataset construction and model architectures?
Submission procedure
We invite the submission of extended abstracts (1000 words including references) prepared for blind review directly to the artmusphi.sciencesconf.org website, on the "New submission" tab by May 1st, 2026.
Acceptance notification may be expected by June 15th, 2026.
