Regulation with enforcement will be difficult, especially since machine learning is more of a tool than an author. An artist should have the right to remove their work from the training set as part of copyright law. Derivative works are very similar to sampling, and it would be very interesting to see what a good algorithm could put together; if not good work, at least something that may inspire new work by a human composer. A computer cannot be inspired, just programmed.
Agree with all of your observations, Jeff. Concerning regulation and enforcement, what do you think of an idea which would put control of provenance into the hands of the artist, placing it inside the DAW? Different from current attribution mechanisms which label already finished works, this DAW-integrated approach could attach provenance information to the building blocks of music itself.
That is an excellent idea. Within the DAW, perhaps a manifest could be made that could include the artist's criteria as well as what tools, plugins, etc., were included in the work. I know an archivist who is troubled by the inability to reproduce digital recordings, due to the complexity and continuous changes to authoring environments.
James. these are both really interesting questions, and make good topics for full posts in the future- thank you! But for now, here are my initial thoughts, and I would be really interested to hear more of what you may be thinking.
Classical Music: I think one important detail that would need to be considered would focus on the performance of AI-generated classical music. Specifically, would an AI composition be rendered by the GenAI itself, or would it be transcribed and performed by humans? The first scenario is an easier answer for me, as I feel the lack of humanity in the composition that I believe many may perceive would be amplified by a synthetic performance, and therefore not be appreciated (likely flat-out rejected by true classical music fans). The second scenario is more complex, as so much meaning is imparted by the performers- would this be enough to overcome the lack of meaning in the composition and connect with the listener? I need to consider this more deeply.
Concerning instrumental non-orchestral music, I was going to write a separate response to this, but as I think about it, much, if not all, of what I said about classical music also applies here (even though a non-orchestral performance may be more realistically rendered by AI).
Thanks for very provocative questions, and I think I will need to consider more invested responses to each!
I was thinking more along the lines of generating a MIDI score, with the assumption that much of the meaning is somehow contained in the relationships between the melody, harmony, structure, and so on. But you're absolutely right: dynamics, timing, and the sound from the instruments are essential. If we heard Beethoven's 5th played at pianissimo, half tempo, and with an ensemble of flutes, it might not have been so iconic! This leads me to wonder whether a meaningful piece just "has" meaning, or if it's constructed in such a way that we can easily "apply" meaning to it.
I think considering any music with lyrics is probably 100x more complex because of all the interactions between words, connotations, memories of depicted situations, etc. So many more extra-musical components and so much room for individual interpretation. But you might also say that it's far less abstract, so from the words alone it's easier to get a sense of the meaning of the song.
Regulation with enforcement will be difficult, especially since machine learning is more of a tool than an author. An artist should have the right to remove their work from the training set as part of copyright law. Derivative works are very similar to sampling, and it would be very interesting to see what a good algorithm could put together; if not good work, at least something that may inspire new work by a human composer. A computer cannot be inspired, just programmed.
Agree with all of your observations, Jeff. Concerning regulation and enforcement, what do you think of an idea which would put control of provenance into the hands of the artist, placing it inside the DAW? Different from current attribution mechanisms which label already finished works, this DAW-integrated approach could attach provenance information to the building blocks of music itself.
That is an excellent idea. Within the DAW, perhaps a manifest could be made that could include the artist's criteria as well as what tools, plugins, etc., were included in the work. I know an archivist who is troubled by the inability to reproduce digital recordings, due to the complexity and continuous changes to authoring environments.
How do you think this might apply to Classical music? Especially music without lyrics? Do you think that might be easier for AI to accomplish?
James. these are both really interesting questions, and make good topics for full posts in the future- thank you! But for now, here are my initial thoughts, and I would be really interested to hear more of what you may be thinking.
Classical Music: I think one important detail that would need to be considered would focus on the performance of AI-generated classical music. Specifically, would an AI composition be rendered by the GenAI itself, or would it be transcribed and performed by humans? The first scenario is an easier answer for me, as I feel the lack of humanity in the composition that I believe many may perceive would be amplified by a synthetic performance, and therefore not be appreciated (likely flat-out rejected by true classical music fans). The second scenario is more complex, as so much meaning is imparted by the performers- would this be enough to overcome the lack of meaning in the composition and connect with the listener? I need to consider this more deeply.
Concerning instrumental non-orchestral music, I was going to write a separate response to this, but as I think about it, much, if not all, of what I said about classical music also applies here (even though a non-orchestral performance may be more realistically rendered by AI).
Thanks for very provocative questions, and I think I will need to consider more invested responses to each!
Hi Paul,
I was thinking more along the lines of generating a MIDI score, with the assumption that much of the meaning is somehow contained in the relationships between the melody, harmony, structure, and so on. But you're absolutely right: dynamics, timing, and the sound from the instruments are essential. If we heard Beethoven's 5th played at pianissimo, half tempo, and with an ensemble of flutes, it might not have been so iconic! This leads me to wonder whether a meaningful piece just "has" meaning, or if it's constructed in such a way that we can easily "apply" meaning to it.
I think considering any music with lyrics is probably 100x more complex because of all the interactions between words, connotations, memories of depicted situations, etc. So many more extra-musical components and so much room for individual interpretation. But you might also say that it's far less abstract, so from the words alone it's easier to get a sense of the meaning of the song.
This stuff makes my head spin.
Looking forward to hearing your thoughts!