... I don’t think the product matters all that much in creativity, especially not at the student level. Creativity is a practice, a process, an experience. Prompt-based generation skips the entire process. That’s why it’s so poisonous for education....
What Ethan has written is longer and more complex than my headline's paraphrase. I urge you to go read the whole thing. My take-away, such as it is, is that Ethan's concern is that when people draw upon artificial intelligence prompts they are using a shortcut to generate a product when musical activity is about the process and the physical and social and cognitive disciplines these processes encourage and refine.
What kind of musical processes? For me this question can be answered easily, counterpoint. Do you want to create a melodic line that can be set in canon against its own retrograde. Bach's crab canon from Musical Offering gives an example that you can have a melody that can be set against reversed version. The beauty and benefit of the music is not just in the music but in the processes and disciplines of mind and body to even be able to create such a little piece of music.
These are the kinds of things that inspire composers across the centuries. It doesn't even have to actually be counterpoint. What if you wanted to write a piece of music where, once you played through the end of the piece, you wanted to flip the score upside down and play it note for note backwards back to the first note you started from? Could you write such a piece and get it to sound okay played right-side up and forward and then upside-down and backwards? Paul Hindemith gave himself this challenge for the prelude and postlude of Ludus Tonalis. I happen to actually like a lot of music by Paul Hindemith so I love that composers try to do these kinds of things. I'm not telling you that you have to be a fan of either Johann Sebastian Bach or Paul Hindemith (if you are, hey, great!). I'm showing you through their music that the kinds of things they delighted in doing as musicians are the kinds of things that AI, even if it "could" do this for you, has robbed you of the process of learning to do yourself if you resort to using AI rather than spending your days or weekends messing around with what you can and can't do on your instrument.
What if I wanted to write a melody that is a 12-bar blues that can be set in canon against itself at the octave? It can be done if you understand how blues riffs work and how oblique motion lets you finesse the IV chord, let alone how John Lee Hooker and other blues masters used second inversion subdominant harmonies because open chord tunings don't give you any other option. Maybe you could tell AI to write a blues canon that can be played on the guitar in open G tuning but what if you wanted to switch to open D tuning? You know that the two-part blues canon that could work easily in open G tuning becomes completely unplayable in open D tuning for reasons no human has to spell out that AI could "probably" compensate for ... but why deprive yourself of your own paths of musical exploration by asking AI to do this stuff for you?
If I were to write a fugue with triple counterpoint that can be played using bottleneck technique (and that's not really an "if") I'd need to know enough about what's possible and not possible on the guitar to pull off such a feat. If I know that the most difference I can have using a glass slide on my left fourth finger is a single fret then I know that I can pull off a B flat at the eleventh fret of the second string and a D natural on the fourth string but that G on the third string? That's completely out of consideration. But the G on the fifth string that's normally A? If I'm really careful with my left hand position and my control of the glass slide that G at the 12th fret could be done.
Years ago I recall reading that the University of Washington made use of a video game called Foldit to do research on enzymes for AIDs research.
...
Proteins, of which there are more than 100,000 different kinds in the human body, form every cell, make up the immune system and set the speed of chemical reactions. We know many proteins' genetic sequence, but don't know how they fold up into complex shapes whose nooks and crannies play crucial biological roles.
Computer simulators calculate all possible protein shapes, but this is a mathematical problem so huge that all the computers in the world would take centuries to solve it. In 2005, Baker developed a project named Rosetta@home that taps into volunteers' computer time all around the world. But even 200,000 volunteers aren't enough.
"There are too many possibilities for the computer to go through every possible one," Baker said. "An approach like Rosetta@home does well on small proteins, but as the protein gets bigger and bigger it gets harder and harder, and the computers often fail.
"People, using their intuition, might be able to home in on the right answer much more quickly."
...
...
Foldit was created in 2008 as a way to 'gamify' protein research. Proteins are essential biomolecules found inside every cell of every organism. Their intricate three-dimensional structures give rise to their diverse functions, which include digestion, wound healing, autoimmunity and much more.
Through gameplay, Foldit players have helped determine the structure of an HIV-related protein and improved the activity of useful enzymes. Until now, however, Foldit players could interact only with proteins that already existed. There was no way to design new ones....
Where mass number-crunching would take time, engineers and scientists resorted to making a literal game and invited people to play that game. Human intuition about spatial reasoning ended up solving a problem in a few weeks that mass number-crunching would take a long time to solve.
There had to be people who knew the strengths and limits of computer-based problem solving to literally game up a compensatory process for the weaknesses in both computer and human resources to a scientific problem.
Now, sure, computing could be so good in 2025 that this kind of Foldit game might not be necessary any more. I don't actually know. To everything there is a season and a time and a purpose for everything under heaven. I've read Ecclesiastes enough times to guess that there is some kind of use for artificial intelligence to help make the lives of people better. Farming out the disciplines that let people make music could even potentially be included.
But the idea that you tell AI to generate images associated with a piece of programmatic music is just stupid to me. That would be like me telling one of my nieces who can draw just fine, thank you, that she should put in a prompt to AI to create images associated with some song. Why not just ask her to draw those things?
What I want music education to do is not to short-circuit cognitive development by handing off work to computers. I want music education to elucidate and clarify how the cognitive processes of, say, developing a blues-based or ragtime-based approach to sonata forms can be formulated. I have very specific, personal goals in mind. What artificial intelligence obviously can do is to act within the parameters of a genre or a form. That's the problem. The challenges and possibilities in music that I see for us in the 21st century are exploring the literal and figurative spaces across and between genres and forms.
I've spent a lifetime thinking about how to create blues-based sonata forms and my hunch has been that the path toward that is blurring the boundary between monothematic sonatas and continuous variation form. As best I understand how music educators have been discussing possible uses for AI it would seem that the prescription is holding to the foundational surmises about genre purity that I think we most need to reject. What is it about human-produced vibrations in the air that makes it "classical" or "blues" or "jazz"? Why should I take as given that a sonata form has to have more than one theme? Is it because of some post-Hegelian dialectical concept of masculine and feminine that need to be "reconciled"? Why?
By settling on monothematic sonata I found that I solved a first-step problem n how to make a blues sonata, and this entailed rejecting the very idea that thematic contrast is necessary or good in sonata forms. I had to swiftly and forcefully reject one of the biggest textbook bromides in the music theory of the long 19th century about what a sonata form even is. But that kind of baseline assumption is what probably has to be built into artificial intelligence for it to be used. That, I think, is the problem--artificial intelligence gets built from the ground up with the kinds of biases and presumed definitions about music that music educators should want us to avoid like the plague, to say nothing of musicians more generally. Giving students assignments where they have to use artificial intelligence comes off like it's force-feeding them a task of relying on a computer tool to presume the kinds of biased cognitive shortcuts in music that, if I understand Ethan Hein's argument correctly, music educators should never be foisting on students of any age to begin with.
As Ethan put it:
I support the impulse to make musical creativity more accessible, but the thing is, it is already extremely accessible. It’s the easiest thing in the world from a technical perspective. Little kids make up songs constantly. The question is not, how do we help people be creative? They are already, from birth. The question is, why do we grind creativity out of kids so thoroughly, and how do we stop doing that? Teaching songwriting and other creative music-making requires only that you disinhibit the strong creative impulse that is already there.
Every baby comes into the world capable of making ear-splittingly loud sounds with their little voice boxes. Professional vocalists and vocal instructors get paid money to teach people how to recover those sound-making abilities most babies come into the world able to do without thinking about it.
1 comment:
I might be more open to the possibilities of AI music generation if it wasn't being promoted so aggressively by the corniest people in the world. I got interested in music technology in the first place because I loved rock, hip-hop and dance music and wanted to know about the tools people were using to make those sounds. I find AI interesting from a STEM standpoint but I have never heard anything made with it that I wanted to hear a second time.
Post a Comment