One day, AI will tutor students in all manner of subjects. Including music.
Oh wait, that day is already here.
“What are we to do?” You might ask.
After posting previously in a forum thread, I’ve mulled it over for a few days and would love to hear some other reasoned thoughts.
Obviously the latest LLMs still answer many, many music-related questions incorrectly.
But the future seems inevitable. So, I see no choice but for some willing individuals to participate by actively developing a music tutoring AI product.
To be clear, I expect AI tutors in every field. There is no reason music will become an exception. By tutor I am envisioning a helpful assistant, not necessarily the primary instructor at formal colleges and universities. And I am not saying these tutors will be a formal contributor to the curriculum. But they will exist and some impressionable minds will pick them up and use them, whether those students go to school for music or not.
So, what are your thoughts?
If you could participate in a music education software development effort (I did for 8 years, and would again) what would you want to teach it?
I’ll give two examples from my own experience. I asked ChatGPT 4o about some chord notations in a course I was taking (from Film Music Notes) and it got the answer wrong two or three times before it got it right, but, eventually I got the answer I expected.
In the above example, I was already taking a paid course, and I had a question, asked, and got an answer that helped me. Eventually. I knew enough to know it was wrong the first couple times. Not all students would.
The other example (mentioned in my linked post above) was when I was exploring a totally new genre / style / musical “subculture” which still feels quite foreign to me. I prompted SUNO to create some pieces, a couple of which were relatively impressive. Then I prompted about seven other AI music creation tools the same way and got absolutely incoherent garbage. The fact that SUNO nearly read my mind was initially enthralling (and I mean that in a vampiric sense — I felt “in the thrall” at first) partly because it was unexpectedly capable.
So, I transcribed two of my favorite pieces, and started dissecting strengths and weak points in the composition.
They were both very fast with lots of notes and the harmony was slightly jazzy. They were both unlike my writing. They weren’t part of my muscle memory in terms of melodic choices and I played them very slowly to try figure out why certain “choices” were made.
The result?
I feel personally compelled to teach AI where I can, because I want it be better. Even when it’s impressive, it makes some garbage (nonsensical) outputs, and it’s this bleeding edge that will require skilled, musically experienced minds to feed it musically coherent reinforcement instruction.
Note: I have not, at the time of this posting, released the musical outputs described above publicly, and if I do I will label the source and explain that it was an experiment — not part of my regular practice, either casually or professionally. I did post them privately to YouTube to see if ContentID could identify source material, but it did not.
Personally, I found the SUNO pieces “eerily” capable of writing what I was prompting but ultimately flawed. Yet I also found them “inspiring” in the sense that they seemed to capture and unlock for me a style/genre I had tried occasionally to understand for nearly two years.
These experiences led me to seriously consider the impact AI will have in music education, and it’s my belief that someone is going to need to guide the companies developing music education products. Well, put it this way: someone will guide those companies.
How should they guide those companies (?) is my main question.