I used to be quite optimistic about the rise of AI-generated music. I thought, “it combines my two biggest interests, CS and music, so this is the kind of stuff I live for.” I even wrote this sentiment in my application for the CS coterm: that the reason I wanted to study more AI was to apply it to the music-making process.
Sometimes the only way to know that you don’t like something is to experience it. I think AI-generated music was one of those things for me.
My interest in AI and music started when many AI tools were being developed that aided in the music-making process. There were plug-ins for mastering, websites for downloading stems, and even algorithms for chopping drum and bass beats. These early applications of AI made me optimistic because they very clearly aided my process and, in many ways, helped me learn how to make music better. Listening to what the mastering plug-in was doing to my track helped me realize the ways that my dynamics and EQ were making my tracks sound unprofessional. A lot of these tools also have many knobs and dials to turn and there’s a distinct and intuitive way that these hyperparameters affect the output of the various programs.
I even started making my own programs. I developed a very rudimentary harmony and melody designer that would essentially randomly choose harmonies and melodies that would most likely sound good. Then I got to play around with those and pick the ones I liked.
But as an artist, the most important tool that I have is my taste. That’s what sets me apart from other artists. And it’s my greatest defense for my music: “I like it.” These tools never got in the way of that, nor did they ever try to. They don’t pretend to have taste, allowing you the flexibility to tweak them to appeal to your taste.
MusicLM is a whole other story.
Words being fed into a black box algorithm that outputs music. Where is the taste in this process?
The music created by MusicLM is still distinguishable from human-made music, but it raises the question what is the difference between human-made and AI-made music? Even if they were indistinguishable, there's still the undeniable fact that humans making music is very different from AI making music. Humans use their taste, instrument-playing abilities, and emotions to curate a vibe through music. AI pretends to be a human who has taste and emotions and makes the most likely song. Listening to the musicLM songs honestly didn’t make me feel anything. It felt dystopian to be listening to a “chill, laidback singing voice” that’s just an approximation of what a chill, laidback singer would sound like.
How much of music is knowing what’s behind the music; knowing that the pain in someone’s voice came from a pain that someone felt; knowing that someone loved another person so much that they couldn’t help but express it through music?
The question that scares me is if we switched the Mona Lisa with a perfect recreation of it and no one knew, would it still have the same impact? If we listened to AI-generated music on the radio, in the club, and at parties but didn’t know, would it still hit the same? It’s scary to think that maybe it would. Or even scarier, what if the knowledge that you never know whether a song is human-made or AI-made actually makes listening to all music a worse experience?
I will never stop making music because the process of making music is why I do it. The creativity, the feel, the expression. But will anyone want to listen anymore? I don’t know.