I have a very pragmatic view on music: all I care about is whether it sounds good or not. That’s it. That’s literally it. And from that standpoint the advent of AI music has been a major net positive for me.
You can choose to ignore that and come tell me about the ethics of using AI or how you care about the story behind the song and so on and that’s all fine but just note that you’re no longer addressing what I’m actually saying here.
I’m also not claiming it’s great - only that it’s good. I’ve yet to discover an AI song that I go back to like I do with many human-made songs but if I want to put on a playlist playing in the background then one consisting of entirely AI-made music reliably provides me well above average track after another whereas with organic music I need to sift through a ton of absolute crap to stumble upon the occasional good track that still rarely exceeds what AI is able to output. I can go on YouTube and put on a 2 hour playlist of Jazz House or similar and I don’t get the urge to skip a single track.
Currently I feel like where AI still trips is lyrics - they still are only plausible sounding nonsense once you actually start paying attention to it but especially for instrumental music it has far exceeded the level of a good average human artist.
Few examples of what I’m refering to: with vocals and without.
Upvoted for unpopularity. Objectively speaking, based on the foundation of how AI models work, their output tends to the mean (average). If the music in the example is considered “good”, then objectively the standard has moved downward a lot. I’ll start with visual AI as analogous to make my point:
Various people and businesses, big and small, have decided that AI generated images fit their needed purpose. I see a few places downtown that have made their store logos AI slop. Is it objectively well drawn image? In a sense that maybe they make fewer mistakes than before, but besides that, “average” is a more objective moniker for it than “good” on a factual basis. Subjectively you can decide whether you like it, I can judge you for it but the decision is yours to make.
Back to AI genned audio, even before it became widespread, people were making Muzak for elevators, phones, commercial jingles etc. It was fit for purpose to keep you occupied in the background. AI soundtracks seem to fit that niche for you and that’s fine. It may be “good enough” for you, but it’s not “good” and I doubt will ever be if judged objectively as music.
I listened to the instrumental track, and sure it is coherent at a very basic level, but the harmonies just keep going back to the same, any semblance of a musical idea or fluorish ends or is replaced with another entirely different one within 4 bars. That’s the reason why I don’t think any song is memorable, without any fundamental changes to the underlying tech I doubt any will. AI generated songs like “We are Charlie Kirk” are memes because they are monuments to conservative tastelessness.
People will tell you not to listen to or support that, I won’t stop you but I’d agree in the sense because I think of it as supporting an industry for elevator and phone hold muzak over real musical artists, that also sucks power and water and pollutes a lot. I personally don’t care about your tastes, you can like it, but calling it “objectively good music” is an affront to art.
I think there are two suitable uses of it: generating a cover or pastiche of two unrelated artists that would never have made something (Frank Sinatra sings “Rock the Casbah”), or completely filler music. Which if the second is meeting your needs, great, but that doesn’t make it objectively good. Perhaps good for Muzak consumers to have a flourishing industry but not for music consumers.
I would claim that much of AI-generated imagery is as well objectively good because much of it is such that we would describe it as good had it been made by a human. However, since AI often has quite a distinctive style, the output of it is usually very recognizable and it gives people the ick. Even for me - for example AI-generated thumbnails on YouTube videos are like a red flag for me despite me not being against the use of AI on principle. I just think there’s a time and place for it and in this case it feels misleading and dishonest.
When it comes to AI-generated music I (and I argue the vast majority of people) simply can’t tell the difference in most cases by the sound of it alone. The only difference between a solid track made by AI and an organic equivalent is the knowledge that one was made by AI and the other wasn’t. In many cases I could present a track made by a human but claim it was done by AI and people would absolutely hate it. That’s what my point here essentially is: they’re not hating the actual thing but the story they tell themselves about it. It does not suck simply because it was made by AI. One can choose to actively try and avoid it for ethical reasons or such and that’s fair but also a different thing. It’s like saying meat tastes bad when in reality they’re just morally opposed to how it’s produced and the taste plays little to no role in that.
I did study both computer neural networks and music so I can tell you pretty definitively that what AI has gotten good at is imitating human artistry.
The problem is that AI style is that it doesn’t have unique repeatable style, it’s kinda just averaging of a bunch of styles, so it will only ever be an imitation, in its current form.
So yeah, I’ve been “tricked” by AI before, thinking it is a real song, and probably you could trick me in the other direction too. However, just like a lawyer would be checking a hallucinated citation, I checked my facts. After I found one cover that happened to autoplay into slightly intriguing enough to want to hear more of the covering artist’s discography, but I wouldn’t be able to because this artist doesn’t exist. I did feel empty and disappointed after that. The variations to the original song I thought to myself tended towards the mean, well before I had any suspicion it could be AI.