The interest in ai music generation is lower than I initially thought. I jumped in but felt the exercise lacked the joy of making music physically or with software like pro tools. With pro tools you control the thousands of knobs which gives you more control. These AI models take away that connection. You can play around with different words to get different results but it's like painting with a shotgun.
No one wants to hear other people's ai songs because they lack meaning and novelty.
AI image and short video generation can create novelty and interest. But when the medium require more from the person like reading a book or watching a movie the level of AI acceptance goes down. We'll accept an AI generated email or ad copy but not an ai generated playlist and certainly not a deepfake of someone from reality. That's what people want from AI, a blending of real life into a fantasy generator but no one is offering that yet.
The impact of AI on creative fields will be pretty nuanced I feel. I don't think we will end up in a world where everyone is just AI generating the media they want. In some cases we will, for example stuff like lo-fi beats on YouTube are already started to be bulk made by AI because really it's just fancy white noise for people to use to work.
Actual music (like what you find on Spotify) I think won't be impacted very much. People strongly identify with the art they consume, and that identity comes from the people who make the art. Those folks might be using AI under the covers for elements of their creative work, but ultimately what people care about is the humanity behind the art. It's the same with film, and traditional art people hang on their walls. We like the actors, the director, the artist, their taste, and who they are. It's why we have celebrities, because we get invested in the people behind the art.
Video games I think will be interesting... I feel they will be more susceptible to being accepted as AI generated. I don't think people identify with them as strongly.
Pop music is so simple, yet so difficult to make a hit. For some artists the music can be mediocre or even fairly bad, and still be a massive hit because pop music is essentially theater and their persona and mystique carries the day.
Some bands were terrible touring artists and rarely put on concerts yet made great careers as studio acts. Steely Dan would be one that produced many hits yet rarely toured, mostly later in their career.
The fundamentals of pop are totally understood. Yet what makes a hit is so fickle and difficult, the bar is extremely high
I could see myself listening to AI music in Spotify, not as the main act but at least as a Plan B.
For example, I like a specific music genre, Italodance, which was popular in the 90s and then disappeared. The problem is that I have listened to all of it, as far as I know. No more is being made. If an AI model could make more for me, with decent quality, I'd probably listen to it.
If you listened to all of it chances are ai will not be able to generate anything that sounds remotely different than what you already are familiar with.
> We like the actors, the director, the artist, their taste, and who they are. It's why we have celebrities, because we get invested in the people behind the art.
But we can't know any of these celebrities as people. We only engage with their images created by marketing. Their stories are as curated and fabricated as the artworks they produce. Transferring these simulacra to AI personalities is merely another marketing problem to be solved.
>The interest in ai music generation is lower than I initially thought. I jumped in but felt the exercise lacked the joy of making music physically or with software like pro tools. With pro tools you control the thousands of knobs which gives you more control. These AI models take away that connection. You can play around with different words to get different results but it's like painting with a shotgun.
You can still pull AI stuff into your music editor and tweak it, although it's harder because it's already mixed. But ironically, this is the exact same problem you have with AI coding to avoid learning how to code - unless you know what you need and how to do it, you're basically relying on AI to one-shot it for you. The nice thing with music and visual art is that it's subjective, so you're the only judge of what's correct. That's why people get super impressed with images in GenAI when it generates 1001 human faces in that setting vaguely resembling what was asked. If you had to generate a very specific thing, it's basically impossible to get it correct.
> No one wants to hear other people's ai songs because they lack meaning and novelty.
Not true. No one wants to pay for other people's ai songs. There are so many AI songs on youtube (mostly lofi or traditional Japanese instrumental) and they cumulatively have quite a lot of view.
The thing is for quite a lot people, music is just something they put in background while doing their office desk jobs. It's just there to make chores a bit more tolerable and nothing more.
> The thing is for quite a lot people, music is just something they put in background w
You are absolutely right. These people only listen to music passively and so it doesn't make a big difference who/what made these tracks. Same for lots of commercial music (cheap TV show soundtracks, commercials, jingles, playlists for restaurants or shops).
But for anyone who actively listens to music and appreciates the style and evolution of certain artists, AI music is not acceptable. The very premise just feels wrong, if not outright insulting.
I like active listening. I can easily spend two hours sitting or lying down comfortably in my headphones, eyes closed, so that I can focus on the music alone. The kind of music I want for that is not (yet) something that AI can generate.
But that same kind of music is also distracting when I'm actually trying to do something, because I keep overfocusing on it. So when I work, I listen to different kind of music. Having AI generate that doesn't feel wrong or insulting in the slightest, nor is it relevant to the other kind of music.
Although I will say that if and when AI is actually able to generate music good enough for active listening, I wouldn't be insulted by that, either.
I think that the tech exists for more interactive AI generation, it's just gonna take time to implement.
I foresee something like your standard production software with heavy AI integration, where you prompt it to make the song you want, but it is made fully step by step in the production environment. You can then manually tweak it or ask the AI to fine tune whatever parameter or slice you want.
Kinda like sitting over the shoulder of someone who knows what they are doing, and working collaboratively with them to accomplish the idea you have in your head. Meanwhile you have practically no idea what all those buttons/lines/glowly bits/sliders do.
Back in the day nobody wanted to hear my FL Studio songs either. Not saying this as a joke, it's just that most people don't care about most people's amateur art from what I've seen.
Identity of the creator and the piece has always been a big part of music appreciation. Imagine how dystopian would be if you ask someone their favorite and the answer is "the music that HitAI generated when promoted 'sad song for happy people' when I was 17, but their model has since changed and my subscription didn't include offline ownership, it's lost forever."
Also image generation, particularly with latest GPT, can be finetuned a lot more than music generation which is nowadays limited to "here's something with those lyrics and genre".
> No one wants to hear other people's ai songs because they lack meaning and novelty.
That's not true. I already found a few tracks that I like. It's actually impressive what Udio can produce. Also ElevenLabs demoed their music generator, and their demo tracks were all quite cool.
I do agree with you that fine controls are missing, and also splitting instruments/voices into separate tracks.
No one wants to hear other people's ai songs because they lack meaning and novelty.
AI image and short video generation can create novelty and interest. But when the medium require more from the person like reading a book or watching a movie the level of AI acceptance goes down. We'll accept an AI generated email or ad copy but not an ai generated playlist and certainly not a deepfake of someone from reality. That's what people want from AI, a blending of real life into a fantasy generator but no one is offering that yet.