I've voiced my opinion on "AI" being used to make "art" before, and I'm going to do it again: I think this is one of the worst things you can use AI for, as it narrows down the possibilities of one of the most enjoyable HUMAN activities, which is CREATING art.
I loathe the vision of a future where anyone can ask a neural network to create music just by describing what they, the user, want to hear, without them necessarily having any musical chops at all, and where some douche will train a neural network to generate every possible melody and chord progression that could be enjoyable to the human ear.
I'm saying.. You had thousands of real problems to solve using AI, and you chose to use it to make living more miserable for anyone who sees music as something more than a mere arrangement of sounds.
Technically speaking this demo is close to brilliant, really well done. Most polished AI music generator I've seen yet.
But the marketing copy really feels off to me.
"Bringing the joy of music making to everyone."
The joy of music making involves taking an active role in the creation of the piece. The joy of music making is not getting a finished piece delivered to your door without any effort. The joy is in the process, not the outcome -- especially if everyone can generate the same result as you. There is no emotional investment.
And I say this as someone who plays acoustic instruments as well as program drum machines, and writes DSP and MIDI sequencers. Whether you're playing a piano or programming a cool drum beat the joy comes from taking an active control of the creation process and finding something new to create, unlocking secrets new and old.
I would accept a marketing angle like "generate free to use background audio for your podcast / video / video game / commercial / whatever".
But IMO there is no joy in pressing a button.
Maybe their other products mentioned on their primary brochure page let the user go a bit deeper, and that I think has potential. Garageband is a great example of giving the user creative control but simplifying a ton of the complexities of finishing a song.
I don't get it at all. How exactly does AI prevents you from creating art? If you want to write your own song, grab a guitar or midi keyboard and go for it. Do you really think AI will make a simple guitar unaffordable in the near future?
Yeah, AI might make "manual content creation" less profitable... but you said creating art is an enjoyment itself, right? If you only do it when it's profitable, then maybe you don't enjoy it... you enjoy the money. Most people are willing to spend time with their SO even it's not profitable. That's how I know they enjoy it.
It’s not that it will prevent you. It’s that no one will want to listen to it.
Do you ask your parents or google?
Do you ask strangers for directions or google maps?
Do you wear sweaters knitted by other people? They’d have to pay YOU to put up with their sweaters and wear them.
No one outside of a niche or your immediate family will give a shit about your music, buddy. They already don’t give a shit about most other things people used to do, after automation replaced it.
If it makes you feel better, already hardly anyone really reads the vast majority of comments and articles people write. It is solely an exercise in releasing your own stress LOL. No one cares!
For the vast majority of artists in any genre no one pays any attention anyway. The threshold where people starts to pay attention to you is really high. If you make art for the sake of adulation, sure, good enough AI generation will make that even harder. But your odds of getting to that stage are already vanishingly small.
That’s true. It is already hard for any humans to rise to the top of the statistical distribution. Soon it will become nearly impossible, as it is in chess for example.
But a great analogy is whether people are interested in watching chess games between players rated 1300. Maybe, for like 2 minutes. Then they get bored and move on. The two players are interested in that game and that’s it !
> I loathe the vision of a future where anyone can ask a neural network to create music just by describing what they, the user, want to hear, without them necessarily having any musical chops at all
As a musician I have to say I don't care at all. In fact it seems very, very cool that people without the massive amount of time and money privilege required to learn music theory and practice well enough to make an enjoyable song, can unleash their creative ideas and hear something they find interesting. It's akin to listener empowerment, and I'm all for it.
When millions of digital compositions a second can be created by swarms of AI, all digital human endeavor will be drowned out. The market cares not anymore that you can knit a sweater.
What this means is everything you offer as a musician will be as irrelevant as renaissance music and classical music is today for the vast majority of people.
Same was said about microphones and amplifiers, the drum machine, sampling...
Music just adapts. New methods, new sounds, new genres. And it won't get easier to create music, it will just get easier to create certain parts of it (anyone can play strings now with an orchestral sampler; that was unthinkable 50y ago). It will still be brutally hard to stand out from the crowd.
Nonsense. Those things didn’t churn out new finished compositions. At best, they added a track using a few parameters.
This is entirely different. Your comment can easily have been generated by an LLM. Operating at scale, human comments and articles would literally become 0.00001% of all comments and articles, and on average probably would be lower quality anyway, in both breadth of “knowledge” and lack of civility. Once people come to overwhelmingly prefer digital content, all those charming platitudes about how human ingenuity / whatever can never be replaced, will melt away. “The crowd” will be 99.9999% automated processes.
Do you want a human to make your fries and smile a fake smile as they give them to you?
Do you want a human to operate your elevators? Produce your clothes and shoes by hand? You are spoiled by the thread count and consistency of machines there. Why not in digital content?
I've never made music as relevant as renaissance or classical music, but that's not my goal. I play for my own enjoyment, turns out some people also like watching me play which is pretty cool.
I think it's easy as an artist to put our own social position ahead of the actual goal: listeners should be able to enjoy music that evokes feeling/joy/relaxation/excitement. Whether that's 300 year old classical, a garage band or AI generated pop doesn't really matter. Lots of people had the same complaints about 3 chord DIY punk. Gatekeeping the creation of art seems elitist and anti-listener/viewer/reader...
I don't need people to enjoy watching me play. I'd still play even if it was just for myself.
That said, I imagine IRL people will always enjoy a demonstration of skill. That's a bit orthogonal to just enjoying the music on its own though since it includes an element of performance. If robot bands could put on super entertaining live shows and produce amazing (better than human) music, I'd be first in line to listen.
Creating art is an enjoyable human activity. I totally agree. But "creating" is subjective. At pretty much every technological step for music in my memory and somewhat before, people have lobbed something like your objection against it.
- Recorded music is bad because it kills live performance.
- Effects are bad because it takes away from natural sound of music. Processing vocals is bad because it lets less skilled vocalists supplant better singers.
- Studio techniques like splicing multiple performances together to get one good one are bad. If a guitarist can't play a solo in one go, it's clearly bad. (IIRC Def Leppard, for example, got a lot of side-eye about the album Hysteria for this.)
- Synthesizers are bad because they're artificial and/or replace actual musicians. (Had a class with a guy who was unbearable about synths because he played horn and if a band could just program a horns track / play horns with synths, he was out of a job.)
- Sampling is bad because it just copy/pastes previous ideas and that's not really creating art... (Raging debate in the 80s about whether sampling was real music, etc.)
I see music as "more than a mere arrangement of sounds." I have spent quite a bit of money over the years to see live music, buy albums by artists I value, and there's still songs that send shivers up my spine every time I play them.
But I don't think an AI that could deliver on creating mashups or generative music is going to make living more miserable. Monetizing everything is what makes living more miserable. Maybe they're intertwined to a large extent. But if I had an AI that could create music that actually matches my descriptions, I don't think that would make life miserable or take away from actual muscians.
e.g. - I'd love to be able to get an AI to create something like a John Lennon song written with Khruangbin, or "what if Miles Davis played with John Entwistle from The Who?" That doesn't take away from any existing music or future music.
I agree with your perspective on this, but my hope is that it won't have a major negative externality at the end of the day--drum machines and drum software have really solid for well over a decade now, but plenty of people still learn the drums, learn them well, and find opportunities to play them.
The odd thing is, the motivation for the research was not really to build this stuff as a product.
If you're going to do machine learning dealing with images, music, or text, a completely natural thing one might want to do for many reasons -- to solve "real problems" -- is to try to approximate the probability distribution of the data.
Having done this, it was seen as a kind of strange curiosity that you can actually sample from this probability distribution, which amounts to generating new images or new songs or new text.
It was only recently and somewhat unexpectedly that these generations actually got good enough to be interesting to people in themselves, and then people started trying to turn them into products.
There will be a middle ground where the technology augments professional tools.
If you can pull open an audio editor, start with 20 tracks and have the ai pull things I like from other songs AND modify each track to match as elements get added, you’ll start to see pros make some amazing things. Similar to how people with a photography background have a huge head start with midjourney.
What you are describing will be more powerful when I can make a playlist of my 10 favorite songs AND write a sentence about why I chose each song, and what features of the song are worth preserving. Then it can be used to match me with existing or generated songs. Current gen streaming radio doesn’t have the context for why I like each song I like. Use the ai to take songs apart and understand their elements.
This video is something I would like generative ai writing to understand. https://youtu.be/QWveXdj6oZU. This tools current rhyming isn’t any better than Kurtis Blow.
My main frustrations of AI and art is that people often frame art as a product the creator has made, instead of something part of the creator.
What I'm referring to here is the parts of self that artists put into their work. For example, many of Raymond Carver's characters in his short stories are blue collar workers in the mid-west. He's writing from experience. He was a blue collar worker from the mid-west.
(I'm not referring to 'separating art from the artist', that's a different conversation).
In many aspects of art, art is certainly a product that the creator is making and intending to do so, but in other aspects it isn't - it's something the creator is making as part of themselves.
In conversations with AI and art, especially with AI creating visual media, I feel like people often forget this point of view.
Does this mean that someone using AI to create art can't put some of themselves into it? No, I don't think so.
Does this mean that art created with AI that has no human element in it, aside from prompts, isn't art? No, I don't think so as well. Art can be art without any human element at all. One can even consider prompts part of that human element.
It's just my opinion that this point of view is often missed in discussions. People sometimes make art because they want to, they have something to say, they have feelings they want to express, or they just want to have fun. AI can be used in this way as well.
I think people view artist's complaints similar to people losing their jobs in the industrial revolution. This is in some way correct, but it also just completely skips over the emotions, soul, and self that people put into what they create.
At the end of the day though, I don't think what I'm saying here is new at all. There are intentional artists, and then there's zombie formalism [0] or songs written by a team whose job it is to make a hit. Sorry for the rambling, I just wish people viewed art more than a product to be sold.
And at the end of the day, maybe my opinion is just my own. I love reading about the process an artist takes and their entire life, but I know not everyone does the same.
This is only my interpretation of things, but I've noticed that people who already listen to music in a way where the music is just "filling a hole" in their sonic space, where the music doesn't get the attention it deserves, where the possible lyrics aren't even paid any attention to, where the artist and the art are never even put together, making separating them from each other a moot point, tend to have a more positive view towards AI-generated music.
They don't spend time analyzing music or the lyrics of a musical piece. They aren't putting the art in the context of the life of the creator, and much less trying to understand the music. They are listening to the music for what it is to them: JUST music.
Contrast this with another way of listening to music, a way that's much closer to my heart than what's described above: imagine the music itself evokes something in you, after that so do the lyrics, after which you decide to read the liner notes, which prompts you to Google something about the artist, because what you found is so interesting that you want to try to figure out WHO is behind the music. Maybe you want to try to learn what in their life prompted them to make the music they did. You might read interviews of the artist, you might watch a documentary – you might be subjected to a totally new world-view because you are getting to know the artist.
One day you come back to the song that prompted all this, and a fragment of the lyrics reminds you of something you read about the artist and it all of a sudden something clicks and you've put the song's lyrics and meaning into a whole new context.
I don't know if I need to say this, but I don't just loathe AI-generated music; I also loathe making music just for music's sake. There's something more in good music than just the music, and I don't want a future where all that's behind the music is a pile of linear algebra.
Couldn't you extrapolate this opinion across all current uses of AI, such as programming, text generation + image generation and argue the same things?
They are all creative, yet it seems a little silly to say we shouldn't use AI to progress in any of these areas.
I'm a music lover, and see no problem with AI generated music. It will co-exist nicely with existing forms of music, and perhaps even teach us something new.
One the one hand, I agree that it feels scummy that so much AI effort is devoted to mimicking human creative output. I strongly believe that as AI-generated images, sounds, and even films continue to improve and proliferate into products and services, a good chunk of people will actively try to boycott AI output, seeking out alternatives that are (somehow) certified to be 100% human-produced (Ah, but the tools those humans need will also become more AI-powered...). There will be a growing humanist movement devoted to protecting "essentially human" qualities, such as creativity, from automation.
My attempt to be pragmatic: the AI art genie is out of the bottle. There's no going back. Even if all AI art progress were magically reset to 0 today, it would arise again at some point as people retried to create software in their image. We need to figure out how to adapt to a world in which the barriers to creating "good" art (however you choose to define "good") are far lower.
The central problem as I see it: the faster than bigger the tech advance, the more unintended and unpredictable societal consequences it will have. Laws come to mind as the main tool to mitigate uses of tech that are collectively deemed harmful. But compared to the pace of AI progress, laws will crawl lightyears behind where they need to be.
A lot of people don't have the motivation to learn all the necessary techniques to get to a capable level of artistry. I have a feeling some of those people are too insecure about themselves to enjoy creating art by hand. Those people will probably be drawn to AI tools instead, where they don't have to think deeply about themselves and their experiences to have a finished result in a few minutes. Those people will probably enjoy whatever process AI gives them, because before they had no process that didn't feel like subjection to self-flagellation for hundreds of hours, but some part of them still wanted to produce "art" anyway.
That is not to say that most people feel that way. In the long run that kind of thinking is detrimental to creativity and a healthy spirit. But it gives those people a voice where they wouldn't have had one before, whether or not that's a net positive overall. It's probably going to create a sort of artificial codependent relationship between the artist and AI that serves as an enabling device for quick results with little to no introspection. In the past that probably would have reached its limit at the sketching phase or something, but now you can get somewhat decent results with that mindset, shattering the ages-old idea that only hard work can create passable final results, and the convincing nature of the outputs will encourage people to entrench themselves further in that mindset instead of taking the harder path.
But of course there are shades of grey. There is nothing stopping professional artists with a lot of confidence from generating a base image and doing significant editing work in Photoshop. Even if there's no joy in it, for a certain lower bar of quality like placeholder/concept art, I think the productivity gains will outweigh arguments about virtue, especially once the tech managers have all awoken to the endless possibilities AI provides.
One more thing, I'm very interested in how humanity as a collective whole can "choose to use" AI for any given purpose. Who's deciding what to use AI for? The "douches" smart enough to uncover the math that's always been there and publicize their results? The laypeople that find the papers online, develop their own versions and spread the results around link aggregators because they think it's a cool weekend project? It's going to be nearly impossible to completely control either group, although the researchers are probably going to adhere to a stronger code of ethics than any random person on the Internet.
But as of right now, what stops the researchers from developing a future AI that you can attach to a skateboard and use to pull off perfect versions of any given trick every time? It would ruin the experience for traditional skateboarders that have spent hundreds of hours learning the same tricks, that's for sure. But what stops them from attempting to develop that if enough prior art was there?
I have a feeling that most regulations and laws are written in blood. I think most AI researchers at this stage don't have enough foresight to look even a couple of months into the future with regards to potential consequences after the release of a new breakthrough. Maybe they figure it's better to release their results right away and let the world at large sort out the effects. It's going to take a few more Stable Diffusions disrupting and dehumanizing other art forms for enough people to shout at governments that something must be done to stop ourselves from becoming too innovative.
And even after that we probably won't have universal international law or code of ethics banning the development of new AI. Not until it's reached the perception of gain-of-function research in the eyes of the public, which would mean widespread association of AI with the life or death of millions. And unless that happens someone that isn't affected by those regulations could release a better AI into the world.
This is great, but please let us use it without vocals! I mostly listen to instrumental, so this was unexpectedly jarring. Even adding "instrument" generates vocals, and the vocals aren't exactly great for the music styles I've tried.
Suggestion: Regarding the "Try Again" button: The name suggests it will try the same prompt again. Perhaps make it do that, and add another button called "New" that does what the Try Again button is currently doing (starts a new prompt).
Ah it would've been good to know that, I also tried a couple different prompts but they all had the same vocals. A small note indicating as such would've helped.
Also tried a couple prompts to get just ambient music, but vocals are always included? An option to include vocals would be nice to have.
This is shockingly good, lots of fun. It does not really get the styles I ask for, and as observed elsewhere only really generates hip-hop style vocals, but man this is very cool.
I asked it for something ambient that made me feel like the world was wondrous and vast. I was aiming for something that'd fit an exploration game like Subnautica, or Venineth. The melody was in the ball park of what I expected, and I didn't expect vocals at all but when they kicked in they were kinda hop-hopish? Lyrics even kinda made sense. It worked better than I thought and it wasn't a combination I'd have ever come up with on my own.
I agree- was asking for a lofi haunting melody with samples from the movie Groundhog Day (a song I wish existed but haven't found it). The lyrics were right on theme but the rap was unexpected. Overall, much more polished than I was expecting
Threw a couple of things at it - no matter what I proposed it wanted to layer a rap vocal on top of it and didn't really sound anything like what I proposed. (e.g., "70s prog rock instrumental about traveling through space" gave me acoustic guitar to a hip hop type beat and vocals.)
I loathe the vision of a future where anyone can ask a neural network to create music just by describing what they, the user, want to hear, without them necessarily having any musical chops at all, and where some douche will train a neural network to generate every possible melody and chord progression that could be enjoyable to the human ear.
I'm saying.. You had thousands of real problems to solve using AI, and you chose to use it to make living more miserable for anyone who sees music as something more than a mere arrangement of sounds.