While it might not seem like it to a novice, music turns out to be a highly mathematical endeavor with precise ratios between chords and notes as well as overall structure of rhythm and timing. This is especially true of popular music which has even more recognizable repeating patterns and trends, making it unfortunately an easy target for modern generative AI which is capable of analyzing huge amounts of data and creating arguably unique creations. This one, called Suno, does just that for better or worse.
Unlike other generative AI offerings that are currently available for creating music, this one is not only capable of generating the musical underpinnings of the song itself but can additionally create a layer of intelligible vocals as well. A deeper investigation of the technology by Rolling Stone found that the tool uses its own models to come up with the music and then offloads the text generation for the vocals to ChatGPT, finally using the generated lyrics to generate fairly convincing vocals. Like image and text generation models that have come out in the last few years, this has the potential to be significantly disruptive.
While we’re not particularly excited about living in a world where humans toil while the machines create art and not the other way around, at best we could hope for a world where real musicians use these models as tools to enhance their creativity rather than being outright substitutes, much like ChatGPT itself currently is for programmers. That might be an overly optimistic view, though, and only time will tell.
It will always (or for a very long time) be the auditory equivalent of AI art, all square in aspect ratio and full of the absolute average qualities and glossy finish of Artstation or Pixiv, because AI is at heart a great averaging machine. None of it will be memorable beyond the next few tracks, all interchangeable. People will listen to it and love it but they won’t talk about particular tracks after a short time.
There will always be a market for talented and inventive humans. AI can pass the Turing test for the average human, but one of the most steadfast aspects of humanity is that it despises the mediocre. Only the shoulders of the bell curve attract people.
The interesting stuff will be talented people using it as a timesaver or an inspiration machine to eliminate the blank page before working in earnest.
This ^^
You say “AI is an averaging machine” But I recently saw an illustration (not AI generated) The depicted how artists, and all people, really pretty much just copy each other anyways.
It’s only after lots and lots of copying that we find our real voice in our uniqueness shines through. However, what about this is really impressive? Your uniqueness is your flaws and that adds beauty to what might otherwise just be a copy of someone else’s work or style.
All it takes is an AI that can inject uniqueness based on its experience to make this a very similar situation. I’m not saying it’ll happen like it does with people, or that it even if AI can be just as creative as people, it won’t be as meaningful, but I think you’re selling it short to think it won’t happen.
Yes, I was thinking along the same lines – how to inject randomness (good or bad, i.e. flaws) to add some uniqueness to the output. I’m sure they’re working on it and the AI output will be better for it.
But it will still be an output by a computer. It’s either champagne, or it’s sparkling grape wine.
I prefer champagne but can’t afford it.
I suspect that you do not know how such systems actually work, and why they fail badly in specific areas. Go and find an AI that can correctly render “An octopus in a spacesuit, repairing a satellite, in orbit above Earth”. Yeah just like the one you can see in your mind’s eye right now, with the tentacles actually inside the suit. AI can’t manage that yet, because it does just average or interpolate between trained states and that is not what is required in this case. True intelligence requires far more from an artist than what AI image generation systems can offer. An artist needs to apply common sense too, based on what they know but can’t see, the hard vacuum of space is incompatible with mostly water/carbon based lifeforms and the point of a spacesuit is to fully enclose the lifeform completely.
https://ibb.co/G9HCPLN
You were saying?
You just proved his point. The tentacles are not in the suit..
All it takes is for AI trained in physics, chemistry and biology to combine with a generative art AI and BAM! Your point would be mute.
So, more of an AGI instead of AI.
Good luck with that.
Moot
Utter BS that just shows you´re afraid of AI.
Specialized models and the group controlling will have the same fame than some artists. If AI can average humans well enough, it can also outperform them.
There will be always a market for talented and inventive humans for sure, but you won´t be sure that those perform without AI help, and vice-versa.
> If AI can average humans well enough, it can also outperform them.
That doesn’t appear to make logical sense. An average is always inside the range of values you’re averaging, never outside. How could an AI outperform people when the best it does is copy them?
You’re showing a simplistic ignorance about how these models produce their content – which is okay, as it creates an opportunity ripe for learning.
The model’s requirements to produce an “average” output is simply a large dataset. Scale that dataset, and fine-tune the model, and suddenly it can output exceptional works of creativity. This is what we’re beginning to see with the current state-of-the-art models.
“AI” may have started as an averaging machine, but the current capabilities are way beyond that, and it’s still on a very steep cure of increasing that capabilities. “Art” should be experienced from a cleanroom perspective, just the product without knowing anything about who made is and how. From that perspective many an AI product would surely induce just as much emotions as human created “art”. We live in wonderful times ;)
No. No it’s not. All the ai is a Google search.
Claiming that art should be experienced in disregard of its context is quite a stretch. What is the attraction of, say, a cuneiform tablet without the context of who created it and when? Art is 99% context. There may be an AI which can generate passable Van Gogh clones, but it will never have lost an ear.
It’s not a stretch, it’s just purposefully ignoring all the societal mythology around art and artists, which is good. Regardless of their historical significance, cuneiform tablets are generally not good art, and the quality of a Van Gogh painting does not rely on his missing ear. Artists get upset with AI because this mythology allows bad artists to come up with elaborate copes for why their trash is still valid. AI, on the other hand, just makes nice pictures. You don’t need to think about it any harder than that. Aesthetics are an optimization problem, like anything else.
>generally not good art
That’s a meaningless statement, since “good” in the context of art depends on the social context it was made in. Art as a pure concept has no such value. A cuneiform tablet is as good as the best painting of the greatest master, because you’re eliminating any cultural or social notion of aesthetics and purpose. Talent, effort, complexity or simplicity, intent, message, concreteness or abstractness – all lose any significance when you try to view “art” in a contextual vacuum.
>AI, on the other hand, just makes nice pictures
Judged by who? You?
>Aesthetics are an optimization problem, like anything else.
Just like creating the perfect shoe. You simply take all the shoes in the world and find an optimally weighted average of them, and that should be the best shoe for anyone and everyone, for every purpose. One size, one shape – just like aesthetics.
Not sure if you’re describing AI generated music or most chart music…
I expect this will work wonders because pop music is already very derivative.
I’m reminded of the White Stripes song Icky Thump. Something like that would never be generated by ai.
It would never be, until it eventually is. Human brains are just wetware AI.
Let’s be fair that is still very useful and disruptive. Think movie soundtracks. If the music sets the tone properly you don’t need to have it remembered for most movies. Elevator and department store music doesn’t need to be great either. A lot of money is out there that needs functional music not memorable
Think soundtracks to youtube hacker videos.
My guess is that you haven’t heard any of the songs made using it…
I listened to a few and some of them even sent a chill up my spine they were so good, just like human made music does every now and then. I’d urge you to keep an open mind to it.
Lol you’re in denial, AI art is 100x the quality of AVG drawer
“None of it will be memorable beyond the next few tracks, all interchangeable.”
That sums up pop music for me. Cheap shots aside, it’ll just be the entry point. I’ve watched the music industry cycle through genres from the consumer side. A new sound comes out (grunge, r&b, rap), it’s new, innovative, fun, has a viewpoint. In a few years the music business people figure out the formula, and you get formulaic groups who parrot the themes of it without any real heart. AI may never produce original music. What it will do is shorten the time between something original being said, and it being cloned and repeated until it’s meaningless, and the original people to create the sound get drowned out by the repetition.
I have similar concerns for CGI and movies. We’re on the cusp of fully generated performers. The visual aspect is there. The voice generation is almost there. One expression nuance can be generated, studios won’t have to pay actors. Hatsune on the silver screen.
Wow, I’m bleak this morning. Hope everyone’s day gets cheerier.
There are people that talk to each other about particular tracks? Why?
‘That might be an overly optimistic view, though, and only time will tell.”
I remember, back in the early days of the Internet, there was much excitement about the possibilities the Internet could bring, including an end to censorship, breaking down walls everywhere.
That proved to be an overly optimistic view, given how massive network equipment companies like Cisco actively participate in government censorship these days.
Call me a pessimist, but I’ve learned that overly optimistic views tend to be just that: overly optimistic.
Yeah, I remember those days. I actually had a hit record on MP3.com, so sad when it all stopped: https://www.youtube.com/watch?v=CDoBAW3NTG0
I don’t like this future as much as the next guy, but that fake Lead Belly song was an absolute banger.
Digital audio workstation software and synths have had arpeggiators for years, with a human picking out the best sounding melody/accompaniment that fits around their tune’s key, there’s formula for working out key changes and chord progressions but it takes a human to decide which parts fit together and why.
Given that music is constantly evolving, if AI could predict the next trend(s) in music, that would be an accomplishment, rather than just aping what might sound good compared to other things that already sound good.
They don’t need to predict. They just spam everything and see what sticks.
That is the same way we humans do it. There is a lot of human music out there that is absolute trash, but producers, listeners, and ultimately time weed out the really poor tunes from the acceptable ones. If you listen to classic rock stations or any other genre from decades ago, they all play what has survived the test of time, or as you say, what sticks.
I approve using this to make silly songs
like todays xkcd (#2910 for future people).
This isn’t legal. Sampling requires consent from both the rights holder and the artist and this database is just a huge collection of samples so there’s no way they got it approved.
According to wikipedia: In 2019, the European Court of Justice ruled that modified, unrecognizable samples could be used without authorization.
Tell that to Aphex Twin or Daft Punk.
Hackaday is the last place I expected to be negative about AI
They took MAH poster job!
It’s HACKADAY. Not CONSUMEADAY.
Of course a website that wants to know what’s inside pieces of tech (rather than only the UX), or cares about right to repair, wants to preserve history, and constantly showcases humans creating things that manufacturing facilities have already done “better” wouldn’t wholeheartedly embrace machine generation.
Until AI is used to control solenoid motorized physical instruments, then AI music will only be FFT overaliased bit garbage.
it sounds better than vocalists who use autotune. when the real musicians sound like robots and the robots sound like musicians, you have a problem.
I disagree, robotic musicians and musical robots are great. FFT trained audio LLMs sound bad quality wise, and compositional murky often.
AI will soon learn to create addictive earworms that the sheeple cannot stop humming.
But would it still be memorable 20years later?. 50 years later? They are still selling the white album and Voodo lounge
Probably. I’d rather listen to AI music than the Beatles.
… your loss.
Long before AI, I knew some pro music composers/producers of commercial and corporate music who would unapologetically copy or lift aspects of pop hits to use in their output of lucrative but forgettable compositions. You don’t need AI to be totally derivative and unoriginal.
On a tangent, “math” and randomness have often been used to generate new compositions. A relatively well-known example is Phil Thompson and his fractal-based music generator Gingerbread.
https://www.last.fm/music/Phil+Thompson/+wiki
No, AI won’t “create” anything that hasn’t already been created and that’s the entire point musicians are screaming at everybody, but of course programmers think they know more about music than actual musicians do, which is genuinely truly hilarious. Like, the sheer hubris is off the charts. Have you had any hit AI songs yet guys? Are people hopping up to buy AI music? Are human artists recognizing it as good music? No? Okay, then you’re not accomplishing your goal. Do you know what Muzak is/was? Do you realize generative art has existed for like.. nearly a century in music and hasn’t taken everybody’s jobs? Do you know who Andy Warhol was and what the entire message of pop art is was? We’ve already been here and does this like, five different times in the art world. We’re really not that impressed. And that’s how it works, you have to impress us and everybody else – and you don’t, then womp womp. Welcome to art. lol. Oh, and none of it makes any sense. You won’t come up with a formula to “make good music”, we’ve been trying that for centuries with music theory and it’s still just a tool which is all AI will ever be for music. AI can never create the cutting edge of music which is what everyone is actually interested in because it simply isn’t creative. If a human being made the music that the AI is making, I’d say it’s really unoriginal. Sounds cool, but it’s ultimately drivel. There’s no unique substance to it whatsoever, let alone the fact that whatever lyrics might be being said have absolutely no soul, no real meaning, and no flavor other than sounding like something you’ve already heard before. Anything else?
its nice to hear frank sinatra performing master of puppets. it really confuses the old people.
I have a tune in my head – but no musical ability. Heck, I don’t even know how to describe the tune, unless I refer to other, similar tunes. It will be interesting to see how how I get, describing it to an AI.
Easy AI composion was mentioned for POP … any others? Some of the stuff kids listen too seems to be very repetitive sequences, and a type of poetry narration. Perfect marriage of formula tunes and ChatGPT. This seems to be a two-way transaction… formula based music, and the discerning ear, or ability of the listener to appreciate complexity and talent? Or settle for less complicated stuff?
There was a pretty good sitar one on their chart when I checked it out. It can do plenty complex stuff.
Here I just say AI is for Algorithmic Instrumentation. That’s been going on since Tin Pan Alley was making “hits” out of what fits in the middle of the bell curve, and yes some of those tunes were available in that non-human digital interface for your piano. Come forward a century and the hits keep on coming with everything a copy-paste idiom coming out of the studios. So I think it will be just more of the same and worse, but then I bailed out of that bell curve stuff long ago.
I created an EP using AI last year. Technically, it was cowritten with it. Four of the tracks were written, or co-written, with AI.
One wasn’t.
If AI is stealing our jobs, and diminishing the role of the artist, then you should be able to tell which it is! :) can you?
https://nodemusic.bandcamp.com/album/co-written
I listened 2 2 of the tracks. Neither sounded likely be remembered in 50 years time, I’m afraid :( …. but most of my comments are generated by AI, so who am i to say?
I can’t tell. All of them are annoying electronic “music” of a kind I’d never listen to voluntarily.
If that’s your comparison for good music, it isn’t surprising that you think AI generated music is acceptable.
Sci fi stories come to life. Maybe AI music will have AI audiences, and develop its own new musical forms that appeal particularly to AI listeners…
The Ray Bradbury story “Night Call, Collect” comes to mind. Anyway we appear to be in an exponential curve of AI progress.
So here is the problem for a part of the music industry that collects fees for the playing of music in public places where it is broadcast e.g. shops, malls, forecourts, etc.
So AI music will not attract a fee :-( and some start-up will offer a music generator for a one off price.
No, AI won’t “create” anything that hasn’t already been created and that’s the entire point musicians are screaming at everybody, but of course programmers think they know more about music than actual musicians do, which is genuinely truly hilarious. Like, the sheer hubris is off the charts. Have you had any hit AI songs yet guys? Are people hopping up to buy AI music? Are human artists recognizing it as good music? No? Okay, then you’re not accomplishing your goal. Do you know what Muzak is/was? It was music mass produced for the sole purpose of commercial use. That’s been around FOR LONGER THAN ANYONE HERE HAS BEEN ALIVE. Is it popular? Absolutely not. Do you enjoy the music companies play when you’re waiting on hold? No? Okay. Great. Do you realize generative art has existed for like.. nearly a century in music and hasn’t taken everybody’s jobs? Do you know who Andy Warhol was and what the entire message of pop art is/was? We’ve already been here and does this like, five different times in the art world. We’re really not that impressed. And that’s how it works, you have to impress us and everybody else – and if you don’t, then womp womp. Welcome to art. lol. Oh, and none of it makes any sense. You won’t come up with a formula to “make good music”, we’ve been trying that for centuries with music theory and it’s still just a tool which is all AI will ever be for music. AI can never create the cutting edge of music which is what everyone is actually interested in because it simply isn’t creative. If a human being made the music that the AI is making, I’d say it’s really unoriginal. Sounds cool, but it’s ultimately drivel. There’s no unique substance to it whatsoever, let alone the fact that whatever lyrics might be being said have absolutely no soul, no real meaning, and no flavor other than sounding like something you’ve already heard before. Anything else?