I've spent the past 24 hours being quite astonished at Udio, which I already consider vastly superior to any AI music generator I have heard to date.
To be clear: I share this for the purposes of awareness and conversation, not endorsement.
The lyrics I gave it were random text relating to a conversation I was having with a friend this morning. Since Udio only gives you 33 second clips, I had to extend it three times (adding an instrumental into, 2nd set of text and final, largely instrumental outro)
BTW, as of this morning, Udio is claiming it is removes artist names (here "Richard Strauss") from the prompts, but as you can clearly hear (if you know his music), they must still be using it in the underlying mechanics
---
Good Morning Soo
The dogs need a walk
Can you hear them?
I hope you have a lovely day
Are you getting the groceries and the meal kits?
Yes, this day is for shopping.
How marvelous!
---
You need to be a member of composersforum to add comments!
Replies
Right- I don't think anyone is claiming the software has the ability to think for itself and produce anything that is authentically original. See the last paragraph of my thoughts. As mentioned, that will be the tipping point, if and when that day comes.
In the meantime, it seems that it's gradually coming up to the point where it could replace a lot of cookie-cutter work: the kind of musical filler that like bad romance novels simply recycle the same worn out plots, changing really only the names of the characters and altering some irrelevant details to make it appear original. I wouldn't say it's entirely a bad thing, since there's not very much value in these sorts of low-effort pieces anyway.
As for the tipping point of being truly original -- I'm probably biased, but my opinion is that it's still a long way off. So far generative AI algorithms are still unable to solve basic math problems -- it can spit out the right answers to specific problem instances that it saw in its training data, but the occasional "hallucinations" expose that the underlying algorithms really only ape the surface resemblances of its training data, while lacking true comprehension of the underlying patterns. It's especially disturbing that, thanks to its uncanny ability to ape surface forms, it can confidently present wrong answers with an air of authority that can mislead the unwary, and even more disturbing that some people are believing the hype and taking these wrong answers seriously. Real computer algebra systems that will actually give you the right answers are meticulously designed with advanced mathematical algorithms that deal with mathematics at an abstraction level far, far beyond the basic primary school arithmetic that current AI algorithms haven't even begun to grok.
Now of course, such basic arithmetic skills can be hand-coded by the AI implementors to coax it to give you the right answers in specific scenarios, but that's cheating because the AI hasn't actually learnt anything from the training data, the human merely coded it manually and retroactively patched it in. Without the human helping hand, the AI remains a dumb interpolation engine. Such manual interventions merely prove that the real underlying AI algorithms are still, at their core, far below what it takes to even begin to approach true human understanding. To get past this point, the underlying AI algorithms must reach the point where they're able to infer such algebraic rules without any manual intervention, purely from being fed the same kind of training data that a school kid would be exposed to. As I mentioned, this would require a fundamental breakthrough, a kind of unprecedented insight and radical change in approach, that go beyond what current algorithms are doing.
Based on what I'm seeing of AI these days, I'm not holding my breath.
To make a long answer kind of short, though;
You're comparing mathematics to art. There is really no "right or wrong" answers when it comes to music that will satisfy a client. If they like the result, then it's a "right answer" versus something that is undoubtedly right or wrong like the answer to a complex math problem.
As you mentioned in your post, and a point I agree with, the software already can do better than shoving a cookie cutter library or generic track in the background. In a matter of seconds.
So on that topic, I have to imagine it will mean less work for composers. Especially if you want a quick little song with lyrics in it, you can really tell this program to say whatever you want and you'll have pretty convincing results in a matter of seconds.
Clients can like and approve whatever shitty music they please. That's right. If the music has been trained on that of other unwitting or unwilling composers, and especially if it's not a composer getting the job but some prompt enthusiast, I'd say it's as close to being objectively wrong as a moral nihilist like myself could accuse anything of.
The point is not about right or wrong answers to math problems. That was only an illustration to show that current generative AI is incapable of reasoning. It's pretty good at aping the process of reasoning, but if you look at it deep enough, you'll discover that it's merely smoke and mirrors. It has the appearances of sophisticated reasoning but lacks the actual substance.
If a client is satisfied with the kind of mindless regurgitation of training data that current AI produces, I'm glad they aren't my clients.
"If a client is satisfied with the kind of mindless regurgitation of training data that current AI produces, I'm glad they aren't my clients."
Well, just to poke the bear a little then -
How is that much worse than a producer, music supervisor, etc., going to a music library with thousands of music tracks that sound nearly the same and picking any old generic track? Composers write that music specifically to be generic with hopes it could apply in a variety of contexts. In my opinion, it's part of the reason why most music we hear now in media is completely uninspired nonsense that (literally) thousands upon thousands of composers could replicate and consciously choose to replicate because it pays the bills. Composers are already regurgitating one another, alongside their own works, very often. And composers writing for libraries already rely on technology to produce generic or formulaic music quickly and cheaply.
I agree with what Rowy said in the sense that this technology is likely going to eliminate composers who make a living off of library music and syncs, those composers that churn out quantity over quality in the hopes of getting placements by sheer volume. Once that happens, only those who survive after being forced to become significantly original or produce expertly crafted material will have a chance at standing out. Why pay thousands of composers royalties and spend potentially hours sifting through tracks and attempting to sync them up when now, or in the very near future, those in the music department can just feed the AI platforms a series of prompts and get something very close to what they need - and even better, they'll own the music and the production company will collect the royalties, assuming the laws don't change or update anytime soon. I'm not saying I support it or endorse it, but if I look at the situation without bias and from the very realistic perspective of cost-effectiveness and time efficiency, especially in media, it's the obvious choice for low-budget, quick turnaround music needs.
Since I really don't deal with music libraries much anymore, for many of the exact reasons mentioned here regarding concerns of blending into an ocean of thousands, originality, and being encouraged to be generic rather than original, I don't even know if it will be a bad thing if they get replaced by AI that can produce music of a similar quality (and probably better in the near future). Still not saying it's original, but it will force the need for more original material on the bigger budget projects that have a need for music that a computer cannot produce (and projects that want music that will not blend into the ocean.)
We may not be there yet, but it is coming. For most media projects involving music, the focus hasn't been "art" or originality for quite some time in that business. It's about money, getting it done quickly and cheaply, connections often, and for the best result possible. Not always, of course, as there are those rare projects that may prioritize the music or focus on the music specifically - but naturally, that tends not to be the norm.
Yes, that's what I'm trying to get at. Like it or not (and personally I don't like it, but it's the harsh reality) the AI hype is going in the direction of replacing low-effort music productions. How far it is along that line of development I can't say, not being personally involved in the industry, but it seems inevitable. The economics of it, the social trend, the tech, it seems unavoidable at this point. (And I'm not saying this is good or bad -- personally I have a lot against the trend, but my opinion does not change the reality.)
On a sunnier note, though: perhaps what will ultimately happen is that this will lead to computer-assisted composition tools where the computer will produce an initial draft of the music based on the composer's prompt, and then the composer takes over and tweaks it, adding the creative element that the machine is unable to create on its own. Rather than a wholesale "type in XYZ at the prompt, get a fully-finished product from start to finish", it would be more like "I want features XYZ in the next 16 bars, give me 3 mockups and let me see if any of them could be of use to me". I.e., use the tech as a toolbox to aid the human, rather than as a replacement for the human. The same way tech (in the broader sense of tools created by humans) has been employed since ancient times to enhance, rather than replace, the human element.
Music libraries, and media using them, goes back some time. It's not a modern facet of the industry at all, and the difference is still that a human created those pieces. You can play devil's advocate along the lines of "modern composers regurgitate each other so why is AI different" if you like; it still IS different. A perceived lack of quality is not equivalent to, or sufficient reason for, generative AI outright replacing humans in the workflow.
It being low effort to you isn't really the point. The most banal filler score to a third-tier TV series has still been composed by someone, possibly even orchestrated and recorded rather than just using samples. Sure, I could shit out that stuff easily, probably most of us could, but why does that make it valueless? Do you think AI should be used on the jobs you think are beneath composers? The composers who want those jobs are unlikely to agree.
You've been complimentary about my music, but others have called it cookie-cutter just as you have the music of others you wouldn't mind being AI-replaced. Should I be replaced?
Sorry, I just fucking hate generative AI and cannot understand any artist, let alone anyone here, getting involved with it for any reason beyond assessing where the tech has got to.
I think it inevitable that within five years you will have "composers" passing off AI "music" as their own work. And I also think that this will cast real doubt on the music of real composers as being entirely not AI. Yep.
Wesley