Skip to content

AI song generation – the value of economic basics

Musicians face an artificial intelligence nightmare from the rise of ChatGPT-like song generators, the world’s biggest record label has said.

Universal Music warned that AI-created music threatened “widespread and lasting harm” to artists and threatened a Napster-style crisis without robust copyright protections.

So-called generative AI models have already caused uproar among illustrators for using human-produced work without compensation to create art. The rise of ChatGPT, which produces authentic-seeming poems and essays, has caused concern from publishers about a tidal wave of AI-generated material.

Both Google and OpenAI, the company behind ChatGPT, have developed software that creates vocals and music in the style of certain artists and genres.

OK, so is this the songocalpyse? Let’s assume yes. We’ve now got an unlimited supply of mildly inventive but largely derivative songs and music. This will screw over all those who produce largely derivative and mildly inventive songs – just about all songwriters.

Is this bad for society?

Well, back to basics then. Songs are a public good. Very difficult to produce one of any quality. But once produced anyone can copy it, that copy does not diminish the amount of it available for any one else. It’s non-rivalrous and non-excludable – a public good. We tend to think that things like that will be underproduced. For if the profit ain’t there then the incentive to do the difficult bit ain’t.

So, we institute copyrights, to provide the excludability and thus produce a possible profit, the incentive and the production.

So, where are we with AI songs? We’ve got no constraint upon supply any more, do we? We’ve moved songs from a public good to a non-economic good. There’s no constraint upon supply (OK, a little bit, the cost of running the AI but Pfft) therefore there’s no need to the profit, nor the incentive, to generate supply, is there?

Abolish copyright.

We’re done.

27 thoughts on “AI song generation – the value of economic basics”

  1. A real musician can beat “AI” hands tied behind his/her back at the bottom of a lake on a lazy sunday.

    But then again.. The music industry has pretty much done away with actual musicians for a while now, so I guess the whole AI thing would be a concern for it.
    Let me get my violinette..

  2. The commodification of music has already happened. First Napster, lately Spotify. Artists have to resort to outlandish behaviour and appearance just to stand out. After the recent BRIT awards, the Twitterati laughed and shared photos of Sam Smith and Harry Styles in ridiculous costumes. But that’s the point: they made you look.

  3. If most of the money to be made for an average musician is from touring then what difference does it make who writes the songs? People go to gigs to see them performed, and they don’t want them performed by a computer.

    Elton John doesn’t write his own songs, the bloke in the background who does so could easily be a computer. The paying punter doesn’t care.

    (Disclaimer – I wouldn’t be seen dead at an Elton John gig)

  4. Tim, it’s the Humanapocalypso.

    Let’s think about this:

    Most voice actors and audiobook narrators are done now. They don’t realise it yet, but machines can do the job far quicker, cheaper and better than most humans are capable of, can give you almost infinite “takes”, and can be tweaked to your heart’s content.

    Pretty soon, human voice actors will be an expensive, handcrafted luxury. Kevin Bacon is expensive, Computer generated Not Kevin Bacon works for free.

    Artists have also been replaced. Algorithmically generated art is more than good enough for the majority of things that currently get commissioned.

    Want illustrations for an article, or a book cover, or just for fun? Playing around with AI prompts for 10 minutes will give you a vast range of options for the cost of about 5p of electricity.

    Want to write some blurb for a website, a marketing brochure, or even a book? You’ve got the idea.

    Computers have just made the majority of human creatives redundant. For their next trick, they’re probably going to automate away most human backend admin jobs. Idk how that’s going to change society exactly, but it feels like it’s the beginning of something profound.

  5. But you do want to reward the non-derivative artists that create new genre. As in other areas the issue is often what is the pipeline and process that gets you those people. Do you need a bunch of ‘apprentices’/’struggling artists’/’perfomers’/’architects’ doing effectively easily replaceable roles simply because if not did you cut off the top of the pyramid where the main value now lies – although notably that presumably changes the approaches of the large labels significantly. Similar issues around many of the creative sectors and quite a few others.

  6. oh and i failed to find a decent AI solution for a range of ‘alien emerging from a [named politician] skin suit’ given the recent headlines so any recommendations welcome!

  7. Aha! “Do you need a bunch of … ’architects’ …?”

    The idea of making most architects redundant warms my heart.

    As for pop music I’ve assumed for years that it’s all churned out by computers: mass produced, like baked beans, but less healthy.

  8. One does wonder whether AI produced sci fi or thrillers will be any good.

    I suppose I’ll have to just wait and see. But since humans can produce some pretty dreadful trash, no doubt I’ll still have to rely on a good author.

  9. This evening I am going – together with 30 or 40 others – to a choir rehearsal conducted by a professional opera singer who is a wonderful teacher and also makes everyone laugh a lot. The music we’ll sing is by Puccini, Verdi, and John Rutter.

    If you’re telling me computers can replace this experience or this content, I am telling you – with all respect – that you don’t know what you are talking about.

    Of course, derivative crappy pop music is quite a different thing.

  10. I would suggest that millions more prefer the pop music and whilst not fans of pizza, vardy and John butter, would have the respect to think – meh each to their own

  11. 1st phase: Machine made/automated X is quick/cheap/convenient but to get the detail and refinement you need a highly skilled human.
    2nd phase: The machines get more refined, the users decide they never needed those frilly detailed bits anyway.
    3rd phase: (Highly skilled?) humans make deliberately wonky stuff just to show it wasn’t machine made; some people will pay more for this just to show they can afford it.
    4th phase: you want wonky stuff, a machine can do that too, nobody cares.

  12. Peter – If you’re telling me computers can replace this experience or this content, I am telling you – with all respect – that you don’t know what you are talking about.

    If you told people from the 1970’s about social media, they’d have thought you some kind of weirdo. Who needs Tinder when you have the Miner’s Club?

    AI isn’t just another set of tools, it’s a (trying and failing to avoid bullshit buzzword bingo here) paradigm shift. The internet has changed human behaviour in a remarkably short span of time, AI offers similar potential to disrupt and reorder the way we work, communicate, and consume “content”.

    Change isn’t always for the better, that’s for sure.

  13. DJC- yeah the value moves to the thing that’s rare. Human music types will move to “hand painted”, “homemade”, artisanal music stuff. Most of them still will make diddly squat but you’ll get a few hits. I love the idea of getting rid of copyright for something that ain’t rare or hard anymore. What i guess will happen is that the definitions will migrate. So that at some point Paul McCartney and his estate benefactors 70 years hence will have been deemed to come up with 50% of AI generated stuff. Would this explain some of the valuations for back catalogues?

  14. I remember looking at the potential of creativity within AI as part of my MSc nearly 20 years ago. While today’s systems are a lot more sophisticated, they are still curtailed by the same boundaries. “Write something that could have been written by Beethoven” was quite doable by AI back then, there is a large corpus of known works and interpolation is easy. “Write something that is generally considered aesthetically pleasing and is an entirely new style” was not possible then and still is not possible now.

    Much entertainment is had in using AI to make things cross genres (“Hey ChatGPT, rewrite Richard III in the style of Jeremy Clarkson”) but note here this is what is being done: do X in the style of Y. Anything “in the style of” is inherently derivative and thus not original or creative in and of itself. What we have instead done is added a new tool for people to be creative — the originality is not in the work itself but instead in the mind of whoever came up with the command that was given to the AI — thus it is the idea of “rewrite Atlas Shrugged in the style of Richard Murphy” that is creative rather than the output itself, just as we credit creativity to Picasso rather than his paint brush.

  15. “One does wonder whether AI produced sci fi or thrillers will be any good.”

    I think AI will be able to do derivative works. It might write the sort of thing you get on Netflix, which is obvious and cliched. It’s never going to write Die Hard.

  16. As Andrew Orlowski likes to point out, ChatGPT (and similar programs) are just the world’s worst autocorrect system.

  17. Puccini, Verdi, and John Rutter/ derivative crappy pop music
    All music is derivative. If it wasn’t derivative, you wouldn’t recognise it as music.
    And how derivative? Depends on one’s perspective. If you’re not into classical opera, it all sounds pretty much the same. Same’s true of all genres.

  18. Steve;

    “Pretty soon, human voice actors will be an expensive, handcrafted luxury. Kevin Bacon is expensive, Computer generated Not Kevin Bacon works for free.”

    Free Kevin Bacon only shifts the issue – there would still be perceived value for those able to prove it was done by Not Free Kevin Bacon.

    In terms of approach – given the thread t’other day WRT Waters/Dark Side of the Moon, there’s always Stock, Aitken & Waterman with the Hit Factory, who basically figured out how to craft pop music that appealed to teenage girls, who formed the larger part of the persistent singles/’45 market. Cue lots of hits, but it’s a form of algorithmic approach. Involving SAW as creators, and Kylie, Jason et al as performers. So there’s a difference in how the rights are split out already.

    “These works could in theory be deemed free of copyright because they are not created by a human author. As such, they could be freely used and reused by anyone. That would be very bad news for the companies selling the works. Imagine you invest millions in a system that generates music for video games, only to find that the music is not protected by law and can be used without payment by anyone in the world.

    While it is difficult to ascertain the precise impact this would have on the creative economy, it may well have a chilling effect on investment in automated systems.”

    So, there seems to be a general approach, consistent with tax incidence, that only natural persons can hold copyright. There may be something, involving model rights and provenance, IANAL, whereby Not Free Kevin Bacon can assign rights to the ML generator, but’s he’s still going to get paid for doing so.

    Tim’s assumption that you can cancel copyright doesn’t hold, as the current state of play seems to indicate that ML systems would never be assigned it anyway.

    The other issue is that these systems are statistically based status quo generators. They’ll have a tendency to trend towards the mean, based upon the contents of the training set. Once ML output enters the training set, I reckon it’ll have a tendency to narrow the SD around the mean, implying that the things are increasingly easily gamed.

    Get the current Hype Cycle out of the way, there’ll be a set of functionality that is useful, and we’ll otherwise reset back to the default “in ten years time” whiuch is at least fifty years away. As usual.

  19. We have yet to encounter the one hundred dollar textbook author arguing that copyright should be abolished. Only those who have nothing to lose would argue against intellectual property.

Leave a Reply

Your email address will not be published. Required fields are marked *