On 09/05/2026 12:12, Kieren MacMillan wrote:
Hi Chris,
If using a tool to generate a score made it not copyrightable, wouldn't using
Lilypond to generate a score prevent copyright?
No more than using a hammer to build a house makes the house belong to someone
else. ;)
The problem isn’t “using a tool” — it’s using a library built almost entirely on other people’s uncredited and unpaid (and in most cases copyrighted) work.
There's a lot of those things around - they're called brains ...
LLMs are famously and notoriously trained using copyrighted works,
without the permission or remuneration required by copyright law.
What law? The EU certainly, and I think most other jurisdictions are
following them EXplicitly treat an AI just like a human - "if it's
freely available on the web, you may freely READ and LEARN from it".
Having an LLM generate actual content for you is then “fruit of the
poisoned tree”, since it’s essentially impossible to provably generate
something non-derivative [in the technical sense of that word].
Well, if that's the case, I would guess pretty much all *your* work is
also "fruit of the poisoned tree".
Indeed, ever since modern copyright law was introduced there was a
reasonably good case to argue that - if you are being pedantic - pretty
much ANY composer five or ten years into his career could not compose
any more without breaking copyright.
It's very hard to argue that your later work is NOT derivative of your
earlier work (and the works you learnt from, but that's irrelevant at
this point), and if you've been selling "works for hire" pretty much all
your later work infringes copyrights you've sold off earlier!
Hope that helps clear things up!
Kieren.
Clear as mud, thanks! :-)
Copyright law is - and has been from the beginning - a vague mess. Not
helped by all the arguments over censorship, free speech, and
"constitutional rights" - not mine for sure!
We all know the Gilbert and Sullivan operetta "The Pirates of New York",
don't we :-)
At the end of the day, the middle of the road attitude is to treat an AI
like a human - the law EXPLICITLY (where I am at least) says it is
perfectly okay for an AI to read any and all works legitimately acquired
(such as downloaded from the open web).
It says absolutely nothing about whether the work produced by an AI is
copyrightable, copyrighted, or in breach of copyright, just like work
produced by any human!
About the only thing (American case law iirc) the law says about that is
"An AI cannot hold copyright". Any copyright in the original text fed in
(training material, prompts, whatever), MAY or MAY NOT leech through
into the final output work. Exactly the same as with training works,
prompts etc fed through a human brain. The only difference is the human
brain may add its own copyright to the mix.
Oh - and as for running your own models, if you're prepared to spend
£500 or so on a graphics card, I think you can get Open Source models
you could feed the lilypond documentation into, and get pretty decent
results out of. All running locally on your computer with nothing being
fed to/from the internet.
(NB - AIs scraping the web, I'm well aware that many of them are
extremely badly behaved citizens, but that's nothing to with copyright.
For those who remember Prince Philip, I'd say it is criminal, but it's a
"stealing electricity" problem - their actions are obnoxious and cause
grief/expense/whatever for other people ...)
Cheers,
Wol
