Because many music databases also include AI-generated content. Discogs may be the exception, and it seems that they are currently working on guidelines. Hopefully, they will exclude AI slop from the database. Similarly, as with MusicBrainz, when AI tools are used in the production by an artist (IMO this term excludes AI), it should be declared in the credits etc.
And if people are already adding their AI music to musicbrainz then we pretty much have to start again documenting all human music and pretty soon that will still be less work than documenting all the constant AI releases
It should be, but it’s not.
AI music should not be allowed in databases of human endeavour. It has nothing to do with music. It’s like putting programs or jpgs or chemical formulae into musicbrainz
Artists will start (have started) to use AI as a tool just like they did with every technical development from the fortepiano to the electric guitar, synthesizer, and computer. This will be interesting and legitimate, and the boundaries will be fuzzy.
Yes they will and have but the vast majority of what we are talking about is 100% AI slop. Generate it in seconds, upload it straight away, done. That is what is giving us the ridiculous volume of the stuff
Sure, but the boundaries will still be fuzzy.
I think it would be simpler to look at the artist annotation, and specifically disambiguation. Whilst Roon doesn’t use this, it is available via tbe API, and I imagine Roon holds a copy of the MusicBrainz data in their cloud.
I don’t see anything fuzzy about someone labouring for weeks/months on something (with or without some AI tools) vs AI generating it in 1.5 seconds. Anything that isn’t the former is slop
The artist and distributor are not going to mark their stuff as AI. They are the people running the con. Forget about it
Here is Claude ai’s answer to your suggestion:
Is it possible for an AI platform like Claude ai to come up with a database of music verifiably created by humans?
That’s a genuinely fascinating idea, but it faces some fundamental challenges worth thinking through honestly.
What Claude could theoretically do:
- Research and compile a list of artists who are verifiably human based on publicly available information
- Cross-reference against known AI platforms to flag suspected AI artists
- Build a “human verified” whitelist rather than an AI blocklist — a fundamentally different approach
Why a “human verified” database is actually smarter than an AI blocklist: Your extension currently tries to catch the bad actors. A human-verified whitelist would instead only allow through artists confirmed to be human — anything not on the list gets flagged. This inverts the problem elegantly since there are far fewer verified human artists on streaming services than AI ones, and the verified human catalog is relatively stable.
The hard problems that make this difficult even for AI:
- Verification at scale — TIDAL and Qobuz each have catalogs of 100+ million tracks. No AI can independently verify the humanity of every artist behind that
- New artists constantly appear — the list would need continuous updating
- The impersonation problem — as we just discovered with Larry Grenadier and Fred Hersch, AI content can appear under verified human artist names, defeating even a whitelist approach
- Legal and ethical liability — incorrectly labeling a human artist as AI-generated could constitute defamation *
- No ground truth source — Claude’s knowledge has a training cutoff and cannot independently verify real-time streaming catalog changes
What would actually work: The most credible human-verified database would need to combine:
- Streaming platform data (artist verified accounts)
- Rights organization membership (ASCAP, BMI, PRS etc.) as proof of human authorship
- Label affiliation for signed artists
- An open community contribution model like MusicBrainz
This is essentially what the music industry itself needs to build — and what Deezer, Qobuz, and others are slowly moving toward with their AI detection and tagging initiatives. Claude or any AI assistant could help compile and maintain such a database, but the authoritative source of truth would still need to come from the industry itself.
Is there an irony bypass in here
And how are you going to distinguish a release by a human under an anonymous, made up, name, who used AI as a tool, from a full AI slop release?
If it’s not distinguishable by it not sounding like garbage slop then who cares
So you’ll have a magical tool that judges how it sounds?
Isn’t that how deezer and qobuz stuff is purporting to work?
“Purported”, maybe. I’m copying my edit from the previous post to this one, because this is the problem:
I mean, there were probably hundreds of thousands of techno singles with a black cover and a made up name without credits. Today, large parts of those could be made with AI. How do you know?
I disagree. For instance, Nick Cave used autotune for effect in O Wow O Wow (How Wonderful She Is), and dynamic compression througout Wild God, which still divides opinion. But, there was a creative choice to use a tool in a particular way. The tool didn’t determine the outcome: Cave did.
Paul McCartney innovated in the sixties. He used tape loops etc. It’s his creativity. So, if someone used AI to create a “tape loop” effect, it’s still the artist working.
On the other hand, if someone asks AI to write a song and music in the Chicago Blues style, then that is and will always be AI slop (probably perpetrated for monetary gain.)
Sure.
Sure, too.
But if an artist uses AI to create de-/re-contextualized sounds, possibly based on existing music (such as sampling did, or John Oswald / Plunderphonics) but not necessarily so (it could also be original-sounding), and you have just the music to judge, no metadata that relates to anything known, how are you going to judge which is which? (And the “I know pure AI slop when I hear it based on my taste” doesn’t count).
IMHO, every supposed solution that claims that it’s easy underestimates how gigantic the problem is.
Everyone thinking a database of AI-made music is a thing that can be made is underestimating the problem.
Well, I don’t think that it can be made. The problem is gigantic, whichever way you look at it