The AI model and a flesh and blood artist both learns by listening to music. To call one “unethical” and the other not is hypocritical.
But yes, it’s possible to train gen AI on royalty-free, public domain et.c. data. For instance Adobe have trained their image generator on their own stock photo library.
Your comment seems confused to me. It’s not the “listening” part that is unethical, but the commercial use of creative works without the payment of royalties.
I won’t let the people making billions from these tools abdicate responsibility for them.
Even if you somehow managed to create a training corpus of the size required these days composed entirely of data for which you have explicit consent, the fundamental value proposition of these tools is still to destroy the livelyhood of artists/writers/support personnel/etc. If you support their use, you are personally contributing to this, and I don’t think I need to detail what I think of people who do this.
Oh, sorry. You may not know how gen AI actually works. It first “listens” to music to be able to create music. As what some people think is unethical is that it listens to music to learn how to create music. The process is the same as with flesh and blood artists. To make a difference between if a person or a software is listening to the music and only call one “unethical” is hypocritical.
Do you hate other tools that help people create music or is just a specific? Do you also hate MIDI instruments, VSTs, step sequencers, arpeggiators, autotune and so on.
And you can get open source generators that run offline. GitHub - multimodal-art-projection/YuE: YuE: Open Full-song Music Generation Foundation Model, something similar to Suno.ai but open · GitHub
That is just a side effect. If you use an orchestra VST instead of hiring a symphony orchestra you also put those out of work.
As I understand it, the digital information in a very large number of music files is being converted to a suitable format, such as audio spectrograms, and then fed to an algorithm that conditions the parameters in a mathematical deep learning model.
Haha, no it isn´t. If this is the way you “listen” to your music, please upload a photo. 
The sheer volume of data, both for training and output by the generative model, and the enormous revenue generated by large commercial parties without any financial remuneration of the thousands of artists that contributed to the vast database required to train the model makes this a completely different case. You may not agree, but there are valid legal and ethical objections to these practices.
Maybe stop moving the fucking goal posts. None of those things are remotely similar and you know it.
I don’t think it is the role of Musicbrainz to police the music here, just like we also store other unethical music like blatant plagiarism, but we don’t need to bend over backwards and lick the boots of the billionaires pushing the “AI” shit.
The existance of the music on the database is not an endorsement of it. There is other music that speaks about violence or other law breaking or unethical behavior. Giving it a rating or reviewing it negatively fits other sites better. There are few ratings on here.
The authors of AI generated music or their fans would have to do the work of adding it, and they likely can’t be bothered to. Few examples of AI music get reposted on other platforms.
I agree. We should list the music regardless of what tools was made to use it.
The problem I was raising, besides ethical, was that it takes (wastes) more time to add, edit and maintain in MB than it took to generate in mass.
Well,t he ethical argument turned out to be bogus, so that leaves only the volume of tracks and releases. That is of course a problem that have nothing to do with the tools used to create the music. For instance Buckethead released 118 albums in 2015. Viper released over 300 albums in 214. I think the problem will solve itself. If nobody is interested in the music they will not get added to MB. It is a choice we make as editors to add or not add a release.
This seems to be going down into a free for all of people attacking the tech (which is mostly fine) and each other (which is not ok). Since I see no actual useful discussion happening here for a while, I’m going to close the thread.
As a summary, and I put on my robe and style leader / community manager hat:
Whatever we all think of AI use for music creation (I’m not generally a fan and have ethical issues with some of it, clearly some others here find it not a problem at all, while others are a lot more opposed to it still than I am), I think there is no real doubt that data about it belongs in MusicBrainz. So nobody should feel they should not enter some specific music they like or find interesting just because AI was used for it, in the same way we don’t block adding data about Russian/Israeli/American music, or extremist, racist, radical, or intentionally offensive music, or any other sort of music that might be controversial or ethically problematic in some way for some of us.
On the other hand, I think tagging AI music as such cannot hurt and should be encouraged. For people who are in favor of AI music, it is no different than a “piano” tag on a piano recording. For those who want to avoid it, it’s a marker to help them do so. As such, it seems like it has no clear downsides. I’d prefer if people would tag it neutrally and then choose what to do with that information rather than using deprecating tags, if only because if there’s a tag people with different opinions on AI music can agree on, it seems to actually help both sides achieve the preferred situation of knowing whether the music is AI.
That said, my personal opinion is that we should encourage people to not add AI music to the database unless they actually find it interesting or enjoyable or care about that specific music in some other way, if only because so much of it is being made that it would be a ridiculous amount of work to document it otherwise. That is not restricted to AI music specifically though; given the amount of music that was being uploaded to the biggest platforms every day even before AI music was a big issue, adding as much of it as possible just to document all the things would be a mess even avoiding AI altogether. AI music just makes it even more so.
A different discussion, that could be had in a hopefully more constructive way, is how to credit AI music and what relationships to use for it, which we will eventually probably need a guideline for. If someone has good ideas, and can discuss it in a calm way, feel free to open a separate, more concise thread about it.
And a final reminder:
