AI in music

[Cato]

56 minutes ago, Al Krow said:

 

There's got to be an opportunity for a streaming platform to be set up that takes only non ai content, and requires some form of validation from uploaders.

 

So far Deezer is the only one I'm aware of that's trying it.

 

Nothing I've read over the years about Spotify or it's owner makes me think he cares about much about the content on his platform other than funneling as much money as possible from the music industry into his own pockets.

 

Unless his revenues start dropping off because people are boycotting Spotify in protest over unlabelled AI I can't see them taking any action.

 

 

Edited Saturday at 13:06 by Cato

[TimR]

15 hours ago, 80Hz said:

 

Well sure, but I think that still involves some critical thinking/reverse engineering, no? And I would think what you're talking about is still beyond most people.

 

Example: I've tried to get an LLM to spit out some code for an audio thing I wanted to do, and no, it didn't work, and furthermore, no, I didn't know how to fix it 😂

 

A lot of kids are being taught programming by assembling pre built units, that do things, together. 

 

We should remember that a lot of programming we do, if we are using a language is also assembling things that other people have already written for us. No one writes in binary or assembler. 

 

Very few of us have built our own amplifiers, leads, guitars. We all rely on the donkey work being already done for us.

 

It just depends on what level of individual human input you're comfortable with. 

 

Seems most of us draw the line at the actual performance.

 

There's some very interesting music created from received telescopic data by NASA's Sonification project. 

 

 

 

 

 

I think all AI produce should be labelled as such. 

 

 

 

Edited Saturday at 13:21 by TimR

[Al Krow]

16 minutes ago, Cato said:

So far Deezer is the only one I'm aware of that's trying it.

Nothing I've read over the years about Spotify or it's owner makes me think he cares about much about the content on his platform other than funneling as much money as possible from the music industry into his own pockets.

Unless his revenues start dropping off because people are boycotting Spotify in protest over unlabelled AI I can't see them taking any action.

 

I did a quick search and this is what Google's AI assistant, lol, came up with:

 

Top Music Streaming Services Avoiding/Limiting AI

• Bandcamp (Most Proactive): Bandcamp is considered the strictest, having formally stated that music or audio generated wholly or substantially by AI is not allowed. They reserve the right to remove AI-generated content, focusing on protecting a space built on human, independent artists.
• Deezer (Best Detection/Labelling): Deezer has taken a leading role by actively identifying AI-generated tracks, aiming to label them and remove them from algorithmic recommendations. As of 2025, they have reported that 85% of AI-generated streams are demonetized on their platform.
• Qobuz (High-Fidelity Human Curation): Known for its focus on high-fidelity audio, Qobuz tends to feature,, heavily curated content, with playlists created by humans rather than algorithms, making it less likely to be flooded with AI-generated, "lo-fi" style content.
• Coda Music: A newer platform that is building tools to label and allow users to actively block or turn off AI artists, promising a more human-centered experience.
• Other Platforms: SoundCloud has stated it does not use artist uploads to train AI and has introduced a "no AI" tag to block unauthorized use. 

Music Platforms to Exercise Caution

• Spotify: While not banning AI, Spotify has introduced measures to target "spammy" AI uploads and vocal deepfakes, rather than AI as a creative tool. However, it remains a primary source of AI-generated content, particularly in personalized playlists.
• Apple Music: Focuses on quality control and proper metadata, meaning they may reject AI content that lacks clear authorship or is deemed low-effort. 

[TimR]

I can't imagine anyone employing people to sit a desk all day, looking at the music you like and then working out what other music you would like. For a start, no human would have that massive encyclopedic knowledge of songs and genres, including album tracks etc.

 

Certain jobs/processes are only possible and have been created because of AI. I suspect this will be the better use. 

 

Self driving cars, where one accident can be analysed and then the scenario be exported to all the other self driving cars so they don't make the same mistake, unlike human drivers who all seem to make the same mistakes over and over again and never learn from either themselves or others. 

[Woodinblack]

4 hours ago, TimR said:

I can't imagine anyone employing people to sit a desk all day, looking at the music you like and then working out what other music you would like. For a start, no human would have that massive encyclopedic knowledge of songs and genres, including album tracks etc.

 

Certain jobs/processes are only possible and have been created because of AI. I suspect this will be the better use. 

 

That really is the weakest definition of AI - ie, its AI in the same way as a spreadsheet is AI. But that is the trouble now, marketing is putting everything done on a computer as AI, as by some definition it is.

 

I would just call that a database. Or probably more normally Big Data.

[TimR]

1 hour ago, Woodinblack said:

 

That really is the weakest definition of AI - ie, its AI in the same way as a spreadsheet is AI. But that is the trouble now, marketing is putting everything done on a computer as AI, as by some definition it is.

 

I would just call that a database. Or probably more normally Big Data.

 

A lot of what we think of as human intelligence is just accessing a database. Very little of what we do requires actual intelligence*. 

 

*Luckily, or most people wouldn't survive longer than a day 

[Woodinblack]

3 hours ago, TimR said:

A lot of what we think of as human intelligence is just accessing a database. Very little of what we do requires actual intelligence*. 

 

Absolutely none of it is just accessing a database. Computers are very good at accessing data, humans are absolutely terrible at it, there is always some kind of interpretation of that data and revision based on unrelated items, rather than literally recalling the data, ie storage.

 

The 'people who liked this also liked...' problem is a big data lookup issue. You have a database of 100 million peoples listening preferences and what they like. User A liked this song by this artist? Well, we have another 5 million that liked that, and of those, 1 million also liked this song, 900,000 liked this one, 800,000 liked this one - thats your list. A straight database search is really easy, and cheap, a neural net lookup requires a ton of more power. 

 

 

[TimR]

41 minutes ago, Woodinblack said:

 

Absolutely none of it is just accessing a database. Computers are very good at accessing data, humans are absolutely terrible at it, there is always some kind of interpretation of that data and revision based on unrelated items, rather than literally recalling the data, ie storage.

 

The 'people who liked this also liked...' problem is a big data lookup issue. You have a database of 100 million peoples listening preferences and what they like. User A liked this song by this artist? Well, we have another 5 million that liked that, and of those, 1 million also liked this song, 900,000 liked this one, 800,000 liked this one - thats your list. A straight database search is really easy, and cheap, a neural net lookup requires a ton of more power. 

 

 

 

I would say the first three hours of my weekday, every day, is exactly the same and can be done on autopilot. I suggest most people are the same. 

[TimR]

Conversely, I've just had an hour long meeting with a couple of people I don't know very well on a personal level.

 

We discussed something very important to all of us and required every ounce of my emotional intelligence to learn and pick up on their micro body language and inferred language. 

 

I'm ready to go back to bed. I certainly couldn't do that regularly for very long.

 

When machines have learned to do that, we are in trouble. Until machines learn to do that, they are not going to be much of a problem to us. 

[SumOne]

I assume some of the backlash against Punk was similar. And probably Rock n Roll before that, and Blues before that. Music isn't all about technical skills, it's about expression - if AI can help people express themselves then it's it always a bad thing? I suppose there is a fine line regarding artistic input.

 

People that studied music for years and struggled to make a living in an orchestra probably looked at punk musicians with the sort of distain people look at people using AI for helping write music now. I suppose a key thing with Punk was the democratisation of expressing yourself through music - whether or not you were lucky enough to have been able to go to a music conservatory or have the time and money to practice. AI supporting artists possibly does similar.

 

Nearly 50 years after punk and we still have music conservatories and orchestras of highly trained musicians and people that appreciate hearing that, and a lot of musicians (e.g. rappers, DJs, pop singers)  that could barely tell you the difference between a major and minor chord. There is room for both.

[Woodinblack]

33 minutes ago, SumOne said:

Nearly 50 years after punk and we still have music conservatories and orchestras of highly trained musicians and people that appreciate hearing that

 

Hmm.. not sure of the relevance as punk didn't try to do what the conservatories were doing, but cheaper, it produced a new form of music. whereas the AI thing is to produce the music without needing the people of the punk and the conservatories at least on recorded materials. 

 

[SumOne]

5 minutes ago, Woodinblack said:

 

Hmm.. not sure of the relevance as punk didn't try to do what the conservatories were doing, but cheaper, it produced a new form of music. whereas the AI thing is to produce the music without needing the people of the punk and the conservatories at least on recorded materials. 

 

 

I think it's just a different flavour of that, or it can be.

 

People said similar with sampling, synths (memories of my Dad's distain at 80's '1 finger keyboard players'), drum machines, DJs, Laptop music etc. 

 

Granted, they did all need people to a certain extent (but lazy people could just use drum machine presets and a synth arp preset), but it depends how people use AI to how much input they have - could be anything from just mastering a track all the way up to making it all via a prompt (and isn't that a bit like the non-musician producers that just tell musicians how to play?).

 

 

[BigRedX]

When AI creates a form of music that is genuinely new, that's when we need to start worrying.