Music and AI - past, present & future
AI-generated music is bound to stay and be interesting
I listened this morning to A Junior, A Senior And I ( AJASAI for the fans ) and it was an episode on music with two guests, Stefano and Aslahan.
After 5 minutes, both guests admit that they don't know much about AI generation, they had still interesting examples and stands, mainly as musicians.
Yet, there were things where both the host and the guest didn't dig into and I thought I'd write about my answers to this on-going question of music and AI.
First of all, it's funny how in the music world, the name Ray Kurzweil does not mean much and it is a big name in the AI world, it might be an anecdotal proof that both worlds do not often talk to each other. Ray Kurzweil is one of the godfathers of AI as he predicted AGI for 40 years now - he is still working at Google around that theme - and Ray made a lot of his money making music, because algorithms for sounds are so complex and require high math aptitudes and data and data transformation. Things essential to build AI systems.
I really appreciated how the musical guests were sincere with the use of AI as a tool - they mentioned how stem selections on a track got really better - and Aslahan insistence that the drive to use AI was mostly businesses decisions. The real proof, not mentioned is that it's before the LLM craze even that labels have been using big data and smart algorithms to do song selections on albums to find the potential hits they ought to back.
It was on AI-generated music that I thought the discussion could touch more grey areas. Stefano and Aslahan both went against AI generated music.
While being an artist's partner, I do understand the value of human-made art, I never really rejected digital art, and AI generated art is digital art.
It is true that using simply an app used by someone else is slightly lazy - it doesn't mean that it has no value.
Somehow, it kinda makes me think of photography - when there was a debate on whether it would replace paintings. And what we saw is that it didn't but it did influence it and is now an art of its own. Photography, to be faire, did have an impact on the business of portraits, but there is still art in portraits ( the British national portrait gallery is amazing if you get a chance ).
How could AI-generated music be art ?
First of all, generation algorithm can be tuned to specifics, through training and post-training.
And the anecdotal simile here is dj-ing & turntablist. There were huge push-backs in the 70s and 80s against them -- and almost with the same criticism that we could hear on the podcast now. "They are stealing music" - "They haven't built the experience musicians/experts have".
Musicians though can't train algorithms and/or tune music, and that's where there might be some distinctive artists coming out of it.
Then, there was the criticism of AI-generated music not being a team effort and not creating an aura with fans. I think it might be cultural blindness, but I can see AI being teams creating digital avatars, like we have in Japan and Korea, that fans do take seriously. And it does take a whole team to run an avatar-based music-dom. Yes it's not exactly all the same skills as solely-man-made music, but it is teams creating a thing that has an aura like musicians. edit: I received messages telling me I was also culturally blind by forgetting Gorillaz and Deadmau5.
Finally, beyond arts, one of the thoughts that I think is important to re-iterate is that intelligence is an emergent capacity of LLMs that was not expected at this level. LLMs do not solely "copy-paste", as the guests hinted at, but do create through probabilistic creation new things. They are also capable of some reasoning. And that's for language base large models. And as Anthropic said, every model trained has its own personality, strength and weaknesses, uniqueness - depending on the training and on the data.
After 5 minutes, both guests admit that they don't know much about AI generation, they had still interesting examples and stands, mainly as musicians.
Yet, there were things where both the host and the guest didn't dig into and I thought I'd write about my answers to this on-going question of music and AI.
First of all, it's funny how in the music world, the name Ray Kurzweil does not mean much and it is a big name in the AI world, it might be an anecdotal proof that both worlds do not often talk to each other. Ray Kurzweil is one of the godfathers of AI as he predicted AGI for 40 years now - he is still working at Google around that theme - and Ray made a lot of his money making music, because algorithms for sounds are so complex and require high math aptitudes and data and data transformation. Things essential to build AI systems.
I really appreciated how the musical guests were sincere with the use of AI as a tool - they mentioned how stem selections on a track got really better - and Aslahan insistence that the drive to use AI was mostly businesses decisions. The real proof, not mentioned is that it's before the LLM craze even that labels have been using big data and smart algorithms to do song selections on albums to find the potential hits they ought to back.
It was on AI-generated music that I thought the discussion could touch more grey areas. Stefano and Aslahan both went against AI generated music.
While being an artist's partner, I do understand the value of human-made art, I never really rejected digital art, and AI generated art is digital art.
It is true that using simply an app used by someone else is slightly lazy - it doesn't mean that it has no value.
Somehow, it kinda makes me think of photography - when there was a debate on whether it would replace paintings. And what we saw is that it didn't but it did influence it and is now an art of its own. Photography, to be faire, did have an impact on the business of portraits, but there is still art in portraits ( the British national portrait gallery is amazing if you get a chance ).
How could AI-generated music be art ?
First of all, generation algorithm can be tuned to specifics, through training and post-training.
And the anecdotal simile here is dj-ing & turntablist. There were huge push-backs in the 70s and 80s against them -- and almost with the same criticism that we could hear on the podcast now. "They are stealing music" - "They haven't built the experience musicians/experts have".
Musicians though can't train algorithms and/or tune music, and that's where there might be some distinctive artists coming out of it.
Then, there was the criticism of AI-generated music not being a team effort and not creating an aura with fans. I think it might be cultural blindness, but I can see AI being teams creating digital avatars, like we have in Japan and Korea, that fans do take seriously. And it does take a whole team to run an avatar-based music-dom. Yes it's not exactly all the same skills as solely-man-made music, but it is teams creating a thing that has an aura like musicians. edit: I received messages telling me I was also culturally blind by forgetting Gorillaz and Deadmau5.
Finally, beyond arts, one of the thoughts that I think is important to re-iterate is that intelligence is an emergent capacity of LLMs that was not expected at this level. LLMs do not solely "copy-paste", as the guests hinted at, but do create through probabilistic creation new things. They are also capable of some reasoning. And that's for language base large models. And as Anthropic said, every model trained has its own personality, strength and weaknesses, uniqueness - depending on the training and on the data.
We are only at the beginning of sound generation, and so yes, it does sound all a bit basic/dumb, like text generated by LLMs were 4 years ago. But sound generation, because of how deep/wide the information is, is more complicated than text generation, and we don't know what intelligence or other side-effects will emerge from truly LLM-like models based on it.
For the future of it, I think that it is bound to stay, and there won't be a clear link between ai-generated and man-made. It's bound to be like dj-ing and photography, where people mix genres and styles. And it's ok for some artists to reject it, a lot of artists, in every domain, are purists, yet reality has brought a full colourful canvas of variety to their domain.