Not long ago, I discovered the AI art generator, Midjourney, and wrote the following post:
Exploring this app got me wondering what the future of AI might hold for music. Luckily, I chanced upon was delivered via algorithm a YouTube video that wondered the same:
Along with this video, I read an article about the recent Beatles documentary, Get Back that talked about how they used machine learning to enhance and restore audio from the old 60's footage used in the doc. Peter Jackson noted in interview that they were able to teach the AI what Paul and Johns' voices sound like so that they could remix things more cleanly. It's pretty amazing stuff!
So, my questions to y'all are pretty open-ended:
- What do you think the future of AI holds for music?
- Have you come across other interesting resources talking about this subject?
- How far off is AI created music... and what does that look like anyway? Do you think AI will be composing completely fresh music or do you think AI will be copying the styles and voices of past musicians?
Top comments (5)
I don't think we're very far off, personally, but it's going to be a discussion of originality and quality as well. We do have some things that already exist for procedural music generation, and I suspect the same transformers we're using for text and image generation can be used to arrange previously known music arrangements together into new pieces, however these segments of reused music will likely be very recognizable and the result will sound derivative.
Imagine a new song that contains the opening chords of Thriller, the common refrain from Africa, and some very recognizable Bon Jovi. It's music, it's AI generated, but is it truly original?
These are just my suspicions as someone who is musically illiterate but a data science student and speaker in the community.
Appreciate ya weighing in here, Matt! I'm fairly musically literate, but lacking in the data science knowledge.
Current procedurally generated music aside, I think you're right that it's gonna be tough for AI to create something original if all the data is pulled from existing songs.
Funny thing though, musicians copy musicians all the time. Whole genres can come into existence by people imitating their musical heroes! However, inspiration and imitation sometimes leads to new breakthroughs and potential sub-genres — which I'm not so sure would happen with AI. Then you also have musicians who just experiment with new sounds constantly and occasionally discover stuff through accident (thinking about how Peter Gabriel is credited as discovering the gated reverb sound of the 80's). I'm not entirely sure if AI is going to create those same accidents... it may, but I think it'll take a human to recognize the accidents as something interesting and see the potential.
It's an interesting field. I think artistically, creatively, and even legally we're just beginning to discover our relationships with AI generated content, be it music, art, paragraphs, or code.
Check out Al Biles and GenJam: genjam.org/. Al Biles is a jazz trumpeter and recently retired Professor. His research was in computer jazz improv. It looks like he primarily plays now as part of a modern jazz quartet with people. But he was previously performing as "Al Biles and GenJam", essentially Al on trumpet with GenJam improvising other instruments as he played. He has some good talks about GenJam with demos, if he still has them available on his site. I believe he even auditioned with GenJam for America's Got Talent, but I think it was with the round they don't air with producers.
Oh wow, this def sounds interesting! I'm gonna dig into this one later on for sure. Appreciate ya sharing!