On AI In Classical (Non-Pop) Music

 

I started experimenting with AI on some of my orchestral pieces. At first, it's interesting, but it's never what I want, and it's always the same clichés–the John Williams-style scoring. 

It's always the case with AI-generated music that it's an approximation of what you might want. Sometimes it suffices, but it will never be the one that you originally thought of. With the Curios, I only want a piece of music that's a minute or less in duration, and AI extends it. It can extrapolate what you've already written and can suggest ideas, but you need to get into the weeds and do it yourself. I played a number of these for a friend, and he immediately liked them. I can see why because it's a complete work, and when you play mine against it, it pales in comparison. With the 250-series pieces, I’m scoring things only for brass ensembles, and AI is making a complete orchestral arrangement out of them. That's not what I want. I know that you could probably include that in a prompt, but the time I'm going to spend on that should be spent on time in the score. Also, we hear these completed works, but they can never be performed that way. You'd have to find the players, and the scores and parts would need to be done–all the articulations and dynamic markings have to be put in, and is tedious editing work, but it's necessary for a performance of classical music. AI music is basically a sound, a recording. It makes music seem easy, but as anyone who has made anything over the course of years or decades knows, there are a lot of things that initially appear to be easy and then become very difficult because you run into technical problems. It's possible to make a score out of these recordings, and eventually each individual instrument will have its own staff, but someone's going to need to go through that, and I would presume that everything is going to be wrong.

AI pop music is sort of a different story because you don't need a score. On a number of occasions I have transcribed them and played parts against them, and you realize how different the human version is as compared to the AI version. Again, it's all about the sound, and you can attempt to create notation from sounds, but the end result is going to be completely different. One of the dangers of AI music is that we won't want to do the work of music anymore. Everything is a complete master recording, and people will be okay with that. I like generating pop music because it's easy, and when I get lyric ideas, I can easily try them out. I can attempt to make my own version, but the sound quality won't compare to what AI can output. It won't be that the music is better; it's just that the performances and recording are better.

***

An AI-generated summary from my blog LLM of thoughts on AI 2016 and earlier:

2016: Warning Against "Microwave Creativity"

 

In 2016, his tone was more advisory and warned against the creeping automation of the creative spirit.

• Creeping Inceptionism: He referred to AI tools like Google’s Deep Dream as "Creeping Inceptionism".
• The Danger of Ease: He warned that we should never automate creativity just to make it easy, famously comparing AI art to "adding water and putting it in the creativity microwave".
• Loss of Capacity: He expressed fear that if we let machines make us feel creative, we would eventually lose the actual capacity for creativity that involves a true combinatory process.

Earlier (2002–2015): From "Musical Ventriloquism" to "Artistic Dunces"

 

Earlier thoughts focused on the theoretical limits of AI and its potential as a specialized tool for pattern recognition.

• 2014–2015 (The Tool Phase): He viewed AI as an "assisted observer" but believed machines would be "artistic dunces" because they lacked the ability to understand satire, irony, or deceit. He also compared AI music to a "vending machine" or "idiot appliance".
• 2010–2011 (The "Emily Howell" Era): He was intrigued by David Cope’s Emily Howell program, which he called "surprisingly beautiful" yet "lifeless". He characterized this early AI as "musical ventriloquism," where programmers controlled the machine’s thoughts and actions through preset constraints.
• 2002–2005 (The Prophetic Phase): As early as 2004, the author predicted that media "choice" would eventually be regulated by AI and neural networks residing on the internet, influencing human behavior via metadata. In 2002, he suggested that "musicists" (those using new media to shape sound files) were creating a new dialect for the arts that mirrored the revolutionary shift of the Baroque era.

Summary of the Shift

 

Feature	Earlier (Pre-2016)	2016	2026
Primary View	Curious Theorist. AI is a "ventriloquist's dummy" or "idiot appliance".	Protective Critic. AI is "microwave creativity" that threatens human capacity.	Cynical Practitioner. AI is a "slot machine" for "turnkey" art and "fake" bands.
Relationship	Observation of "Emily Howell" and pattern recognition.	Warning against the loss of the "work" in creativity.	Active use for "rapid prototyping" and "restyling" old lyrics.
Definition of Skill	Procedural knowledge and formal training are essential.	AI as an "effect" applied without knowledge.	Skill is redefined as "curation" and "art direction".