Kamal Ghorui
A few months ago, a musician reached out to tell me what our software had done for him. After a stroke, he couldn’t remember the notes to the songs he had spent decades composing. The melodies were gone from his memory, and the recordings sat on his hard drive like locked boxes. Our AI tool turned those recordings into sheet music. For the first time in years, he could see and play his own music again.
Recently, a 94-year-old composer with a doctorate emailed me. He had improvised more than 280 songs since 2000 and spent decades heading a university music department. Now he works against the clock. He records his improvisations and runs them through our tool so he can ensure his musical legacy will survive him. “Time is running short,” he wrote, “to preserve my songs for posterity.”
That story has stayed with me because it has almost nothing to do with the AI debate consuming the music industry right now. That debate is real and urgent. Generative AI tools can now produce a fully produced pop song from a text prompt in seconds. Streaming platforms are being flooded with AI-generated tracks. Artists whose voices and styles were used to train these models have received next to nothing in return. The legal and economic questions are serious, and they deserve serious answers.
But in the urgency of that fight, we are making a mistake. We are treating all AI as if it were the same thing. It isn’t. And the difference matters enormously for the future of music.
The AI dominating headlines is generative AI (systems that create entirely new music from scratch). A user types a prompt, and a track appears. Companies like Suno and Udio are advancing these tools rapidly, while technology giants like Google are releasing increasingly powerful music-generation models. These generative systems raise hard questions about copyright, compensation, and what happens to working musicians when the cost of producing a song approaches zero.
But there is a second category of AI that has received almost no attention, and it works in the opposite direction.
Call it assistive AI. Rather than generating new music, assistive AI helps people understand, access, and work with music that already exists. It is the difference between a machine that writes a novel and software that teaches you to read.
I run a music education technology company (Songscription) that builds this kind of tool: AI that converts recordings into playable sheet music. When I was in high school, playing baritone saxophone in the pit orchestra for the school musical, I did this by hand. I would sit with the score before rehearsal, painstakingly transposing parts into my notebook. For the jazz band, I’d rewind the same five seconds of a famous improv again and again, trying to identify each note by ear. It could take hours to transcribe a single line.
Today, because of developments in assistive AI, the same task takes seconds. That compression of effort is not trivial. It is the difference between a student who keeps going and one who gives up.
The stories from our users reflect that. One composer wrote that it was the first time she had ever been able to create sheet music for her own composition: she had the music in her head, could play it by ear, but had never had access to notation tools that worked for her. Another message simply said: “I never thought this would be notated. Thanks for fulfilling a dream.”
None of these people was replaced by a machine. They were handed a key back into their music.
This distinction matters because the policy responses to each kind of AI should be completely different. Generative AI may require guardrails: clearer rules around training data, meaningful compensation for artists whose work was used to build these systems, and an honest reckoning with what happens to musicians when content becomes nearly free to produce.
Assistive AI deserves something else entirely, I’d argue: encouragement.
The stakes are real. Many households include someone who plays an instrument, yet millions who want to learn never do. Research shows the barriers are access and friction, not talent. When friction is removed, more people learn. When more people learn, communities gain something that cannot be streamed or generated: live music, made together in a room.
I believe that the most exciting future for music is not machines that write songs for us; it is that technology that helps millions more people learn to play them, and in doing so, fills more stages, more living rooms, and more lives with something no algorithm can replicate.
The musician who recovered his songs after his stroke and the 94-year-old composer didn’t need a machine to create music for them. They needed one that could help them find their way back to their own. That may be the most promising future for AI in music: not replacing musicians but helping millions more people become one.
END OF ARTICLE
