I used AI to make this song. The results, and its implications, startled me

San Francisco Chronicle​

14 hours ago

Earlier this month, a band called the Velvet Sundown surpassed 1 million monthly listeners on Spotify. Its profile features two albums, some gauzy cover art and a genre-appropriate backstory. The music was credible, the branding effective and the effect convincing.
Only afterward did we learn the truth: There was no band. Every note, lyric, image and faux-autobiographical gesture was the work of generative artificial intelligence.
The project was presented as human, then rebranded as 'synthetic music guided by human creative direction,' a soft euphemism for automation. Whatever creative intentions, aesthetic sensibilities or prompting skills that may have guided the project, the resulting songs still arose from full substitution because no embodied musical practice took place. No performance, no skill-based interaction with the medium, no causal chain running through human hands in the act of making.
We are in the process of crossing a threshold. AI systems trained on decades of creative labor now can plausibly simulate artistry, threatening to render the human creator redundant. Skills that once required years of practice are instantly mimicked by code. Musicians now face an existential question: Has their craft, refined through long hours of practice, collaboration and performance, already become obsolete?
Listeners also face a paradox. A song moves you, but its 'singer' cannot feel. What, then, has moved you? The structure, the timbre, the rhetorical form? Or the knowledge that somewhere, behind the notes, another person is manipulating generative AI with whom you might be joined in experience?
There is historical precedent for this.
In the industrial era, machines hollowed out skilled labor. Weaving, typesetting and machining were all transformed into logistical problems that were, by and large, solved through automation. What disappeared was not only employment but a relationship between work and meaning. The rhythms of human labor were replaced by the monotonous hum of mechanical replication.
Today's creative automation occurs on cloud servers. The new assembly line runs through graphics processing units and datasets on which a million songs become the raw template material for one more product.
I've experienced this firsthand. Last year, I subscribed to Suno, an AI platform that generates music from text and audio prompts. Its marketing promises 'a future where anyone can make great music,' reframing skilled composition as a technical barrier waiting to be removed. My early experiments were forgettable. A prompt request for a Weimar-style cabaret polka returned a tune that sounded less like 'The Threepenny Opera' (1928), and more like elevator music for a mid-tier chain hotel in Leipzig, circa 2004. A Krautrock-inspired instrumental was more on point, but still nothing to write home about. Over time, however, Suno's responses improved dramatically.
Then, in July, I decided to write a song combining the old way with the new generative tools. I uploaded lyric ideas (finished by ChatGPT) and a low-fi guitar recording from my phone to Suno. After several cycles, what came back startled me. The track, titled 'Ashes on the Heath,' didn't just resemble my idea; it felt like it was the idea, polished. The vocal delivery seemed emotionally real. It almost felt like a collaboration. [Listen to the song below.]
Of course, the emotion wasn't real. Generative AI doesn't really collaborate. It doesn't know you or have feelings about strum patterns or drum fills. Most importantly, it doesn't make mistakes, which are integral to the process of real music. AI replicates genre conventions and timbral signatures. It does not live through them. It reproduces the statistical contours of prior performances while living nowhere inside them. It can only reconfigure forms once used to produce meaning.
Generative AI is not magic; it's logistics. Platforms like Suno do not understand or feel music. They approximate its grammar. They rely on vast archives of recorded music, often pulled without permission or compensation. AI's 'cheap' outputs aren't cheap because the cost of making art has vanished. They're cheap because the cost for training is being invisibly offloaded onto past labor that was never compensated. They don't feel or resolve tension. They don't second-guess themselves. They produce form without the living content that once made form matter.
Other audio tools based on machine learning — e.g., Izotope's noise reduction plug-in and Logic Pro's stem separation — have already become normalized in music production. But generative AI crosses a different line. It doesn't just assist the process. It threatens to supersede it entirely. The result could be substitution rather than enhancement.
While some might argue that generative AI democratizes creativity, this framing obscures a deeper reality. These systems make possible the instantaneous imitation of creativity. Authorship too easily collapses into prompt-entry. It is more akin to configuration from a catalog. Yes, AI can and will enhance creative projects, making new outcomes possible. No, it will not mean the same thing if no human struggle threads it together.
When I initially shared my AI-assisted track with a few friends, I withheld the fact that the vocals had been created by a machine. The reception was warm. Some listeners were moved to tears. Then I shared the track with some other friends, disclosing the AI vocals upfront. The reception was quite different.
'There's no tell in the vocal that it's a heartless construct sent by fascitarian tech overlords to extract phosphorus from the working class,' one wrote. 'So: 1) good job, and 2) ohmygodwhatishappening?'
Another said simply, 'Nice track — but I draw a hard line on AI for music creation, and I'm sure you know why.'
I do know why. Because music isn't just sound. It's a relationship based on tension, memory and embodiment. It's the moment in the take that no one planned, but everyone went with, or the phrasing that makes the line land.
I don't think AI will erase human music. Music is social; it comes from bodies in time. But the economic logic is changing. A tireless synthetic co-worker, charging a modest monthly subscription fee and trained on the unpaid labor of artists whose livelihoods it now threatens — this is all new.
The point is not to romanticize a pre-generative-AI past. It is to understand that the problem is not the tool but the system that deploys it to cheapen labor and consolidate cultural production on private platforms. The question is not only, 'Can AI make art?' but also, 'Who controls what counts as art and under what terms?'
Platforms like Suno are not only generating songs. They are scripting the future conditions under which music will be made, circulated and valued.
To what extent these systems serve as creative tools or substitutes for creativity will decide whether culture remains an open field of human meaning or a closed loop of recombinable parts. That decision is, in principle, ours to make. But the conditions for exercising that agency are eroding quickly.