6 min readNew DelhiAug 11, 2026 03:48 PM IST
First published on: Aug 11, 2026 at 03:48 PM IST
By Charu Pawar, Vagmi Joshi and Arghya Mondal
There is a particular kind of tune that has been colonising Instagram this month. Within days of the first episode of Samay Raina’s India’s Got Latent airing, “Oye Hoye Kya Scene Hai” had exploded to half a billion views and spawned millions of reels. And yet almost no one could pretend the lyrics made any real sense, or that the voice singing them didn’t sit somewhere in the uncanny valley, almost human but not quite.
For the unversed, that’s because it was an AI-generated song, which means someone manufactured half a billion views’ worth of virality on a lazy afternoon, simply by typing a few words into a prompt box and waiting a few minutes for the machine to hand back a song.
So, who actually made this song?
Every catchy AI song is, in some sense, built out of the bones of songs that came before it. This is because an AI music model has been trained on an almost unimaginable quantity of existing music until it has absorbed enough of the pattern to produce something new that still sounds hauntingly familiar.
So if no human being actually composed the song, who owns it? Under Indian copyright law, ownership requires a human hand. So a fully AI-generated song raises a difficult question about who, if anyone, can claim ownership of it.
Neither the platform that generated it, nor necessarily the person who typed the prompt, nor the thousands of uncredited musicians whose original work may have quietly trained the model in the first place.
However, this absence of ownership is not neutral, as it does not protect the people whose musical fingerprints are buried in the model’s training data. The ambiguity covertly works in favour of whoever has the resources to run the model at scale, and just as silently against whoever contributed the labour that made the model possible to begin with.
The prompt box promised everyone a music career
Making a song used to demand years of training, but now, anyone with a phone can produce something that sounds worthy of virality in under a minute. This, of course, comes with ethical questions like whether a skill this layered can simply be replaced with a prompt box, or whether what’s being handed over is just the appearance of the skill, stripped of everything that made it meaningful in the first place.
Even if the model can technically generate a song, recommendation algorithms on Instagram or Spotify still get to decide what gets pushed to the top of the feed. This is where the work of anthropologist Nick Seaver becomes useful, as he describes the real goal of recommendation algorithms not as exposing listeners to better or a wider range of music, but as “capturing” them, i.e. keeping them engaged for as long as possible.
The algorithm is simply trying to hold your attention, the same way a hunter’s trap doesn’t care about the animal’s well-being, only about whether it stays caught.
AI may well be democratising the technical skill of making a song. But it isn’t democratising taste, and it isn’t democratising whose music actually gets heard. It is simply automating the same mechanism that was already deciding that for us. So can we really call it democratisation?
The making of an AI song
Maybe AI music isn’t as new as we think
The real question is, if a song with no human backstory, no lived context, no body behind the voice can move people just as much as one that has all of that, what were we actually responding to all along? Was it ever the person, or was it a pattern of intervals and syllables the person happened to produce, one that the algorithm has now learned to produce without them?
This is especially relevant for the Indian music context, which is usually associated with playback singing for films. Our most popular songs have long been interpretations of the film’s story, rather than that of the singer’s. In that sense, AI music isn’t a rupture from Indian popular music’s history so much as its logical next step.
The playback system already separated the voice from the personal story behind it; the advent of remakes and nostalgia/reel economy already treated songs as fungible to whatever sells best in a 30-second loop. AI is a faster, cheaper version of a commercial logic that was already in place.
Theodor Adorno and Max Horkheimer’s idea of “culture industry” feels strangely prophetic here.
They argued that popular music under capitalism was never really the spontaneous, individual expression it claimed to be. Rather, it was standardised and built on repeating formulas, so it could be mass-produced. Seen this way, AI-generated music isn’t the rupture it appears to be. The industry was “administering” music long before an algorithm existed to do it more efficiently. AI simply makes it literal, automated, and impossible to ignore, stripping away the last illusion that a human was ever fully in charge of what got called popular taste.
When the human behind the song disappears
The philosopher Martin Heidegger warned decades ago that technology’s real danger was never the tool itself, but the way it quietly retrains us to see everything, rivers, forests, eventually art, as raw material waiting to be called up on demand. A song, under that gaze, stops being a song and becomes an infinitely regenerable piece of content, available endlessly.
Oscar Wilde once wrote that life imitates art.
These days, the machine has learned to imitate us so precisely that we have stopped asking whether life was ever really in it to begin with. And maybe the real punchline is not that the fake was convincing. It is that so many of us, once we found out, simply didn’t mind.
(The authors of this article are Development Practitioners)
Making a song used to demand years of training, but now, anyone with a phone can produce something that sounds worthy of virality in under a minute. 
The making of an AI song
Share
