Posts / ai
The Machine That Claimed My Song
A musician uploaded a video of her own recording to YouTube. An AI music company claimed copyright over it, and the video came down. Her own song, removed because a machine decided someone else owned it.
I read that over breakfast and felt the old tiredness settle in. It’s a story where nobody has to be a villain for it to go wrong, which is what makes it so reliable.
The mechanics are easy to guess at. Content ID, or something like it, matches audio against a catalogue. An AI company with a catalogue of generated music files claims the match. The generated material sounds a lot like human music, because that’s what it was trained on. The human gets flagged as the copy. Somewhere in that chain a person could have looked at it and said “hang on”, but nobody was paid to.
One commenter put the whole thing in two sentences. These companies took music without asking, used it to make more music, and now they can go after the original artists when their work resembles the output. I can’t improve on that. It’s the order of events that stings.
Another commenter offered a thought experiment I keep chewing on. Generate a hundred million tunes, then point a second AI at YouTube to claim anything with a familiar melody. You could, in theory, bring the whole system down. They said it as a joke, more or less. It doesn’t feel like one. Melodies are a small space, music theory is finite, and a system that rewards claiming first will eventually be gamed by someone with enough compute and no shame.
I work in IT. I’ve spent years around automated systems that act on a match and ask questions later, and the rule I’ve seen hold is that the cost of a false positive lands on whoever has the least power to contest it. A big label has a rep at YouTube. A solo musician has a form and a waiting period. Her video is down while she waits, and for a small artist the first few days are most of the audience.
I’m fascinated by generative AI, and I mean that sincerely. I’ve used it for work, and some of it is remarkable. But this is the part that sours it for me. The technology isn’t the problem. The problem is that it arrived in a world with no real rules for who owns what, and the people with the most lawyers got to write the defaults. One commenter said it’s the system and the people in charge who allowed this, more than the AI. They’re right. I hold that view alongside another one, which is that the tool is also built to make this outcome cheap and easy. I don’t think those two views cancel out.
There was plenty of “end all AI” in the thread too. I understand the impulse. I don’t share it, partly because I don’t think it’s available. What we could have is a platform that treats a claim against a real person’s upload with some suspicion when the claimant is a company whose output is machine-generated and unverified. That’s not a radical idea. It’s the sort of boring safeguard that gets built after a few of these stories pile up, and I’d like it to happen before the pile gets much bigger.
My daughter makes music on her laptop, badly and happily, the way teenagers should. I’ve started telling her to keep dated copies of everything, project files and all. It’s a strange thing to say to a sixteen-year-old. It’s evidence that she made it.
I don’t know how this gets fixed. Probably slowly, through some court case I’ll read about in a year, after the artist has already moved on. But I’d like to think the song gets put back up and someone at the company is made to explain themselves. That seems a low bar, and I’d like us to clear it.