Daniel Benjamin Miller:
Since YouTube is the dominant video platform, a lot has already been said about its copyright claims system. Most of that has had to do with questions about what constitutes fair use (in movie reviews, for instance). Google’s AI is mostly trained to recognize the presence or non-presence of content in your video, not whether or not your use is fair (something that varies between jurisdictions, by the way). But what I’m going to discuss here is much more clear-cut: the performance of public domain music.
While there’s certainly artistic merit in new compositions, there are also a lot of us who enjoy performing existing classical music. Luckily for us, there’s lots of that in the public domain. The boundaries of what’s fallen out of copyright depend on the country, but virtually everything published in the nineteenth century and earlier is free for everyone to use.
Unfortunately, Google seems not to be aware of this fact. Recently, I uploaded a production of Gilbert and Sullivan’s Trial by Jury to YouTube, and it instantly got hit by four copyright claims. I went to go contest them — something I’ve done before for this kind of thing, and always with success for public domain music — but that really shouldn’t be necessary. Trial was first published in the 1870s. Gilbert and Sullivan died in 1911 and 1900, respectively. As far as I know, this opera isn’t under copyright anywhere in the world.
Daniel notes that the YouTube AI correctly identified the songs, but nevertheless claims that they're melodies of copyrighted songs. It's hard to know how the AI came to that conclusion, because it's a black box that only Google has the keys to. I have to suspect that its training led it to presume that the audio is from a professional production, which would most likely have copyright on it from being recorded for sale, streaming or broadcast.
Unfortunately, as Daniel points out in the final paragraph, it's unlikely that Google will put in the necessary work to identity performances of works that are in the public domain for the simple reason that most people performing such works aren't in a position to sue to force Google to change their AI.