It is designed to stop AI from learning from it. It is intended to work as another form of copy protection.

Maybe, but AI isn't people and doesn't do learning, it does "training" which we can roughly anthropomorphize as learning.

It doesn't really seem like an IP thing to me - it's like making music that's intentionally hard to classify in a particular genre on purpose or something. Maybe that defeats Spotifys recommendation algorithm but that's totally unrelated to any rights to stream it or so on.

You cannot both have an algorithm that stops AI and allows humans to listen in the long run.

Eventually encoders/classifiers will get an 'anthropomorphic' loop of human ear capabilities to use as a prefilter and identify when sounds differ from the expected human hearing and identify the noise.

I mean this is what adversarial reinforcement learning is, humans cannot beat AI on this in the end.