As I see it, you shouldn’t, while you enjoy the results of your model in your room or share it with friends. But if you open the model or start distributing the results of it, then it makes sense that you compensate the original author as a derived work in some way (kind of like we do when sampling or covering).

It is tricky indeed, and the gray area is quite big

What I tend to do is put myself in their shoes - how would I feel if such and such happened. How would you feel if someone took your voice and started making tracks/money out of it…

I remember when videos of Jay-Z reading the Bible began surfacing on YouTube a few years back. Of course, he issued DMCA takedown notices, but ultimately couldn’t stop them.

According to the article, the bill being proposed would:

… require developers to be transparent with copyright owners about using their material to train AI models.

While I feel the artists have a right to earn from their works, it’s a tricky issue. If I’m listening to Dua Lipa and want to train a model to sing like her from music that I bought and paid for, why should I pay her again?

But that’s speaking from a low-scale indie level. The corporations of course would love such a form of argument because it allows them to plunder without recourse. And that’s bad for artists and bad for creators.

The question is: where is the middle ground? And is there one, if any?

This whole AI thing is a tricky debacle. Wading in this war makes me feel like a Luddite, so I’ll just use AI and watch the others fight.