[microtonal]

it's not a responsability of the Ai model

It's the responsibility of the maker for the AI model. They train the AI model on copyrighted material. So they should clear the licensing for using that material.

IANAL, but since I also work in the ML domain, I tried to find out how this works when you have to follow EU laws. Past rulings [1] have considered 11-word snippets to be a copyright violation (which is by no means the lower bound). So, it is likely that if a copyright holder in the EU can show Co-Pilot or ChatGPT to reproduce a non-trivial fragment of code or sentence, that a copyright holder can sue them successfully.

However, the sad fact is that these models are made by well-funded entities. So, they'll bury small copyright owners in lawyer busywork until they go bankrupt and settle with big copyright holders. So one possible net outcome will be that large entities can do large-scale copyright violation while individuals and small companies can't. We have seen this story before. And it helps to entrench big companies even more.

I hope that the EU comes with some regulation to level the playing field. So either make it illegal for everyone (enforced by EU courts) or legal for everyone.

[1] https://www.theregister.com/2009/07/31/ecj_rules_11_word_sni...