This is going to be a challenge because the principle, not only the regulations, have to adapt. The case of software copyrighted as literature is interesting because it shows that even though its very different it Was easier Just to adapt what existed.

Veliz considers the use of such data to be problematic because it violates privacy. And then makes the point to broaden the notion of privacy. Your suggested use of Plagiaristic seems to go round that way also. How would it compare to such projects?

The law is not a side issue. Many countries have copyright that are human centred. Its intersting how for example software, books, and tradrmarks get protected by the law in virtue the understanding of what it is to create them.

But does it apply to any kind of AI training on data that is not 'cited'? It seems to hang on what status the relevant pattern has. Images seem to have patterns that can plausibly be Plagiaristic. Text as well? Less obvious and in cases like games liek chess or go? Stockfish is plagiaristic?

I dont even find the notiong of Plagiarism very usueful in an academic context. Id rather ralk about intellectual dishonesty (ie, getting a gohstwritter). Maybe useful in a legal context. Are you thinking of that sense?

Why think this?Plagiarism applies to agents, capable of neglecting or choosing not to cite. No? Maybe a diff notion that captures the idea that AI uses other's work in a problematic way would be better? Saying that it's Plagiarism seems to explode the notion: All your base are belongs to Plagiarism!