* an AI model is not a backup of the contents of all of the books in the sense that it would preserve their contents or similar such it might e.g. be useful for future generations
* Meta has (allegedly) been unfairly benefiting / profiting off of the copyrighted work of others by illegally reproducing copies of their work. Not just in the AI model sense[1], but actually (allegedly) downloading them directly from pirate repositories in a way that isn't straightforwardly fair use and even uploading some amount of this pirate data in return.
I feel like the parent commenter may have been making the typical argument for preservation of copyrighted materials, and I'm amenable to it... when it's regular people or non-profits doing that work, in a way that doesn't allow them to benefit unfairly or profit off of the hard work of others (or would be connected to such a process in some way).
Plaintiffs allege that Meta didn't just do all this, but also talked about how wrong it was and how to mitigate the seeding so they might upload as little as possible. So no matter how you slice it they allegedly 1) knew they were doing something at least a little bit wrong and 2) took steps to prevent the process that might otherwise have preserved the copied materials for the public interest.
And I feel like you probably knew all this, but maybe I'm missing something.
1: the typical argument wherein the model wouldn't exist without the ingested data, a lot of it is still in there, it is of course a derivative work and the question is really how derivative is it and what part of the work can they claim is their own contribution
