Hacker News Reader

Comment by littlecranky67 | Hacker News Reader

Well they have to destroy them after scanning, else it is a copyright violation. Maybe not in the US, but in a lot of jurisdictions you are allowed to make copies/scans of Books, CDs etc. but if you sell the media, you have to delete the copy.

Given that anthropic was just fined a $1bn fee for torrenting/piracy, we cant really blame the companies for shredding the books after a scan. We should blame copyright laws.

View on HN

> Well they have to destroy them after scanning, else it is a copyright violation.

I think this is the "an undercover cop has to tell you they are police if you ask them" of AI law

So push back and explain how that user is wrong?

I wouldn't characterise it as 'undercover cop...' but perhaps it might be fairer to say that until it's tested in court, no-one knows, so they (Amazon etc) are taking the 'safe' path of destroying the books?

Well, trying anything in court is definitely what they want to avoid. I mean, what is your suggestion here - should they buy the book, scan it, then sell or give it to someone else? That IS a copyright violation in itself, no need to try anything in court.

Whether or not AI training or replication of the text through the model is a copyright violation, is a whole different but unrelated question.

I don’t think it’s a sure thing that training a model on a work is considered a duplication in the sense that ripping a CD might be for example - unless I’ve missed something, that’s what hasn’t been decided in court. I think they see destroying the book as the safest course of action for them.

But they do not delete the scan after training, they keep all the training data for future use and future training.

Yes, and I imagine that’s why - in the minds of their lawyers - doing anything with the book (other than destroying it) is arguably less safe than destroying it

Reply on news.ycombinator.com