Comment thread splits over whether AI book-scanning is fair use
2 Sep 27 11:53 AM · 1d ago · 1 post · 10 comments · 1 source · development 2 of 3
A lengthy reddit discussion argues over whether destructive scanning of books to train AI models constitutes fair use/transformative use or unlicensed commercial copying, citing ongoing lawsuits and publisher licensing deals.
“currently looking to purchase large volumes of secondhand books”
Innodata representative, AI data company representative (via LinkedIn message) · reddit ↗Anthropic AI company operating Claude chatbotPatrick Hempelmann Owner, BMV Books (Toronto)Peter Sellers Owner, Sellers & Newel (Toronto)Innodata New Jersey-based AI data companyZoom Books Pitt Meadows, B.C. book reseller and recycler
The whole story articlespostscomments the bright band is this development · numbered dots are the others · click one to jump
What people said 10 voices · verbatim
-
https://lawyerinfo.ca/guides/money-taxes-ip/intellectual-property/copyright-exhaustion-first-sale-doctrine-equivalents-in-canada/ Copyright Exhaustion: First Sale Doctrine Equivalents in Canada . While the United States refers to this legal concept explicitly as the "First Sale Doctrine," Canada approaches this through the legal principle of the…
-
Format shifting for personal use it generally considered fair use. LLM training is generally transformational, they are using the book to create a map of how language is used in practice, rather than caring about any particular detail for phrasing. On the other side it's commercial use which weights away from a finding of fair use, and the…
-
I don’t think you could argue that using the books for AI is the same as copying it. In reality the text was fed to a model and based on the output of the model a matrix of numbers were tweaked. If it were straight forward then we would have seen some action in the courts by now. Of course since I’m not frothing at the mouth with hatred of AI in…
-
The act of scanning a book is “copying” it, so you need to have a copyright or an exception to do that. To scan for yourself is fair use, to scan for commercial purposes is absolutely not allowed. There also has been action in courts, major publishers have sued and gotten licensing contracts out of these companies
-
it’s so crazy to me that one of the main things people use to detract against ai, a thing which famously has loads of terrible things against it, is the utterly ridiculous notion of intellectual property
-
This is still a better use of old books than assholes buying them up and organizing them by colour on shelves for decor to cosplay as intellectuals.
-
Along with a whole host of others, yea. Considering the people who’d have to do the updating, though, I’m not holding my breath lol
-
Copyright law does explicitly protect against unlicensed use of creative works, I’m not sure which jurisdiction you are in?
-
Right, I’m saying that scanning it to feed to my AI model is personal use and not redistribution.
-
There is action in the courts right now in many countries including the U.S. and Canada.
All 3 developments of Anthropic's mass book-scanning for Claude sparks bulk-order… →
RedditGoogle News