From our other sites
The story
Nippon TV News reporting has revealed that online secondhand bookstores across Japan have seen a wave of bulk orders since around August this year. Booksellers report selling “100 books a day” and, on some days, sales “five times the usual,” fueling speculation in the industry that buyers are collecting training data for generative AI. Follow-up reporting confirmed export records showing more than 50 tons of books shipped from Japan to the United States. On 5ch, users debated whether even handwritten cursive (kuzushiji) Japanese-bound books and rare volumes are being fed into AI training and then discarded, and puzzled over why physical secondhand books — rather than e-books — are being targeted.
Tue, 9/22, 10:30 AM
Nippon TV News NNN
Online secondhand bookstores across Japan have seen a wave of bulk orders since around August this year. Booksellers interviewed say they’re “selling 100 books a day” and that “sales were five times normal on some days,” fueling industry speculation that the goal is to gather training data for generative AI (artificial intelligence). Further investigation turned up export records showing more than 50 tons shipped from Japan to the United States. Is it acceptable for books to be consumed and discarded for AI training?
Source: topics.smt.docomo.ne.jp / Original article here
What people said
That's probably true for printed type, but there's no way AI can handle Japanese-bound books written in flowing cursive script, right?
I don't think it's impossible — you just need to train it on the patterns.
Just give it a character-recognition app as a subroutine.
Because they'll do everything — swap out screws, oil the parts, whatever it takes!
This is already happening in China — it's called 'human distillation.' Feed an employee's entire email history into the AI, and it learns exactly what work they do and what instructions they give — then that person becomes unnecessary.
I hope that world gets here soon. Humans aren't needed.
Judging by the Shiba Ryotaro Memorial Museum, he did have a large personal library, but not quite to that extent — sounds like the story just kept getting exaggerated.
Tengyu Shoten, maybe. That could be the same shop behind the '10,000-book inquiry' in western Japan too.
It's not a person doing the reading, remember.
AI has already gotten pretty good at Japanese cursive script (kuzushiji). It'll probably end up surpassing human experts eventually.
China's got its own AI too, so… need I say more.
Honestly, this is exactly the kind of thing Japan should be doing — building its own AI and cataloguing and preserving this information itself. If they're going to waste 7.2 billion yen on PR campaigns, they should spend it on this instead.
With stocks, past price movements aren't that useful as a reference anymore — AI itself has become one of the players moving the prices now, not just an observer of them.
You mean quants. Financial engineering has a long history — Black Monday in 1987 and the repeated crashes since then are often blamed on quants. Forty years on and it's still like this — useless.
They cut the spine off and scan it. The disassembled book gets thrown away. There's suspicion that a lot of books — including rare ones — are being discarded this way, and it's becoming a real problem.
These days there are scanners that turn pages without cutting the book apart. There are videos of them on YouTube.
Because some information gets stripped out when books are digitized.
Apparently e-book terms of service prohibit using them for AI training… plus they're scooping up old books that were never digitized in the first place.
I think that's exactly the problem.
Start with abolishing the tipping system — that's where this conversation needs to begin.
This isn't just Japan — it's happening worldwide.
Same with sociology books, isn't it? They take what's trendy overseas and use the language barrier to mark it up and sell it here.
Background and Key Points
The unauthorized scraping of online text and images to build generative AI training datasets has repeatedly sparked controversy across various countries. Physical secondhand books often have murky copyright status, whereas e-books are typically restricted from AI-training use under their terms of service — a distinction some thread posters pointed to when noting that buyers seem to be “scooping up secondhand books rather than digitized ones.” That said, this report rests only on booksellers’ testimony and export statistics; it has not been confirmed that the buyers are actually using the books for generative AI training. While the thread raised concern that even cursive-script (kuzushiji) materials and rare books might be cut apart and discarded, the article itself does not confirm the actual scale of disposal or exactly which books are involved — worth keeping in mind.
*This article is compiled and summarized from the 5ch (Breaking News+) thread “Mysterious Bulk Orders Hit Secondhand Bookstores… AI Training? Export Records Show ’50 Tons of Japanese Books’ Bound for the US.”
Leave a Reply