Archives for the month of: November, 2025
Thad McIlroy provides a lengthy and thorough analysis of what’s going on in the struggle between LLMs and publishers over the use of content and AIs’ need for it in order to train themselves.
Like so many commentators he assumes book publishers are at fault for their failure to push their interests in AI licensing. Maybe. Maybe not. Our business is mature, and even if slightly fewer units are being purchased we can still survive through increased prices and reduced expansionist dreams. Payments for AI training usage represent a nice bonus, but are hardly fundamental to the economics of the business. The way forward is cloudy, and just because the judge in the Anthropic case concluded that in principle the use of books to train AIs could be regarded as fair use, this doesn’t mean that the issue is settled. Personally I suspect it probably is, but one legal aside doesn’t conclusively settle the law. There are apparently fifty-four lawsuits on the subject currently working their way through the system.
We often forget that since ebooks and digitization have only been around for a few years, it remains true that thousands and thousands of books are not available in any digital format. It’s obviously a lot easier for an LLM to license an electronic version of every book than to go out and buy a second-hand copy and scan it, a procedure which Mr McIlroy estimates would cost an AI company about $25 a book. So if reading for training really is fair use, publishers should be flocking to offer their books at $25 a pop. Anticipating action there are now appearing brokers who offer to mediate between publishers and AI companies — Mr McIlroy lists thirty of them. He tells us “I’ve been talking to many of them, and they’re mostly serious and thoughtful about the challenges the publishing industry faces. But it’s early days, and the paths to content licensing remain uncertain.”
Most books reside in the Knowledge step on the Information Pyramid. A few, the very best books, may make it to the top tier, and embody real wisdom.
Datacamp provides definitions for the steps.
- Data refers to raw, unprocessed facts and figures without context.
- Information is organized, structured, and contextualized data.
- Knowledge is the result of analyzing and interpreting information to uncover patterns, trends, and relationships.
- Wisdom is the ability to make well-informed decisions and take effective action based on understanding of the underlying knowledge.
Mr McIlroy explains “What AI is very good at is knowledge: “the result of analyzing and interpreting information to uncover patterns, trends, and relationships.” That’s AI LLM’s not-so-secret power.”
The real threat/promise, we are told, comes with RAG, Retrieval-Augmented Generation. RAG will potentially deliver something like wisdom but it needs to be trained on everything relevant to the topic it is engaging with, and way too much of this material is now concealed behind paywalls. Licensing will be focussing on this area.
Well, straight off, I think that all this manic attention to AI in books is a bit of a diversion. I can just about see how a prize-giving organization would want to avoid awarding their prize to a book entirely written by a computer, though I wonder just how such a production could get through the editorial and publishing process without someone noticing. And anyway if it really was brilliant, why should such a book not win an award? Furthermore, there are lots of kinds of non-fiction books where refusing to use the organizing power of AI in compilation would be just dumb and time wasting. Do purists want us to disable spellcheck?
However, The Guardian tells us about New Zealand’s Ockham Book Awards refusing to consider a couple of fiction book because their covers used AI in their design — there’s no suggestion that AI was used in writing the texts.
These are perhaps not the most wonderful designs ever. Maybe the humane judges disapprove of fitting cats with human dentures. Ms Smither, Angel Train‘s author, is quoted as saying that “the designers spent hours working on the cover of her book, which features a steam train and an angel ‘half-obscured in the smoke’, inspired by artist Marc Chagall’s figures”. (It would have been nice if the AI design had actually included a non-diesel train to generate that steam — not smoke — though.) To my mind it is impossible to see what deleterious effect these covers could have on the content of the books or of a reader’s enjoyment of them.
So the judges were foolish. What if the publishers had just submitted the books without their jackets? Or maybe they could just bang out a temporary straightforward typographical cover showing just author and title. “Nicola Legat, the chair of the book awards trust, which administers the Ockham awards, said the trust takes a ‘firm stance on the use of AI in books’.” NB — IN books not ON books! Do bear in mind Ms Legat, that publishers design covers, while authors write books.
This decision is just silly. It’s sort of like objecting to the use of metal saws for cutting down trees because it’s just too easy — people should be content, and proud, to use a flint handaxe just like our stone age forefathers did.
Another extravagant AI problem comes to our attention via Plagiarism Today. I must confess I wondered if it was an April Fool’s trick when I first read the piece, but we’re still deep in winter. Apparently AI has been generating knitting patterns which turn out to be impossible to make on your own little knitting machine. Punch cards have for ages been driving power looms and knitting frames, so in a way AI shouldn’t have any trouble. I would, however, have imagined new patterns were not something likely to be in big demand in knitting — after all a sweater’s a sweater. But no; apparently “fiber artists” will pay good money for new ideas, and they are feeling gypped.
We’re so used to them that we almost don’t think they ever had to be designed. I like to see 3:33, 5:55 etc. on the alarm clock. Unfortunately you have to lie awake to observe that phenomenon!
There’s a mirror in our elevator and I often end up looking at the floor indicator via the mirror and noticing that just before we get to the ground floor we pass floor E then floor 5. (Almost stranger is the fact that in USA the ground floor is floor 1, whereas in Britain it’s 0.)
So where did these numbers, made up from seven little straight lines, come from? Wikipedia gets right down to it: “Seven-segment representation of figures can be found in patents as early as 1903 (in U.S. patent 1,126,641), when Carl Kinsley invented a method of telegraphically transmitting letters and numbers and having them printed on tape in a segmented format. In 1908, F. W. Wood invented an 8-segment display, which displayed the number 4 using a diagonal bar (U.S. patent 974,943).” You’d be hard pressed to use this same technique to create letters though: most of them would work, but for D, K, M, N, Q, R, V, W, X, Y, and Z we need a diagonal or two, so you’d probably need to end up with at least an eleven-segment representation. Luckily for us Liquid Crystal Display has come to the rescue, so your computer monitor doesn’t have to create chunky square letters from tiny straight lines.
Do I detect a tendency for Shelf Awareness and other bookstore social media to feature ever more photos of cute cats among the books?
For example this guy at Blue Cypress Books, New Orleans.
I regard this as a dangerous trend. Many people prefer dogs to cats, and we need to be careful not to alienate any part of our audience, especially, heaven forfend, those afflicted by some sort of cat allergy. Don’t be directing any more people to your online rivals!
We have a cat patrolling our local liquor store. I’ve never been able to work out what data he’s collecting, but I always regard my haul as subject to cat-scan.
The Scholarly Kitchen sounds the alarm.
The first scholarly journal ever issued was published by a learned society. The Philosophical Transactions of the Royal Society was established in 1665 as the first scientific journal. Is 360 years enough? There have grown up plenty of university press and commercial academic publishers of journals, so surely there isn’t any existential reason for publishing to be carried on by learned societies. There’s also of course absolutely no abstract reason why publishing should not continue to be carried on by learned societies.
The learned societies exist to promote knowledge in their own subject area, and publishing that knowledge could be regarded as an integral part of that promotion. For many learned societies their journals may be regarded as an important benefit of membership; but I suppose this could ultimately be evened out by a change in subscription rates. There seems already to be underway a trend away from doing it all by yourself. “Between 2015 and 2023, the number of wholly self-published UK societies dropped from 68 to just 44 — a 35% decline in eight years. The bulk of this decline (17 societies) moved to outsourced models with single publishing partners, while five ceased publication entirely.”
We do, of course, have to acknowledge that as time passes circumstances may change, rendering something which we’ve always done no longer essential. Whether or not the knell is sounding for learned society journals is something we can’t really judge from the outside, but costs must constantly be rising. If every learned society stopped publishing tomorrow we’d no doubt experience a temporary log-jam as papers accumulated at “commercial” publishers awaiting publication. We’d probably be able to work through that in a year or two, or three, or four.
However, as reported elsewhere, submissions to scholarly journals are way up — so maybe fewer academic journals is the last thing we need right now, even if revenue is showing a tendency to decline.
The Scholarly Kitchen brings us news which at first glance would seem to be “good news”. Submissions of papers to academic journals are way up. Which has to be good, doesn’t it? More authors in your subject area has to mean more potential readers for your journal. And, if you get more submissions you can be more selective — yes? If you’re more selective, you’ll have a higher proportion of excellent work, won’t you? All sounds great.
Well of course the gigantic fly in this pot of ointment is the ease with which AI can generate much (or often all) of your journal paper. This means in fact that much of the excess in submissions is in all probability almost worthless. So far, so obvious. The vast problem arises however when you consider the work (cost) involved in sorting out which papers are any good, and which are nonsensical, or worse — because harder to detect — just flawed. Journal editors are being overwhelmed. Obviously book publishers risk facing the same sort of problem. Does this herald the final melting away of the famous slush pile? Let’s just shift the problem onto literary agents!
Another hand binding technique here demonstrated by Kathleen O’Connell.
If you don’t see a video here please click on the title of this post in order to view it in your browser.
Here’s another binding by Linda Hanauer, showing the Coptic link stitch in association with ribbon binding. The boards are cloth-covered with an end sheet pasted on the interior.
The whole makes a sort of comment on traditional hand binding techniques, foregrounding the structural elements which hand binders have usually gone to such pains to conceal.
When the Authorized Version of the Bible (referred to as the King James Version in America) was printed in 1611, it was printed by the King’s Printer. This was merely the latest manifestation of a self-interested tradition whereby any new version of a big book like the Bible would be protected by its promoter, a member of the court usually, who would give the right to print the book to his favorite printer. Royal sponsorship, ergo royal printer: no formal agreement, no action by the king, no discussion or questioning. The book went to Robert Parker and nobody thought anything about it. That’s just the way things were. The same sort of assumption process lead to the status of the KJV as Crown Copyright in Britain. No formal decision about this ever took place: it’s just that as everyone has always behaved as if the copyright was owned by the Crown, the copyright ended up being owned by the Crown. English law allows custom and practice to gain the force of law.
Over time the King’s printer was joined by the University Presses of Cambridge and Oxford as the three “privileged presses” allowed to print the Book of Common Prayer and the Authorized version of the Bible in Great Britain (minus Scotland, which was still in 1611 a separate country with its separate legal system). This permission comes from letters patent issued by the Crown defining the charters of the two university presses. The Cambridge University Press charter from 1534 already gave it the right to print “all manner of books”, and Oxford University Press received the right when it was chartered in 1636. In early years the university presses tended not to act themselves on this licence, often subcontracting the work to other printing houses. Bible production eventually gathered pace in both presses underwriting the large expansions they both experienced in the late nineteenth and twentieth centuries, and ultimately facilitating the opening of their New York Branches.
Now you might assume that “owning” the right to print such a money-maker would lead the printer to take great care over accuracy. You’d assume wrong. Right from the get-go error was rife. One printing has in one place Judas rather than Jesus, while three lines of text are repeated elsewhere. Using defective copies as copy text for reprints compounded the problem over the years. In The Cambridge History of the Bible, Vol. 3, Michael Black tells us that it took two and a half centuries for the error bug finally to be slain by “Dr Scrivener of Cambridge who produced what was essentially the first true critical edition in the Cambridge Paragraph Bible of 1873, which was successful only because the editor took account of the whole history of the text, and all the vicissitudes to which printers had subjected it”. It had been calculated as part of the clean-up process that there were 24,000 variations between then current versions and the original translation.
The Daily Heller brings us several images of Jewish manuscripts from an exhibition, Jewish Worlds Illustrated, at the Grolier Club. The items on show come from the Library of The Jewish Theological Seminary. The exhibition runs till December 27th.
Bearing in mind the possibility of divine retribution for scribal error in the Jewish manuscript world we have to assume total accuracy for all the items on display. Here’s a scroll of The Book of Esther, written out in Italy in the mid-eighteenth century, showing at the opening an illumination containing a bare-breasted woman. Apparently such potentially disrespectful breast-baring was possible because the text does not include the name of God, and so avoids the usual proscription of any type of illustration under such circumstances.
____________________
* I did not know this before I just looked it up, but the label “people of the book” derives from Islam. Ahl al-Kitāb refers to those religions—Jews, Christians, Zoroastrians, and Sabians—who rely on a divine book. During wars of conquest a certain measure of respect was afforded to these reading religions.
Making Book 