“The machine only wants our files, the scans and the PDFs, of the stuff that was already halfway to weightless” is the bullshit I keep hearing. But this week it turned out it wants the actual books as well. Somewhere near Las Vegas there’s a warehouse where people slice the spines off real hardbacks so a scanner can eat the pages faster, and that picture has sat with me all week, because it’s the whole argument in one gesture. The thing that made the work is being fed to the thing that copies it.
“The machine only wants our files, the scans and the PDFs, of the stuff that was already halfway to weightless”
I call bullshit.
That warehouse is real. Emanuel Maiberg and the team at 404 Media hid a tracker inside a rare book and followed it to an Amazon site outside Las Vegas, known internally as VGT3, where staff cut the bindings off books, scan the loose pages at speed to build training data, then bin what’s left. Amazon’s buying books in bulk to do exactly this. Read that back slowly. A company pays people to destroy first editions so a model can approximate the writing inside them, and the writers were never asked. For a form like ours, where the object matters, where people still queue for a signed variant and keep long boxes under the bed, it lands as something close to sacrilege. The copy gets treated as the valuable thing. The book is just packaging to be stripped.
There’s a line in my own book I wrote well before any of this, about how comics used to teach patience, it was something like; “A child inherited a box of old issues from an older reader and with it inherited a way of reading. That apprenticeship is weaker now.” I meant it about scans replacing back issues, about care quietly changing shape. I didn’t picture the older reader’s box being couriered to a scanning plant and guillotined. The inheritance isn’t just going quiet, some dirtbag sanctioned turning it into feedstock. That story has actually boiled my blood because I do more or less all the prep on my books. The research, the writing (badly), the type-setting (badly), the editing (badly), the covers, the formatting etc etc. So I know how much soul goes into making a book. To hear about that being literally chopped up for pulp is really sad.
The same appetite shows up as a spreadsheet in the ad world, which tends to tell you where comics are heading a year or two early. D&AD, the design body that’s been handing out those little yellow pencils for over sixty years, published its Ai and Creativity Report on the 18th, built from talks with nearly two hundred creative leaders across thirty countries and a read of more than ten thousand award entries. Creative Bloq covered it. The finding that stopped me is the plain one. Agencies are cutting the junior roles at the very moment Ai swallows the repetitive tasks those juniors used to learn on, which the report calls the loss of “the training ground that builds the judgement required for future creative leaders.” Their chief executive David Patton named the quality cost too: “Ai has changed what bad work looks like. It’s no longer lazy or derivative in any obvious way, it’s competent, polished, and generic.” Same shape as the rare books. You save a bit of money now by hollowing out the thing that made the work good later.
I’ve been chasing one thread for weeks, which is what a fair test of human authorship might actually look like, and this week the labs showed their hand. Anthropic is rolling out a watermark for text, and the mechanism tells you everything. When Claude is picking between near-synonyms, say “grey” or “overcast” after “cold and,” the system quietly nudges the choice using a private key, so only Anthropic can later spot the pattern. Jason Koebler at 404 Media took it apart, and his point, which is mine too, is that the whole scheme assumes one word is as good as another, that the choice between them is noise to be exploited rather than the actual craft. The writer Jeff Jarvis put it sharper: “Anthropic declares words fungible, language random, choice meaningless.” Frank Miller told a London room a fortnight ago that it’s our mistakes that make us. The machine’s idea of provenance is to treat the deliberate word and the lazy one as the same coin.
Which sits oddly next to the other Anthropic story of the week. On the 17th the music publisher Round Hill, which looks after more than two hundred thousand songs including “Iris” and “Total Eclipse of the Heart,” sued both the Ai song generator Suno and Anthropic, saying they trained on its catalogue without asking and scraped around the locks meant to stop them. Its boss Josh Gruss said, “We are not against artificial intelligence. We are against the idea that you can build a business worth billions on top of other people’s creative work and pay the creators nothing.” Yahoo Finance has the filing. So the same company building a clever tool to prove whether words came from a machine is in court for helping itself to other people’s work without consent. Provenance for its outputs, amnesia about its inputs. One of those it built on purpose. The other it’s fighting.
I was due to sit on a small-press panel at Caption in Oxford a couple of weekends ago but had to skip to head out to Silicon Valley on a fact-finding mission (I like to go to the ‘gemba’) but I heard the mood at Caption wasn’t despair, it was stubbornness. That’s the bit I’ll be watching next: whether that refusal stays a mood or hardens into something a publisher or a scanning plant has to actually reckon with. If a shipment of books can be tracked to a guillotine, someone else’s can be tracked too.
Further reading
404 Media hides a tracker in a rare book and follows it to an Amazon site that shreds hardbacks for Ai training data: 404 Media
D&AD’s Ai and Creativity Report warns that cutting entry-level roles destroys the training ground for future creative leaders: Creative Bloq
Anthropic’s text watermark, and why treating synonyms as interchangeable misunderstands writing as a craft: 404 Media
Round Hill Music sues Suno and Anthropic over songs used to train Ai without consent: Yahoo Finance




Another informative, but desperately sad read.