On Theft
Right now I am writing from Box World. One half of the room is taken up by stacked boxes with KITCHEN or BOOKS scrawled across the top, the other half with boxes-to-be flattened against the wall, next to parcel tape, scissors, bubble wrap, old newspapers, random screwdrivers, all of the detritus of moving house.
Part of moving house is taking an accounting of all of the things you own in the world, and deciding whether you still want them. To that end I’ve spent the last couple of weeks getting rid of hundreds of books. An empty shelf is a sad thing, and I am living in a flat of empty shelves. Only the unread remain — all the collected mass market paperbacks, charity shop punts, unfinished-books-by-friends (sorry) and liberal coffee table books from well-meaning relatives have been taken laboriously on foot, one heaving Bag for Life at a time, to the charity shop down the road. They didn’t even have a book section when I first went in there. They very definitely do now.
Buried underneath a stack of oversized books too tall for the shelf (Masterchef companion recipe book from 1999; Illustrated Guide to British Moths; a Private Eye annual) I found a small pile of dog-eared folders. Inside, having made their way through about five consecutive house moves, were my belongings from school. Textbooks, diaries, qualifications, awards. Sat cross-legged on the floor, a box half-full and abandoned, I took some time to read what was in there. In between two folders, not attached to anything in particular, was a page torn out of a workbook, and on it was a short story I wrote when I was 14.
Writing is a practice, not an output. If you end up with a finished product which you can make a living from, all the better, but it’s not the point. You should write like you hike, or play the guitar, or swim. In my own lazy, meandering way, I’ve been writing since I could hold a pen. There were periods where I managed enough momentum to write 1000 words a day for several months, and some middling unfinished novels to show for it. There were also periods where I was too depressed to do much but play Football Manager. One way or another, though, I’ve dedicated the majority of my life to reading and writing.
I am a librarian by trade, and have been for getting on for a decade. My grandfather was a writer too, though I never met him except through his autobiography (my dad finds it a frustrating read — too ironic and slippery, too unwilling to truly share anything — very strong evidence for heredity). This year I performed a paid poetry gig, I wrote a zine, I’m going to try and get a pamphlet published and I want to get my latest unfinished novel over the line.
There, bona fides done: I don’t care if an LLM uses my work.
I have complex views on LLMs (I won’t call them AI because they aren’t). I think it’s likely that they’ll have a profound effect on the creative industries through the undercutting of the day-to-day work which supports personal creative practice. A lot of authors are copywriters. A lot of illustrators do commissions of dogs. The world we live in is hostile to making art for anyone who isn’t wealthy, and the economic consequences of widespread LLM adoption will be felt. It will make the artistic sphere smaller and more privileged than it already is.
I also think the ensloppification of the internet, as part of the general right wing project of information destruction and the degradation of the intellectual sphere, will make it harder and harder for people to learn and find things which are true and authentic. I think LLMs produce an awful lot of crap and though there is a soft meritocracy to art and production, it isn’t enough to stand in the way of the total bludgeoning force of so much crap pouring out at once. You cannot stand in the way of a tsunami with your palms outstretched.
I don’t like the committed anti-art positions of those who spend their whole time selling a snake-oil version of “AI” either, but I also think they are simply wrong. I have never seen a work of art produced by an LLM which is better than art made by a skilled human. I have seen a lot produced which is as good as art made by an unskilled or untalented person, which is regrettable for the economic reasons above but not an inherent threat to great art and creativity. As far as the present iterations of LLM art go, there is nothing a fraction as good as a talented human artist. There are obvious reasons for this — the nature of an LLM is such that novelty is impossible, only recombination, the production of an average according to a prompt. Great art comes from innovation. There is also an issue of capacity — an LLM cannot think and it cannot remember, so it cannot construct a narrative, it cannot develop a theme. There are technical limitations to how many prompts it can process at once, and a novel is a thousand prompts, shifting and weaving and intersecting. All it can produce is the texture of good writing — a plausible sentence, a pithy phrase — over a short burst. We have nothing to fear.
(There is a debate as to whether it can be called art at all, but most of this is flatly wrong: art requires intentionality, yes, but prompt engineering is intentionality. That a tool is often lazily involved doesn’t mean it isn’t art, just that it’s very bad art. An identical argument was once leveraged towards photography, or towards the use of synthesisers or computer production in music, that a tool is doing the creative work. The answer is as it always was — a tool used lazily and thoughtlessly produces bad outcomes, but it isn’t conceptually not art. A person who fills a pendulum with paint and then chooses an angle of release and set of colours is still making art. The project of limiting what can and cannot be art is anti-art and anti-artists — it should be enough to say that the output is bad and lacking artistic value.)
There are also compelling ecological arguments, though some rebuttals too (the bulk of the power usage is in the training of the data sets — an infrastructural investment, essentially — and day-to-day use is comparable to any other server load, whether it’s Netflix or Claude). I am not particularly qualified to comment on that aspect.
As far as my ownership over my writing goes, and the degree to which it is “stolen” or “plagiariased” by an LLM which uses it as part of a training data set — I have no problem at all. I don’t believe that intellectual property protects artists, and I don’t believe that a homeopathic quantity of a text ingested into a data-set as wide as humanity constitutes plagiarism in any normal sense. The strongest case I can make is that individuals should be compensated to the tune of the price of a copy of their book which was used. This does, in fairness, make the fullness of Meta’s theft quite large, but none of it impedes the individual writer’s ability to make money from their work. It sticks in the craw to see that sort of wide-scale appropriation from a very evil organisation, but I think the outcomes of arguing for more robust intellectual property protections are far more damaging for humanity.
There are several organisations, such as LibGen and the Internet Archive, whose existence is solely for the benefit of humanity. I don’t think that is an exaggeration. Perhaps this speaks to a slightly old-school-internet mentality I have: information wants to be free. The greatest gift you can give someone is knowledge. I am a librarian and I am left wing and I believe in the maximal transmission of knowledge, books, articles, databases, art and culture.
Both of LibGen and the Internet Archive are under continual assault by rights holders. By academic publishers, who make the highest margins in the world in order to gatekeep knowledge, who charge academics to publish their own work, who gouge public bodies and rip off everybody possible. By commercial publishers and bodies whose claw-like grip on intellectual property is designed to squeeze every last drop of blood from the stone. We live in a world of DRM takedowns and lawsuits about sampling. The strengthening of intellectual property rights — the turning of art into a commodity — will never benefit the individual artist. And, specifically, the redefinition of plagiarism to mean simply the ingesting, the use towards a data set, the inclusion of an indetectibly tiny quantity of a work, could have extremely damaging implications.
Do I own the full text of my novel? Perhaps. A chapter, certainly. A paragraph. Do I own a sentence? Do I own a word? Do I own a structure or a theme? What do I owe to writers who have influenced me? Whose phrasing I mimic and whose ideas I reproduce? How substantial does an appropriation have to be before it is plagiarism? What happens if this new model of intellectual property is accepted and people start looking at our work for the use of licensed property, however minimal? Who is being empowered by the assertion of property rights?
Perhaps I’m not looking in the right places, but I don’t see these questions being answered. I see something which amounts to a moral panic — something which is unfocused and not always based in fact. I don’t think the use of LLMs is inherently evil, though I see many things about it which could be bad for humanity. Also many which might not — the use of LLMs for accessibility purposes can be empowering, so too with machine-learned translation, as well as the manifold clinical uses. The technology is too broad to be spoken about as a single thing. Is someone having an LLM producing their shopping list the same thing as Israel using the technology to commit genocide? No, and it elucidates nothing to pretend that it is.
My objections to LLMs are always first and foremost derived from my politics — I am a trade unionist and I believe that the fruits of automation should be shared with workers, not shareholders. I believe we should organise in our workplaces to ensure that we don’t suffer job losses because LLMs are taking on some aspect of the work. This is the same principle to hold to any automation of this kind — whether it is the mechanical loom or a chatbot. I think it’s very possible that we’ll look back in a few years time and be able to categorically say that LLMs and their widespread adoption were a disaster.
As far as my writing goes? Claude, ChatGPT, Grok, Gemini and DeepSeek, fill your boots. There’s enough to go around.
