Nora Roberts Had 204 Books Stolen…
By ChatGPT…
Without permission, without payment, just 204 novels stolen whole cloth and used to train the AI stuff so lazy-ass writers can use her work to pretend to make their own crap better.
Nora just discovered this and is not happy, to say the least. I stopped counting when I got to 28 of my novels stolen without permission or payment.
What is bringing this up tonight is that tomorrow I will be pushing a bunch of Kickstarters by writers that are going now. I will put them on the Kickstarter free class and send out a letter to all of the writer’s signed up in that class, all 1,500 of them.
And I will put them all here as well.
But Kickstarter now has a policy that if you used any kind of AI in your book, you must declare it and if you don’t and they find out, they will cancel your project and ban you from the site.
There would be two projects I would have promoted here, but they used AI. I will only promote writers doing their own work, not writers who steal from other writers.
I would highly suggest that if you see a writer declaring on Kickstarter that they have used art or text AI (at the bottom of the story), cancel your pledge. Don’t support writers who steal from other writers.
The hundreds of major lawsuits are going to take care of this soon enough, but for now, stand your ground.
24 Comments
Harvey Stanbrouigh
The “writers” (and I am NOT one of them) are at secondary fault. I read about the suits, but I don’t read the suits. Never passes Carpenter’s WIBBOW. But eventually I hope the courts prosecute the scuzzy humans who steal the work to train the machines.
Amy
That is horrifying!
But how can a writer tell if their novel has been ripped off?
dwsmith
There is a program that finds that. Web site has been set up. Just google it. I don’t want to post it here.
Kevin McLaughlin
At this point it’s pretty safe to assume that if your book was ever in US public libraries, someone probably used it for training. One of the major sources for books was the Google Books collection, for some AIs (like Bard).
KDP also has terms in their TOS allowing them to use all KDP books for “indexing” – a technical term for one form of AI training.
Philip
I started reading a book a couple weeks ago where the first paragraph was clearly a mix of distinct lyrics from two different hit songs. Not quotes or references, just wholesale lyrics presented as prose.
I read on another paragraph out of curiosity, and the phrasing was awkward.
I realized I was reading an AI book — garbage from the plagiarism machine. Returned it for a refund and reported to Amazon.
Total disgrace.
Kristi N
I wish I could be surprised by the theft, but I’m not. I saw this happening in a TV newsroom as new graduates were hired on. Their attitude was “I want to use it, so I’ll take it” and they wouldn’t stop even when the wrongness was pointed out. The only thing that made them stop was the general manager bringing in the top lawyer from the corporation to lecture on the consequences of intellectual property theft by employees of a commercial business. Once he uttered the words “you will be fired”, the theft stopped.
On a different note, I came across an AI written post on FB, and read it with increasing disbelief. It was at the skill level of a desperate teenager with a new thesaurus and a seven page paper due at the end of the week. Florid phrasing, wrong word choice, lack of logical progression in the narrative…I could go on, but it was horrifically bad. I agreed with Kris when she wrote about AI being mediocre at best, and using it drags the craft of the writer down to that level. After reading that post, I agree even more. AI doesn’t understand, it can only parrot meaningless bits and pieces from its memory banks.
Balázs
Uhh, I didn’t really understand this whole theft thing until one of my computer genius friend explained me how AI art works. He stated that a correct use would be when someone write an AI programme for himself or herself, and teach it via his or her own materials. The AI could be useful to work with in that context, learned with their own materials, used their own articles and texts, but with this there is no theft involved. Sadly, no one do the thing like this.
(And I’m not sure. This still lazyness.)
But when someone taught their AI with other person’s work, and use them as if it was their own… It’s not the same. It took some time for me to understand why and how it is bad.
At the very least I understand now…
Alexander Boukal
Horrible that your, Nora Robert’s, etc novels have by stolen by those tech-savvy idiots who don’t realize that they’re stealing artists’ work. Can’t wait to hear the sweet music of the copyright and ai lawsuits play out.
Does that new AI policy on Kickstarter only refer to ai generated text and art, or does that include using ai editing programs such as Google Docs, Grammarly, ProWritingAid, etc to find typoes? Do you need to declare if you are using one of those editing programs?
dwsmith
Just AI generated text and art.
blitchfield
I still won’t use Grammarly because their TOS states that they store and keep whatever text you have them check. True for both the free and paid versions. They’re using it to further train their AI on writing patterns. So you’re opting in by using their tool. Unlike GPT, they are upfront about it.
On the other hand, AI does take grammar and typo checking to a new level. There’s nothing gramatically wrong with this sentence, for instance: He attends pubic school. MS Word will not catch the error. Takes specific info from an AI training model to spot something like that.
Kevin McLaughlin
This isn’t true, thankfully. Grammarly got accused of this by Cat Rambo earlier this year, but it was found to be a false allegation.
All grammar checkers, including MS Word, are based around the same DLM (deep learning machine) technology as GPT. They’re just earlier, simpler versions of the tech. But Grammarly and PWA claim they paid for or licensed all training works.
Kevin McLaughlin
I’d suggest making an exception for anyone using the Adobe Photoshop AI. Reason being, Adobe got legal licenses for all of the art used in training their AI, and they are actively paying ongoing licensing fees to everyone whose work was used.
I’m impressed, because Adobe generally is a pretty crappy corporation (aren’t most of them?), but they are the SOLE developer of an image AI that has gotten legal licenses and consent for all training works and paid all artists whose work was used.
I’ve always made my own covers anyway, and I have to tell you it’s SUPER nice to know their AI was ethically trained – especially since that AI also underpins about a quarter of all Photoshop features at this point, like healing brush, smart selection, scaling, and all kinds of other stuff.
Jason M
That’s really good to know, Kevin.
I make my own covers too — about 25 so far this year — and the Photoshop AI generative fill has been an absolute lifesaver when images don’t stretch far enough, as well as for spines of print covers.
Kerridwen Mangala McNamara
The mess here – and I am NOT defending AI users, I agree that this is AWFUL – is that it’s perfectly reasonable to train a *human* writer or artist by having them copy great works in their field… to learn techniques, and so long as they *don’t* pass off that work as their own (attributing, etc, as appropriate). This is how the Great Masters learned to paint, for example.
The thing is that, so long as a *human* is going to produce their own work for real – and NOT simply copy someone else’s – even if they try to mimic another creator’s style, their own unique voice will inevitably come out. We’ve all seen wonderful art with acknowledgement of the creator’s major influences. That’s how the State of the Art progresses.
Someday AIs may develop their own “voices” based on the variety of influences they’ve been trained on… not just Nora Roberts or whomever. And not with the intent to pass off something as someone else’s work or to steal their readership/viewership as a too-close-for-comfort mimicry, but as their own voice.
And that will be a much more complicated problem to consider. Because what really *is* the difference then?
(Side-note: I nearly ruined 2 family gatherings this year by starting to get into an argument with a computer scientist who is on one of the major AI development teams. He sees all the beautiful potential… and I – as a former genetic engineer, who used to be on the opposite side of a very similar sounding argument – am seeing the dark side.)
C.E. Petit
Kerridan, I can’t really agree with this, but it’s because of the difference in the way a human copies and a large-language-model-system-based-on-current-technology copies.
When a human copies — especially in the visual arts — the making of the copy is mediated by the brain, and no part of it is retained later. The process looks something like this:
{perceived work} –> {brain} –> {determination of steps to replicate} –> {application of skill to replicate} –> {replica}
When a Von Neumann processor (and that’s all processors in use today, or on the near horizon; it is theoretically possible that a quantum processor, or some other type of processor we can’t yet imagine, will work differently) “digests” a work to “train” a large-language-model or other generative “AI”† system, part of the copy is retained later. This process looks something like this:
{source work} –> {local replica created} –> {processor} –> {analysis} –> {add analysis to existing analytic database}
Notice that the place where there is a complete replica is completely different: It is only after all human work is finished, but before any computer analysis is performed. This is due to the nature of Von Neumann processors, which necessarily (due to their architecture) feed exact copies of source data through the processor’s registers; the processor performs all operations on the registers.
Copyright law already recognizes the result of this. You may recall, not quite a couple decades back, the Grokster matter (MGM Studios, Inc. v. Grokster, Ltd., 545 U.S. 913 (2005)). Indeed, it’s been implicit in copyright law for a quarter of a century now, see 17 U.S.C. §§ 512(a), (c) (distinguishing between “transient” and “stored” copies of infringing copyrighted material — but treating all computer manifestations not already transformed to indices as potentially-infringing copies).
† I prefer “Enhanced Eliza,” but I’m old and actually remember Eliza. These are not “artificial intelligence” because they have no capability to go beyond the scope of a directed response; if one asks for “an explanation of current issues with selecting a Speaker of the House to show to my grandmother,” they won’t (a) ask if granny speaks/reads English, (b) ask how much your granny knows about American government structure and contemporary politics, or (c) ask if granny is a veteran of either the John Birch Society or the Weather Underground. They won’t even ask if granny will be upset or confused by use of technical terminology.
dwsmith
Thanks, C.E. Very clear. Really appreciated.
Fabien Delorme
There are now AI models that are being trained ethically and are open / auditable, meaning anyone can control how they’ve been trained and ensure they haven’t stolen anyone’s copyright (there’s enough public domain / permissively licenced data on the web to train such a model anyway).
I still personally wouldn’t buy something written that way (why would I give money to a lazy writer who doesn’t even own the copyright on the produced content), but I’m mentioning them because I’m pretty sure these ethical models will soon put out of business the companies whose models were trained through copyright theft.
Lorri Moulton
And yet, there are authors who can’t wait to hand their stories over to Google and Apple to get their “free” AI audiobooks. I have not looked at those Terms of Service, but one wonders what exactly they’re exchanging for those “free” audiobooks.
I’ve been so fortunate to find some really terrific narrators willing to do a royalty share on ACX. Not the perfect solution, but one I’m happier with than using a non-human voice. Cheers, Dean…and thank you for sharing our Kickstarter campaigns!
Xander Koolen
I always wonder if the companies that sell AI training sets even paid for the copies of the e-books they have in their training sets. Considering how many books get published every day, this would get crazy expensive, super fast, so my guess is they actually scrape the internet – piracy websites included – and just pretend this is all sort of legal under the startup motto of “move fast and break things.”
Yeah, Dean’s advice of waiting until the courts have settled the matter and there is some sort of compensation for artists and writers is very sound.
And even then, I still wouldn’t use AI for my writing. Because here’s the thing: Why would I want to automate the fun parts?
Catherine
Exciting news about so many kickstarters! It has become my favorite way to support writers. Alerted to the new books, plus can buy goodies at the same time. I can really see the value after being a supporter for a few years.
Having left a group that was a major part of my life because of the AI stuff, and unsubscribed from more newsletters than I want to count when they advertised AI tutorials (teaching how to use the stuff is theft just a much), I’d be more than a little lonely and depressed without your vocal blog and position. It has been eye-opening in a rather depressing way. On the plus side, it is a good way sort through to find some real gems of people who won’t do such things.
Regarding creating the free audio books – I would not do that with a ten-foot pole. The Apple terms of service don’t seem all that clear, and Draft2Digital had no idea what Apple was doing with the IP it creates and says it has a right to in the terms of service, despite agreeing to push it on their site.
I would also NOT make an exception for Adobe AI anything right now. They accept and sell AI work from third parties. All people have to do is click a box. Or not. Frankly, I wouldn’t take any new stock images from unknown artists until things shake out.
dwsmith
Catherine, with you all the way. Not doing Adobe or anything until this all settles out.
Kevin McLaughlin
Then my advice would be to avoid Photoshop *entirely* and switch to Affinity or something instead.
The Adobe AI now underpins about a quarter of all Photoshop features.
Healing brush is and always has been generative AI; but now even stuff like the scaling tool utilizes AI for greater accuracy.
It’s functionally no longer possible to use Photoshop to make a cover without using AI. If you want to avoid AI use entirely, avoid Photoshop. Affinity is an awesome backup option.
dwsmith
AI is something that has been and will be a part of our lives going forward. Two area that are critical here in late 2023. Licensing and paying providers for training. Adobe is known for doing that. Second are things like ChapGTP and others like it, that can create entire product with a few prompts. They have not paid for licensing and have not even bothered to try, stealing work whole cloth from everywhere. They are scum.
The second is decisions on how much of a product can be protected under copyright if touched by AI. If made up out of prompts, then none, if only used a spellchecker, then sure. Somewhere in the middle is the answer and that is going to take a lot of work by the copyright office and the courts to figure that out.
Sheila
“Kickstarter now has a policy that if you used any kind of AI in your book, you must declare it and if you don’t and they find out, they will cancel your project and ban you from the site.”
Good for them. All sites should not allow any form of “AI” use at all. I refuse to use any so-called AI, my Photoshop is an older version, no AI on it. I don’t use Grammerly or anything else like that. I don’t consider basic spell checkers to be any kind of AI. And we truly don’t actually have AI anyway, despite everyone falling on the side of calling computer programs AI.