wonderful. just do the hard, hard work of creating our own stuff and let the chips lie.
or is it let the chips fly? and are those cheddar or salt and vinagar chips cuz if they are i am definitely not letting those darlings fly anywhere but into my mouth. Sorry, love chips. Sorry, sorry.
I recently read a fairly major author who talked about how she uses AI to plan her books, write all the marketing copy, write marketing plans. She said she even has it write some characters whose voices she has trouble writing. Another few years and I believe she'll be letting it generate everything. In the same article, she complained about the terrible things people say to her about it, and how she's had to put all her comments behind a paywall. Then I read articles like these. You know AI was trained on *all fanfiction*, right? That's why everything it spits out sounds like a third grader wrote it?
The industry is going to have to have a reckoning with it. I don't know which author you're referring to but I do know there was a recent scandal with a romance author.
I do know some very talented fanfiction authors so I hesitate to dunk on fanfic but I get the gist of what you're saying.
I'm sure she's getting terrible comments, and while I may agree with the sentiment behind those comments (stop using AI, please) I don't think we should act like star wars fans and bully people off the internet. (Not saying that's what you're advocating for, just a general reaction)
Interested to hear how her publisher handles this, because like Parker said below, I'm 90% sure that AI work cannot be copyrighted
Oh, she's indie and supporting herself on Patreon, so there's no publisher oversight. If I told you her name you'd probably know exactly who she is. She's been a writing guru and podcaster for a decade.
Not to dunk on fanfic (I write it too), but when you average out the vast majority, it comes out to ... Not great. Apparently they also trained AI on hospital records obtained illegally, so there's a lawsuit going about that, too.
Great insights. I think there is also some question as to whether a work made with generative AI can even be copyrighted, at least when it comes to visual art. I'm not sure if the question has been settled for fiction. Either way, I just don't see any point in using AI for any part of the fiction writing process.
Thanks for commenting. I'll respond to a few points here.
"You’re making an argument I see made often, that training AI amounts to stealing/plagiarism. But is that the case?"
Yes, it is the case. The creators of OpenAI did not pay for the use of those books for commercial purposes. The copyright act is kinda specific. I'm not a lawyer, full disclosure, but I am somewhat familiar with it due to being a college professor. It's my understanding that if you conceivably wanted to code an AI program yourself, solely for your personal use, any never, ever share it with anyone else or make money off of it, this would be fine in the eyes of the copyright act in regards to half of this issue -- that the authors and publishers of the works used to train AI weren't paid a licensing fee for the use of their materials. You can pretty much do whatever you want, so long it's for personal use (see burning cds, for example, if you're old enough to have done that). You could even give burnt cds to your friends, but you could not sell them for a profit.
Even in education, which most people generally view as a net good for society, we were limited in how much of a particular work we could run through the copy machine for in-class distribution because, if a work really is that valuable, students or the university should purchase that book for the ideas contained within, not simply copy the book without any royalties paid to the author/publisher.
OpenAI is neither a single person doing this for their private use or an institution of higher learning. They are a for-profit company that used those authors' works without paying a licensing fee to train their LLM and make money. Without those authors' works, the LLM wouldn't be functional. In that situation, a company *must* pay.
See also: NFL games. You can watch an NFL game in your house, or even with friends. But you can't rent out a theatre, charge admission, and stream an NFL game on redzone for 300 people in your community. It violates the license agreement. Private vs public use is an important distinction here.
***
"I say this partly because, for all the hue and cry, I’ve yet to see any clear examples where someone can say “ChapGPT spat out text clearly drawn from this novel.”
Yes, ChatGpt seems to have fixed that issue, but again, the problem with using it isn't necessarily that it *might* spit out a worse version/similar version of a work that looks a lot like, say "IT" by Stephen King (though, I do admit, I do use that example in the article). Again, it comes down to the LLM NEEDING to be trained on those books to function, which were acquired WITHOUT giving the authors of those works an opportunity to sell the rights of their book for the purposes of training AI.
***
Another, larger issue here is that Text and images generated through LLMs like Chat GPT cannot be copyrighted under US copyright law. In 2022, the US copyright office added that a work must be created by a human to be copyrightable:
"The Human Authorship Requirement - The U.S. Copyright Office will register an original work of authorship, provided that the work was created by a human being. The copyright law only protects “the fruits of intellectual labor” that “are founded in the creative powers of the mind.” Trade-Mark Cases, 100 U.S. 82, 94 (1879)."
A press would have no interest in publishing a work that cannot be copyrighted because they they then cannot sell the rights. I don't know how plugged into the publishing industry you are, but the sale of first publication rights, audio rights, film rights, and rights in other languages make up not a small portion of both author income and publisher income, so, unless the US copyright act changes, I do think that publishers will continue to exclude AI work.
***
"I’d argue that when we release a book into the world, we give up some ownership of it.*
I do agree with this. Author intention only matters to the author. This was famously documented when Dr. Seuss threatened to sue anti-abortion groups who co-oped the "a person's a person no matter how small" for the pro-life movement. The line was originally meant to criticize the bombing of Hiroshima and Nagasaki. Seuss didn't like that his words were being used out of context.
HOWEVER, Authors still do retain ownership of the rights of their work until they're sold. Like I said before, audio, video, first print publication, other language rights, use of the work for education, etc. You or I cannot make a film adaptation of our favorite novel and charge money for people to see it without the author's permission. (Note, the money exchanging hands is the real issue here). We cannot upload an audiobook to spotify or audible without paying for the rights to use and charge money for that work.
Finally, for me this is simple. When we teach our students to write argumentative essays, we teach them to cite their work properly. If I read a bunch of essays, write my own, and don't cite the ideas I took from previous essays, that's plagiarism. In much the same way, chat GPT not paying authors for the rights to use their work to train their AI model is also theft, and using a LLM trained on stolen work is also engaging with that stolen work and profiting off of stolen work.
Outside of the copyright issue, I think it comes down to your question of commercial products being able to create anything.
In a sense, all story is derivative in some way because all story engages in dialogue with what has come before. Your example of studying Martin Scorese is exactly what we ask students to do with their expository papers at university - the student (or analyst) is describing a work as it appears. That's how discourse happens and how the cultural lexicon is deepened.
Yet, in this case, it's not a human studying, it's a LLM, which by definition is not human. That matters in regards to copyright law and in the larger context of art. It appears this is what you’ve been missing in your questions: in both previous comments you’ve compared AI’s “learning” to your own real, human learning. You are by far not the first person I’ve seen do this and I think it sort of begs the question as to why we’re comparing a complex algorithm to ourselves, humans. If you read Ted Chiang’s article “Why AI isn’t going to make art,” he addresses this misunderstanding of intelligence and how comparing AI “learning” to our own learning is a false equivalence.
Aside from the ethical/pedagogical concerns about AI (which have been explored in the posts mentioned in my article as well as various others) what this comes down to is how the initial LLM was trained, as I’ve said.
If open AI paid the proper licensing would AI be "ok" to use in the way you described -- as a PRIVATE learning tool rather than a generative tool?
My answer in a vacuum would tentatively be "yes,” as a private learning tool. But as an educator it's still not advisable in the context of deep or meaningful learning. In my experience, students learn far better when they discover the answer on their own rather than it being handed to them via AI or any other instructor -- but that's a pedagogical conversation about education, not an AI conversation.
In regards to the other part of your question -- using AI to generate a plotline and changing all the words, again, it comes down to how the LLM was trained. If openAI used properly licensed work, generating a plotline is something that I personally wouldn't do but it would “solve” the “stolen work” issue. Unfortunately, that's not what happened when these programs were built. And any use of AI continues to be engaging in theft/plagiarism. You’ve said you feel this claim is dubious. I encourage you to read the lawsuit linked above and accounts from other authors or artists in the community.
In the larger conversation of gaining knowledge/talent, I think that AI is a paradox. In a perfect world of completely ethical AI training, I'm skeptical of how helpful it would be for writers who have not mastered their craft already. I've spent half my life learning about writing. I've read dozens of craft books, earned my MFA in writing popular fiction, attended workshops, hired developmental editors, attended conventions, and only in the last 2-3 years have I felt I've truly learned enough about novel writing to properly understand, manipulate, and break the “rules” or “conventions” of the craft. In short: if I had access to an ethical AI early in my career it would have more hindered than helped me because I would not have understood the depth of the knowledge presented to me.
Now, as someone who has mastered some aspects of the craft, I still don't use it because I don't need to. Nor do I even want to. Even if it saved me "time" in one phase of the process, I'd spend that time in other phases verifying the information that AI gave me. And I'd much rather pull 80,000 words out of my brain that have been formed into a plot that I created with characters that I like than change 80,000 words of an algo-generated novel so it's passing as "original." I suspect it's not as big as a time saver or as powerful of a writing tool as people thing.
How many craft books are out there that provide analysis of the hero’s journey or detective PI structure? And yet, publishing remains an incredibly difficult field to break into and despite ready access to plot structures we’re not seeing a resurgence of the hero’s journey or hard boiled detective novels.
Again, I want to make it clear that these thought experiments exist in a vacuum and don’t take into account the other ethical concerns about AI’s impact on the environment, the theft of labor from other professionals, and the psychology behind why we consume art in the first place.
wonderful. just do the hard, hard work of creating our own stuff and let the chips lie.
or is it let the chips fly? and are those cheddar or salt and vinagar chips cuz if they are i am definitely not letting those darlings fly anywhere but into my mouth. Sorry, love chips. Sorry, sorry.
I recently read a fairly major author who talked about how she uses AI to plan her books, write all the marketing copy, write marketing plans. She said she even has it write some characters whose voices she has trouble writing. Another few years and I believe she'll be letting it generate everything. In the same article, she complained about the terrible things people say to her about it, and how she's had to put all her comments behind a paywall. Then I read articles like these. You know AI was trained on *all fanfiction*, right? That's why everything it spits out sounds like a third grader wrote it?
The industry is going to have to have a reckoning with it. I don't know which author you're referring to but I do know there was a recent scandal with a romance author.
I do know some very talented fanfiction authors so I hesitate to dunk on fanfic but I get the gist of what you're saying.
I'm sure she's getting terrible comments, and while I may agree with the sentiment behind those comments (stop using AI, please) I don't think we should act like star wars fans and bully people off the internet. (Not saying that's what you're advocating for, just a general reaction)
Interested to hear how her publisher handles this, because like Parker said below, I'm 90% sure that AI work cannot be copyrighted
Oh, she's indie and supporting herself on Patreon, so there's no publisher oversight. If I told you her name you'd probably know exactly who she is. She's been a writing guru and podcaster for a decade.
Not to dunk on fanfic (I write it too), but when you average out the vast majority, it comes out to ... Not great. Apparently they also trained AI on hospital records obtained illegally, so there's a lawsuit going about that, too.
Great insights. I think there is also some question as to whether a work made with generative AI can even be copyrighted, at least when it comes to visual art. I'm not sure if the question has been settled for fiction. Either way, I just don't see any point in using AI for any part of the fiction writing process.
Hi Wil,
Thanks for commenting. I'll respond to a few points here.
"You’re making an argument I see made often, that training AI amounts to stealing/plagiarism. But is that the case?"
Yes, it is the case. The creators of OpenAI did not pay for the use of those books for commercial purposes. The copyright act is kinda specific. I'm not a lawyer, full disclosure, but I am somewhat familiar with it due to being a college professor. It's my understanding that if you conceivably wanted to code an AI program yourself, solely for your personal use, any never, ever share it with anyone else or make money off of it, this would be fine in the eyes of the copyright act in regards to half of this issue -- that the authors and publishers of the works used to train AI weren't paid a licensing fee for the use of their materials. You can pretty much do whatever you want, so long it's for personal use (see burning cds, for example, if you're old enough to have done that). You could even give burnt cds to your friends, but you could not sell them for a profit.
Even in education, which most people generally view as a net good for society, we were limited in how much of a particular work we could run through the copy machine for in-class distribution because, if a work really is that valuable, students or the university should purchase that book for the ideas contained within, not simply copy the book without any royalties paid to the author/publisher.
OpenAI is neither a single person doing this for their private use or an institution of higher learning. They are a for-profit company that used those authors' works without paying a licensing fee to train their LLM and make money. Without those authors' works, the LLM wouldn't be functional. In that situation, a company *must* pay.
See also: NFL games. You can watch an NFL game in your house, or even with friends. But you can't rent out a theatre, charge admission, and stream an NFL game on redzone for 300 people in your community. It violates the license agreement. Private vs public use is an important distinction here.
***
"I say this partly because, for all the hue and cry, I’ve yet to see any clear examples where someone can say “ChapGPT spat out text clearly drawn from this novel.”
Yes, ChatGpt seems to have fixed that issue, but again, the problem with using it isn't necessarily that it *might* spit out a worse version/similar version of a work that looks a lot like, say "IT" by Stephen King (though, I do admit, I do use that example in the article). Again, it comes down to the LLM NEEDING to be trained on those books to function, which were acquired WITHOUT giving the authors of those works an opportunity to sell the rights of their book for the purposes of training AI.
***
Another, larger issue here is that Text and images generated through LLMs like Chat GPT cannot be copyrighted under US copyright law. In 2022, the US copyright office added that a work must be created by a human to be copyrightable:
"The Human Authorship Requirement - The U.S. Copyright Office will register an original work of authorship, provided that the work was created by a human being. The copyright law only protects “the fruits of intellectual labor” that “are founded in the creative powers of the mind.” Trade-Mark Cases, 100 U.S. 82, 94 (1879)."
Source: https://www.bloomberglaw.com/external/document/XDDQ1PNK000000/copyrights-professional-perspective-copyright-chaos-legal-implic
A press would have no interest in publishing a work that cannot be copyrighted because they they then cannot sell the rights. I don't know how plugged into the publishing industry you are, but the sale of first publication rights, audio rights, film rights, and rights in other languages make up not a small portion of both author income and publisher income, so, unless the US copyright act changes, I do think that publishers will continue to exclude AI work.
***
"I’d argue that when we release a book into the world, we give up some ownership of it.*
I do agree with this. Author intention only matters to the author. This was famously documented when Dr. Seuss threatened to sue anti-abortion groups who co-oped the "a person's a person no matter how small" for the pro-life movement. The line was originally meant to criticize the bombing of Hiroshima and Nagasaki. Seuss didn't like that his words were being used out of context.
HOWEVER, Authors still do retain ownership of the rights of their work until they're sold. Like I said before, audio, video, first print publication, other language rights, use of the work for education, etc. You or I cannot make a film adaptation of our favorite novel and charge money for people to see it without the author's permission. (Note, the money exchanging hands is the real issue here). We cannot upload an audiobook to spotify or audible without paying for the rights to use and charge money for that work.
Finally, for me this is simple. When we teach our students to write argumentative essays, we teach them to cite their work properly. If I read a bunch of essays, write my own, and don't cite the ideas I took from previous essays, that's plagiarism. In much the same way, chat GPT not paying authors for the rights to use their work to train their AI model is also theft, and using a LLM trained on stolen work is also engaging with that stolen work and profiting off of stolen work.
Sorry for the novel. Hope this helps.
Outside of the copyright issue, I think it comes down to your question of commercial products being able to create anything.
In a sense, all story is derivative in some way because all story engages in dialogue with what has come before. Your example of studying Martin Scorese is exactly what we ask students to do with their expository papers at university - the student (or analyst) is describing a work as it appears. That's how discourse happens and how the cultural lexicon is deepened.
Yet, in this case, it's not a human studying, it's a LLM, which by definition is not human. That matters in regards to copyright law and in the larger context of art. It appears this is what you’ve been missing in your questions: in both previous comments you’ve compared AI’s “learning” to your own real, human learning. You are by far not the first person I’ve seen do this and I think it sort of begs the question as to why we’re comparing a complex algorithm to ourselves, humans. If you read Ted Chiang’s article “Why AI isn’t going to make art,” he addresses this misunderstanding of intelligence and how comparing AI “learning” to our own learning is a false equivalence.
Aside from the ethical/pedagogical concerns about AI (which have been explored in the posts mentioned in my article as well as various others) what this comes down to is how the initial LLM was trained, as I’ve said.
If open AI paid the proper licensing would AI be "ok" to use in the way you described -- as a PRIVATE learning tool rather than a generative tool?
My answer in a vacuum would tentatively be "yes,” as a private learning tool. But as an educator it's still not advisable in the context of deep or meaningful learning. In my experience, students learn far better when they discover the answer on their own rather than it being handed to them via AI or any other instructor -- but that's a pedagogical conversation about education, not an AI conversation.
In regards to the other part of your question -- using AI to generate a plotline and changing all the words, again, it comes down to how the LLM was trained. If openAI used properly licensed work, generating a plotline is something that I personally wouldn't do but it would “solve” the “stolen work” issue. Unfortunately, that's not what happened when these programs were built. And any use of AI continues to be engaging in theft/plagiarism. You’ve said you feel this claim is dubious. I encourage you to read the lawsuit linked above and accounts from other authors or artists in the community.
In the larger conversation of gaining knowledge/talent, I think that AI is a paradox. In a perfect world of completely ethical AI training, I'm skeptical of how helpful it would be for writers who have not mastered their craft already. I've spent half my life learning about writing. I've read dozens of craft books, earned my MFA in writing popular fiction, attended workshops, hired developmental editors, attended conventions, and only in the last 2-3 years have I felt I've truly learned enough about novel writing to properly understand, manipulate, and break the “rules” or “conventions” of the craft. In short: if I had access to an ethical AI early in my career it would have more hindered than helped me because I would not have understood the depth of the knowledge presented to me.
Now, as someone who has mastered some aspects of the craft, I still don't use it because I don't need to. Nor do I even want to. Even if it saved me "time" in one phase of the process, I'd spend that time in other phases verifying the information that AI gave me. And I'd much rather pull 80,000 words out of my brain that have been formed into a plot that I created with characters that I like than change 80,000 words of an algo-generated novel so it's passing as "original." I suspect it's not as big as a time saver or as powerful of a writing tool as people thing.
How many craft books are out there that provide analysis of the hero’s journey or detective PI structure? And yet, publishing remains an incredibly difficult field to break into and despite ready access to plot structures we’re not seeing a resurgence of the hero’s journey or hard boiled detective novels.
Again, I want to make it clear that these thought experiments exist in a vacuum and don’t take into account the other ethical concerns about AI’s impact on the environment, the theft of labor from other professionals, and the psychology behind why we consume art in the first place.