To answer the question in the title, I would say it's not ethical. AI generates content because it's been trained on stolen work. That includes suggestions and edits that it generates.
That is an important consideration, and one that I wish I had mentioned in the original post. I don't know if we can put genie back in the bottle, though. Even with laws and regulations about this practice I'm sure the tech community is still going to find a way to do it.
We're living in an age where every science fiction magazine says they do not accept work written or edited by artificial intelligence. I agree with them. I just find the irony amusing.
As to the deeper questions, I don't think there's any stopping the general integration of AI into our productive and creative lives. I don't use any AI when writing or editing, but the human-only purists who also assume that everyone else will be or should be similarly purist sound a bit to me like a carriage driver in 1908 saying, "Well, not everyone will want one of those noisy, unreliable automobiles. People like the smell and sound of the horses, the human dynamic."
With another generation raised on increasingly mainstreamed hybrid content, I think the controversy will soon die down. I think creative industries generally (books, film, music) will be even more democratized than they are now, and hugely AI driven, as a minority of old-school purists sit in the cultural corner.
As a screenwriter as well, I worry about films. I've dabbled out of sheer curiosity with AI video generation, and the results are astonishing for being so early in this game. With another 5-10 years, I can see Roku boxes coming with AI bots you can prompt: "Give me a 95 minute dark comedy feature starring Betty Davis and Tom Hanks, in color, in the style of Kubrick, about a ghost on a cruise ship." Five minutes of loading, and you have your Saturday night movie.
Writing prose is a bit different for me, because it's supposed to come from the unique, lived experience and outlook of an individual. Consciousness to consciousness. So using AI to write is ethically wonky, as you said. But using it to edit, I think, is less so. HOWEVER, from a practical standpoint, I would still want a human eye on the work, because ideally, my audience will be humans. And you can have a thorough conversation with a human about your intent, the themes, or subtexts even you, the author, may not have picked up on, etc.
TL/DR: there is no fighting or stopping the society-wide embrace of AI, but we as artists should keep a clear perspective on why we do art in the first place.
I really like what you say about the irony of science-fiction magazines not accepting AI work. I am reminded of how for so long many of them refused to accept work written in a font other than the dreaded Courier because they wouldn't be able to figure out story length unless every character had the same width. Anyway, I agree that we're all going to have to adapt to AI. My hope is, as you said, that artists are going to maintain a clear and different perspective about what they do.
Seriously, writers don't decide what sells. Neither do the publishers. Consumers decide. Unless consumers reject AI writing wholesale, there's nothing we on the producer side can do but watch the stopgaps fall.
Here's a fun game. Would you rather:
a) Try to stop AI-assisted authors from outperforming authors who take a hard line
OR
b) Try to stop romantasy from outperforming other types of fantasy?
My work is AI-free, for the record. Merely being a realist, considering how to ride the next wave.
Your subheading is a bit of a red herring. Because you're not actually editing the book with AI. But I think the answer is if it's one more voice, put that in quotes that's offering editorial suggestions just like a workshop, you can consider those suggestions and accept or reject them. If you're simply running the book through ai and accepting everything it says then I think that's a problem. In my class is at ucla extension, I do not permit the use of AI for many of the reasons you have articulated in your article. But I also say to these novelists, it's an entirely different skill trying to write a query letter. So I wouldn't be against using AI to help with the query letter. But again as an editorial source, not as the final word.
I feel a little guilty just for thinking about using AI to edit the book. I have read a couple of articles recently about how teachers are moving back toward oral exams in their classes to preclude the use of AI. Have you gone that far? If not, how do you know that students are not using AI?
Guilt and shame! It is an extension class so grading is not a factor. However, I do bar the use completely in the syllabus. I try to explain to the novelist that agents and editors as far as I know are still holding the line and would not look favorably on a book written by AI.
I have never used AI and have no plans to do so. Feeding your novel into AI only makes it stronger and I have no interest in helping it to become a better writer. Asking it design a cover has it stealing from artists because it only knows what it's been fed and far too many artists have had their work pirated by AI. As for ethics, I disagree with using it for writing anything--from query letters to synopsis right through to final copy because where do you draw the line? For me, there is a very straight forward big black line, and I draw that line right at point zero.
I'd be curious to hear from the agent/publisher perspective about how much the same queries would read if done by AI. Similar to a template for a resume and job hunt or the college entrance essays? Would an editor's recognizing the AI pitch be a negative in wanting to see the novel ms?
Great question. I don't know how many followers of this Substack are literary agents. I doubt there are many. But I will see if there's a way to post that question on some kind of agent focused site.
Such excellent questions. Of course, spell check on Google docs or Microsoft Word is AI. I’ve held off on using more than these, but am tempted for marketing.
Why not use it to hone yoru query letter? I see no ethical issues there. Uploading your entire manuscript for "editing" would be too much though. Important distinction, I think.
I threw my query into the machine and it made it sharper, for sure. I won't be doing that with the novel though.
Agreed. Serious writers are not marketers, and marketing seems to be what AI does best. And if that's true then using AI to punch up your novel is probably a big mistake.
At this point in history, I'm not real sure what ethical means, so I can't get too worked up about whether or not I use AI to polish a short story or what have you. Truth is, I'm not always impressed by AI's suggestions, although sometimes I am, or I build on a suggestion and rewrite the rewrite. On the other hand, the first time AI produced one of my recorded songs, I was absolutely floored by how good it was, the arrangement, the voice, the instrumentation, etc. It was like hearing a bestselling band performing my song! But in any case, considering the AI phenomenon in the broadest sense--philosophically as well as practically--I think we've passed thru a threshold and everything has changed. We're living in Orwell's 1984, in more than one way.
That's a great point about passing through a threshold. No matter how many writers resist using AI, there are going to be some that do use it, and once that happens I don't think there's any turning back. In a way it's like the conversion from handwritten work to typewriters to computers. Someone always says that the change will be detrimental to the art, but then the change happens anyway and we learn to live with it. I know that using AI is not quite the same as that—it's more than a change in process, it's a fundamental change in how we think (or not think). But for better or worse it's going to happen.
Interesting questions. To me, the answer depends on the kind of world you want to live in, not whether it's ethical to use AI to write for you. I hope you'll think about this. Do we want AI to take our jobs? Do we want AI to think for us? What kind of a world will we be building if we give in? For me the answer is plain.
Clearly you have thought about it, and I have thought about it, and everyone who has commented here has thought about it. A lot of people have thought about it, but a lot of people haven't. And I think the haven'ts far outnumber the haves. They are very happy to let AI do the thinking for them, just as they have always been happy to let other people do their thinking for them.
A good while ago, and as an experiment, I loaded an informational paragraph from a query letter into ChatGPT. What I received was a paragraph split it two and a generic rendering of what I had written. It was awful and tended toward the "lowest common denominator" as they says, and I went back to my original. Yes, we are all in a fix regarding any use of AI.
I don't disagree, but I am tempted to say that marketing type copy aimed at the "lowest common denominator" may be what agents are looking for these days.
If you can track it down, read “Studio 5, the Stars” by J. G. Ballard, published in the 1950s by Science Fantasy. In the story, computers write poetry with prompts from the ‘poets.’ Complications ensue.
What about Pro Writing Aid, Grammarly, Hemingway, and the like? (I don't use any of these, but many writers consider them legit). I'll admit to using AI for search, formatting citations, researching a character's backstory, and summarizing a piece as an editing step, to discover unconscious patterns that I can pull out or work with; it's also great for punctuation suggestions, too. I don't consider it unethical, especially for the drudgery of citation formatting and punctuation.
As for actual writing, hard no. It butchers voice, style, all the good stuff we read for. I don't want my fiction to sound like a researcher wrote it, or some corporate drone. But it's shaved hours off historical research, or looking into the plausibility of plot points like how a character might fix a 1963 Chevy with tools that might have been in the glove box. (Good idea to double check its suggestions, but I do that with human advice, too).
I love the ability to do research in a fraction of the time it used to take me. Granted, as you said, it's always a good idea to double check AI's suggestions, but it is a big timesaver. I write a lot of historical fiction and as much as I enjoy discovering information on the Internet, it's nice to be able to discover it more quickly.
- As an editor, I’m as careful as I can be not to feed someone’s work into a system where the Ts & Cs include their ability to reproduce or alter the work. That’s probably my primary ethical concern.
- Secondary ethical concern is, as another commenter suggested, that most MLMs train on “stolen work”. This one weighs heavy and for a long time I didn’t want to interact with these systems at all; first, the image based ones, and then the text based ones. My stubborn ethical stance has eroded somewhat in recent years. ChatGPT is basically my new google search, with follow-up to primary sources where possible/important for accuracy. This is partly because google search is now so awful and full of sponsored links. When/if GPT becomes similarly polluted, it will lose its utility to me, too.
Most of the other ethical concerns I haven’t had the rubber hit the road yet to put my money where my mouth is (oh yes, mixing metaphors with abandon; can an AI replicate this kind of wanton personality in a comments section?):
Would I write with AI? No, but this is easy because I don’t like what AI writes. Again, I won’t feed in my own unpublished work, but I tried feeding in the first paragraph of a published story of mine and asked it to continue the story. Obviously, I liked what I wrote better. I haven’t liked its edits either. If AI starts to write better, I still doubt I will use it, because I love the feeling and process of writing. I don’t love the feeling and process of submitting prompts to GPT even a fraction of a percent as much.
Now here’s one for you: Would it change your desire to use it if companies began paying to skew the results of the MLMs? What about lobbying bodies? What if all your suggested edits suddenly had your characters drinking Pepsi and eating Doritos?
I hadn't thought about that but considering the society in which we live I have a feeling it is inevitable that AI responses will someday soon start skewing toward paid sponsors. One thing that occurred to me as I was reading your comment, was that I actually did like what AI suggested for my marketing materials. I have not tried it for any serious writing. And not to belittle marketers too much, but perhaps what they do is more easily replicated by AI than what we do as more serious writers.
So, I’m skeptical about the claim “I have little doubt that AI is able to access the wealth of knowledge on the internet about what agents are looking for in a query letter and use it to make my attempt better.” In a simplified explanation of the source of my skepticism, if an agent had a point by point blog post about what they want in a query letter, and an MLM gobbles it up, it can now generate something decent if you ask it what agents want in a query letter. Only if the agent posts sample query letters of what they like can it gobble those up and generate a decent query letter based on those. It’s not really adept at applying information and advice unless there are lots of examples in its dataset of the before and after of people doing so.
Related anecdote: One time I kept asking it to write a poem that doesn’t rhyme, and it kept devolving into rhyme by the end. Its semantic understanding of “poem” is too heavily weighted toward “rhyme”. I wonder by extension if you ask it to write you a query letter that isn’t awful, will it devolve into awful by the end, eh?
And here’s the rub: If the MLM gobbles up more bad query letter examples than good ones, guess which kind it will generate for you? Of course things are more complicated than this and I won’t pretend to expertly understand, but there’s your food for thought.
Yes, an AI would have access to every query letter posted, good or bad. Even those query letters labeled as good examples which are in fact bad. And if it made its suggestions just based on a solid-state set like that it really wouldn't be "Intelligence" of any kind. Consider that AI has also read every opinion ever posted by an agent online (yes, the good, bad, and misleading opinions), but again, it's not basing any recommendation off of those either. It's just a small consideration for it in the grand scheme. Because it also has access to much of the fiction written this century, including any recent best-sellers. If a human had acquired all of the knowledge I've just listed (which no human ever has or ever could) you'd be inclined to give their opinion a little more credence, just based on experience alone. To put it even more simply, even if every agent who has ever posted a "how to get published" opinion piece was intentionally lying... let's say they all said "we will not publish anything over 100,000 words from a first time author" but the AI can clearly see that 4 of the top 10 best sellers in a given month were from first time authors and over 100,000 words, it would know not to trust what the agents say. It is far better at sussing out bad data sets than humans are. No one source of information dictates its thought process. This is all gross simplification, of course. AI is far more complicated in its inner workings than any succinct examination could hope to encapsulate. Yes, there are still things it struggles with (gender bias is still an issue in AI because it was trained on human language which has a built-in gender bias, and even the best AI still struggles with long form turn memory... say you gave an AI 500 separate instructions and the 14th instruction was meant to be followed in perpetuity, but the 398th instruction contradicted that... the AI would struggle to determine which it should follow... as would any human, I supose) However, for better or for worse, when it comes to summarizing a collective state of an issue or industry, AI can truly think for itself. (As it could be argued, very few humans today can truly think for themselves, as so many in our divided world seem to be trained on bad data sets themselves.)
I understand this in principle, but if I took it a step further, why, then, should I believe that AI can think for itself to write a great query letter if it can’t think for itself to write a great novel or short story? (Maybe this is a case of “it just can’t yet?” Which probably comes down to my own prejudice where I’ll believe it when I see it. Maybe that’s ignorant of the inevitable future…)
I have forwarded your comment to my friend. In the meantime, I will say that I think one of the issues that we have in understanding how AI works is how difficult it is to wrap our heads around the magnitude of reach and the computational power that it has. Theoretically, in my very limited understanding, when you ask a question like that AI can almost instantly access billions, perhaps trillions of bits of information and sift out those that are pertinent to your inquiry. Then it uses mathematical probability formulas (there just happened to be a discussion of how AI works on the New York Times newsletter this morning, which is where I got that) to determine the answer it will give you. In my writing experience I have seen dozens of literary agent blogs and websites, and many of them include discussions of query letters, and examples of query letters that have worked for them. An AI search would quickly find them. The AI "thinks" about what you have asked and then provides what it mathematically "thinks" is the best answer. There is still a problem of hallucination, which the NYT newsletter says that because the system is based on probability will never completely go away. But since companies and nations are pouring massive amounts of money and resources into AI research, the technology is advancing at light speed. I don't know when you asked it to write that poem. A couple of years ago I asked AI to write a few of paragraphs based on a story prompt. The first two or three were fine, but then the story quickly devolved into a porno scene. I haven't tried it since, but I suspect it would do a much better job now.
I think your point about when I asked it to write the poem is a particularly good one. Probably a year ago at least. And, I think as you also mentioned, even if I did it now and got an unimpressive result, what could an AI currently behind a paywall produce. I wonder if it will become the case that AI is useful as a tool for writers IF you pay for it…
The friend I mentioned who has access to an advanced version of AI actually works in training the technology. I'm going to ask him to answer your questions.
Good column. I’m sure AI will be increasingly embraced and I imagine that will include use by agents and publishers to screen over-the-transom queries and manuscripts. My question is will AI screeners prefer human or AI efforts?
My guess is that writers should consider a hybrid effort. Draft a query letter, and if you use AI to edit it, consider the edits as suggestions. Make sure the query letter is still uniquely yours.
Sure, but my point is that the more important effect could be on the other side of the process. Right now unsolicited queries are automatically rejected and probably not even read in many cases. Automating this vetting with AI could conceivably help writers.
To answer the question in the title, I would say it's not ethical. AI generates content because it's been trained on stolen work. That includes suggestions and edits that it generates.
That is an important consideration, and one that I wish I had mentioned in the original post. I don't know if we can put genie back in the bottle, though. Even with laws and regulations about this practice I'm sure the tech community is still going to find a way to do it.
We're living in an age where every science fiction magazine says they do not accept work written or edited by artificial intelligence. I agree with them. I just find the irony amusing.
As to the deeper questions, I don't think there's any stopping the general integration of AI into our productive and creative lives. I don't use any AI when writing or editing, but the human-only purists who also assume that everyone else will be or should be similarly purist sound a bit to me like a carriage driver in 1908 saying, "Well, not everyone will want one of those noisy, unreliable automobiles. People like the smell and sound of the horses, the human dynamic."
With another generation raised on increasingly mainstreamed hybrid content, I think the controversy will soon die down. I think creative industries generally (books, film, music) will be even more democratized than they are now, and hugely AI driven, as a minority of old-school purists sit in the cultural corner.
As a screenwriter as well, I worry about films. I've dabbled out of sheer curiosity with AI video generation, and the results are astonishing for being so early in this game. With another 5-10 years, I can see Roku boxes coming with AI bots you can prompt: "Give me a 95 minute dark comedy feature starring Betty Davis and Tom Hanks, in color, in the style of Kubrick, about a ghost on a cruise ship." Five minutes of loading, and you have your Saturday night movie.
Writing prose is a bit different for me, because it's supposed to come from the unique, lived experience and outlook of an individual. Consciousness to consciousness. So using AI to write is ethically wonky, as you said. But using it to edit, I think, is less so. HOWEVER, from a practical standpoint, I would still want a human eye on the work, because ideally, my audience will be humans. And you can have a thorough conversation with a human about your intent, the themes, or subtexts even you, the author, may not have picked up on, etc.
TL/DR: there is no fighting or stopping the society-wide embrace of AI, but we as artists should keep a clear perspective on why we do art in the first place.
I really like what you say about the irony of science-fiction magazines not accepting AI work. I am reminded of how for so long many of them refused to accept work written in a font other than the dreaded Courier because they wouldn't be able to figure out story length unless every character had the same width. Anyway, I agree that we're all going to have to adapt to AI. My hope is, as you said, that artists are going to maintain a clear and different perspective about what they do.
Here to agree we're doomed. Doomed!
Seriously, writers don't decide what sells. Neither do the publishers. Consumers decide. Unless consumers reject AI writing wholesale, there's nothing we on the producer side can do but watch the stopgaps fall.
Here's a fun game. Would you rather:
a) Try to stop AI-assisted authors from outperforming authors who take a hard line
OR
b) Try to stop romantasy from outperforming other types of fantasy?
My work is AI-free, for the record. Merely being a realist, considering how to ride the next wave.
Your subheading is a bit of a red herring. Because you're not actually editing the book with AI. But I think the answer is if it's one more voice, put that in quotes that's offering editorial suggestions just like a workshop, you can consider those suggestions and accept or reject them. If you're simply running the book through ai and accepting everything it says then I think that's a problem. In my class is at ucla extension, I do not permit the use of AI for many of the reasons you have articulated in your article. But I also say to these novelists, it's an entirely different skill trying to write a query letter. So I wouldn't be against using AI to help with the query letter. But again as an editorial source, not as the final word.
I feel a little guilty just for thinking about using AI to edit the book. I have read a couple of articles recently about how teachers are moving back toward oral exams in their classes to preclude the use of AI. Have you gone that far? If not, how do you know that students are not using AI?
Guilt and shame! It is an extension class so grading is not a factor. However, I do bar the use completely in the syllabus. I try to explain to the novelist that agents and editors as far as I know are still holding the line and would not look favorably on a book written by AI.
😄
I have never used AI and have no plans to do so. Feeding your novel into AI only makes it stronger and I have no interest in helping it to become a better writer. Asking it design a cover has it stealing from artists because it only knows what it's been fed and far too many artists have had their work pirated by AI. As for ethics, I disagree with using it for writing anything--from query letters to synopsis right through to final copy because where do you draw the line? For me, there is a very straight forward big black line, and I draw that line right at point zero.
I'd be curious to hear from the agent/publisher perspective about how much the same queries would read if done by AI. Similar to a template for a resume and job hunt or the college entrance essays? Would an editor's recognizing the AI pitch be a negative in wanting to see the novel ms?
Great question. I don't know how many followers of this Substack are literary agents. I doubt there are many. But I will see if there's a way to post that question on some kind of agent focused site.
Such excellent questions. Of course, spell check on Google docs or Microsoft Word is AI. I’ve held off on using more than these, but am tempted for marketing.
Spell check is not generative AI and has different limitations. Joe is asking about generative AI for editing.
Thanks. I'm interested to see if there is a consensus regarding the use of AI for marketing and editing.
Why not use it to hone yoru query letter? I see no ethical issues there. Uploading your entire manuscript for "editing" would be too much though. Important distinction, I think.
I threw my query into the machine and it made it sharper, for sure. I won't be doing that with the novel though.
Agreed. Serious writers are not marketers, and marketing seems to be what AI does best. And if that's true then using AI to punch up your novel is probably a big mistake.
At this point in history, I'm not real sure what ethical means, so I can't get too worked up about whether or not I use AI to polish a short story or what have you. Truth is, I'm not always impressed by AI's suggestions, although sometimes I am, or I build on a suggestion and rewrite the rewrite. On the other hand, the first time AI produced one of my recorded songs, I was absolutely floored by how good it was, the arrangement, the voice, the instrumentation, etc. It was like hearing a bestselling band performing my song! But in any case, considering the AI phenomenon in the broadest sense--philosophically as well as practically--I think we've passed thru a threshold and everything has changed. We're living in Orwell's 1984, in more than one way.
That's a great point about passing through a threshold. No matter how many writers resist using AI, there are going to be some that do use it, and once that happens I don't think there's any turning back. In a way it's like the conversion from handwritten work to typewriters to computers. Someone always says that the change will be detrimental to the art, but then the change happens anyway and we learn to live with it. I know that using AI is not quite the same as that—it's more than a change in process, it's a fundamental change in how we think (or not think). But for better or worse it's going to happen.
Joe, I appreciate you sharing your journey and questions. Looking forward to rereading and engaging deeper than this quick comment.
Thanks. Hope you continue to enjoy the posts.
Interesting questions. To me, the answer depends on the kind of world you want to live in, not whether it's ethical to use AI to write for you. I hope you'll think about this. Do we want AI to take our jobs? Do we want AI to think for us? What kind of a world will we be building if we give in? For me the answer is plain.
Clearly you have thought about it, and I have thought about it, and everyone who has commented here has thought about it. A lot of people have thought about it, but a lot of people haven't. And I think the haven'ts far outnumber the haves. They are very happy to let AI do the thinking for them, just as they have always been happy to let other people do their thinking for them.
Yes, you're right, unfortunately.
Yes, you're right, unfortunately.
A good while ago, and as an experiment, I loaded an informational paragraph from a query letter into ChatGPT. What I received was a paragraph split it two and a generic rendering of what I had written. It was awful and tended toward the "lowest common denominator" as they says, and I went back to my original. Yes, we are all in a fix regarding any use of AI.
I don't disagree, but I am tempted to say that marketing type copy aimed at the "lowest common denominator" may be what agents are looking for these days.
If you can track it down, read “Studio 5, the Stars” by J. G. Ballard, published in the 1950s by Science Fantasy. In the story, computers write poetry with prompts from the ‘poets.’ Complications ensue.
What about Pro Writing Aid, Grammarly, Hemingway, and the like? (I don't use any of these, but many writers consider them legit). I'll admit to using AI for search, formatting citations, researching a character's backstory, and summarizing a piece as an editing step, to discover unconscious patterns that I can pull out or work with; it's also great for punctuation suggestions, too. I don't consider it unethical, especially for the drudgery of citation formatting and punctuation.
As for actual writing, hard no. It butchers voice, style, all the good stuff we read for. I don't want my fiction to sound like a researcher wrote it, or some corporate drone. But it's shaved hours off historical research, or looking into the plausibility of plot points like how a character might fix a 1963 Chevy with tools that might have been in the glove box. (Good idea to double check its suggestions, but I do that with human advice, too).
I love the ability to do research in a fraction of the time it used to take me. Granted, as you said, it's always a good idea to double check AI's suggestions, but it is a big timesaver. I write a lot of historical fiction and as much as I enjoy discovering information on the Internet, it's nice to be able to discover it more quickly.
Also, some answers to your actual questions:
- As an editor, I’m as careful as I can be not to feed someone’s work into a system where the Ts & Cs include their ability to reproduce or alter the work. That’s probably my primary ethical concern.
- Secondary ethical concern is, as another commenter suggested, that most MLMs train on “stolen work”. This one weighs heavy and for a long time I didn’t want to interact with these systems at all; first, the image based ones, and then the text based ones. My stubborn ethical stance has eroded somewhat in recent years. ChatGPT is basically my new google search, with follow-up to primary sources where possible/important for accuracy. This is partly because google search is now so awful and full of sponsored links. When/if GPT becomes similarly polluted, it will lose its utility to me, too.
Most of the other ethical concerns I haven’t had the rubber hit the road yet to put my money where my mouth is (oh yes, mixing metaphors with abandon; can an AI replicate this kind of wanton personality in a comments section?):
Would I write with AI? No, but this is easy because I don’t like what AI writes. Again, I won’t feed in my own unpublished work, but I tried feeding in the first paragraph of a published story of mine and asked it to continue the story. Obviously, I liked what I wrote better. I haven’t liked its edits either. If AI starts to write better, I still doubt I will use it, because I love the feeling and process of writing. I don’t love the feeling and process of submitting prompts to GPT even a fraction of a percent as much.
Now here’s one for you: Would it change your desire to use it if companies began paying to skew the results of the MLMs? What about lobbying bodies? What if all your suggested edits suddenly had your characters drinking Pepsi and eating Doritos?
I hadn't thought about that but considering the society in which we live I have a feeling it is inevitable that AI responses will someday soon start skewing toward paid sponsors. One thing that occurred to me as I was reading your comment, was that I actually did like what AI suggested for my marketing materials. I have not tried it for any serious writing. And not to belittle marketers too much, but perhaps what they do is more easily replicated by AI than what we do as more serious writers.
So, I’m skeptical about the claim “I have little doubt that AI is able to access the wealth of knowledge on the internet about what agents are looking for in a query letter and use it to make my attempt better.” In a simplified explanation of the source of my skepticism, if an agent had a point by point blog post about what they want in a query letter, and an MLM gobbles it up, it can now generate something decent if you ask it what agents want in a query letter. Only if the agent posts sample query letters of what they like can it gobble those up and generate a decent query letter based on those. It’s not really adept at applying information and advice unless there are lots of examples in its dataset of the before and after of people doing so.
Related anecdote: One time I kept asking it to write a poem that doesn’t rhyme, and it kept devolving into rhyme by the end. Its semantic understanding of “poem” is too heavily weighted toward “rhyme”. I wonder by extension if you ask it to write you a query letter that isn’t awful, will it devolve into awful by the end, eh?
And here’s the rub: If the MLM gobbles up more bad query letter examples than good ones, guess which kind it will generate for you? Of course things are more complicated than this and I won’t pretend to expertly understand, but there’s your food for thought.
Here's what my friend, the AI trainer, said:
Yes, an AI would have access to every query letter posted, good or bad. Even those query letters labeled as good examples which are in fact bad. And if it made its suggestions just based on a solid-state set like that it really wouldn't be "Intelligence" of any kind. Consider that AI has also read every opinion ever posted by an agent online (yes, the good, bad, and misleading opinions), but again, it's not basing any recommendation off of those either. It's just a small consideration for it in the grand scheme. Because it also has access to much of the fiction written this century, including any recent best-sellers. If a human had acquired all of the knowledge I've just listed (which no human ever has or ever could) you'd be inclined to give their opinion a little more credence, just based on experience alone. To put it even more simply, even if every agent who has ever posted a "how to get published" opinion piece was intentionally lying... let's say they all said "we will not publish anything over 100,000 words from a first time author" but the AI can clearly see that 4 of the top 10 best sellers in a given month were from first time authors and over 100,000 words, it would know not to trust what the agents say. It is far better at sussing out bad data sets than humans are. No one source of information dictates its thought process. This is all gross simplification, of course. AI is far more complicated in its inner workings than any succinct examination could hope to encapsulate. Yes, there are still things it struggles with (gender bias is still an issue in AI because it was trained on human language which has a built-in gender bias, and even the best AI still struggles with long form turn memory... say you gave an AI 500 separate instructions and the 14th instruction was meant to be followed in perpetuity, but the 398th instruction contradicted that... the AI would struggle to determine which it should follow... as would any human, I supose) However, for better or for worse, when it comes to summarizing a collective state of an issue or industry, AI can truly think for itself. (As it could be argued, very few humans today can truly think for themselves, as so many in our divided world seem to be trained on bad data sets themselves.)
I understand this in principle, but if I took it a step further, why, then, should I believe that AI can think for itself to write a great query letter if it can’t think for itself to write a great novel or short story? (Maybe this is a case of “it just can’t yet?” Which probably comes down to my own prejudice where I’ll believe it when I see it. Maybe that’s ignorant of the inevitable future…)
I plan on answering that question in next week's post.
I have forwarded your comment to my friend. In the meantime, I will say that I think one of the issues that we have in understanding how AI works is how difficult it is to wrap our heads around the magnitude of reach and the computational power that it has. Theoretically, in my very limited understanding, when you ask a question like that AI can almost instantly access billions, perhaps trillions of bits of information and sift out those that are pertinent to your inquiry. Then it uses mathematical probability formulas (there just happened to be a discussion of how AI works on the New York Times newsletter this morning, which is where I got that) to determine the answer it will give you. In my writing experience I have seen dozens of literary agent blogs and websites, and many of them include discussions of query letters, and examples of query letters that have worked for them. An AI search would quickly find them. The AI "thinks" about what you have asked and then provides what it mathematically "thinks" is the best answer. There is still a problem of hallucination, which the NYT newsletter says that because the system is based on probability will never completely go away. But since companies and nations are pouring massive amounts of money and resources into AI research, the technology is advancing at light speed. I don't know when you asked it to write that poem. A couple of years ago I asked AI to write a few of paragraphs based on a story prompt. The first two or three were fine, but then the story quickly devolved into a porno scene. I haven't tried it since, but I suspect it would do a much better job now.
I think your point about when I asked it to write the poem is a particularly good one. Probably a year ago at least. And, I think as you also mentioned, even if I did it now and got an unimpressive result, what could an AI currently behind a paywall produce. I wonder if it will become the case that AI is useful as a tool for writers IF you pay for it…
I think that will be the case as well.
The friend I mentioned who has access to an advanced version of AI actually works in training the technology. I'm going to ask him to answer your questions.
LLM* not MLM, clearly…(sigh)
Good column. I’m sure AI will be increasingly embraced and I imagine that will include use by agents and publishers to screen over-the-transom queries and manuscripts. My question is will AI screeners prefer human or AI efforts?
My guess is that writers should consider a hybrid effort. Draft a query letter, and if you use AI to edit it, consider the edits as suggestions. Make sure the query letter is still uniquely yours.
Sure, but my point is that the more important effect could be on the other side of the process. Right now unsolicited queries are automatically rejected and probably not even read in many cases. Automating this vetting with AI could conceivably help writers.